A comprehensive survey on machine learning for networking: Evolution, applications and research opportunities

The majority of proposals for congestion inference apply diverse supervised learning techniques, mostly for prediction.

“As depicted in Table 13, the majority of these proposals apply diverse supervised learning techniques, mostly for prediction.”

The evaluation of ML-based AQM schemes was based only on simulations of wired networks, including single and multiple bottleneck topologies.

“However, the evaluation was based only on simulations of wired networks, though including single and multiple bottleneck topologies.”

All the classifiers perform binary classification of packet losses.

“In addition, all the classifiers perform binary classification of packet losses.”

TCP variants built upon ML-based CWND updating mechanisms perform better in terms of throughput and delay than standard and non-ML TCP versions particularly enhanced for specific network conditions.

“TCP variants built upon these ML-based CWND updating mechanisms perform better in terms of throughput and delay than standard and non-ML TCP versions particularly enhanced for specific network conditions (e.g.TCP-FeW and TCP-Cubic).”

The other half of ML-based CWND updating approaches use CALA or an own RL design on a wider range of network architectures including satellite, cellular, and data center networks.

“Whereas, the other half either use CALA or an own RL design on a wider range of network architectures, including satellite, cellular, and data center networks.”

Half of the ML-based CWND updating approaches apply FALA, CALA, or Q-learning on resource-constrained networks such as WANET and IoT.

“Half of these ML-based approaches apply FALA, CALA, or Q-learning (including the FK function approximation) on resource-constrained networks (i.e.WANET and IoT).”

The other half of ML-based AQM schemes employ reinforcement-based methods for deducing the increment in the packet drop probability.

“The other half of ML-based schemes employ reinforcement-based methods for deducing the increment in the packet drop probability.”

Some ML-based TCP also show improvements in packet loss and fairness.

“Some of the ML-based TCP also show improvements in packet loss and fairness.”

Half of the ML-based AQM schemes apply supervised OLS and NN for predicting future time series values of either traffic volume or queue length.

“As depicted in Table 11, half of the ML-based AQM schemes apply supervised OLS and NN for predicting future time series values of either traffic volume or queue length.”

All ML-based AQM schemes improve and speed up queue stabilization over non-ML AQM schemes for varying network conditions.

“All these ML-based AQM schemes improve and speed up the queue stabilization over non-ML AQM schemes for varying network conditions.”

The majority of ML-based classifiers were tested using synthetic data collected from simulations.

“The majority of the MLbased classifiers were tested using synthetic data collected from simulations.”

TCP variants built upon ML-based classifiers outperform standard and diverse non-ML TCP versions such as TCP-Veno and Burst-TCP.

“TCP variants built upon these ML-based classifiers outperform standard and diverse non-ML TCP versions (e.g.TCP-Veno and Burst-TCP).”

The evaluation of ML-based CWND updating mechanisms has been only based on synthetic data collected from simulations and experimental testbeds.

“The evaluation has been only based on synthetic data collected from simulations and experimental testbeds.”

Admission control has leveraged ML extensively in a variety of networks, including ATM networks, wireless networks, cellular networks, ad hoc networks, and next generation networks.

“As shown in Table 14, Admission control has leveraged ML extensively in a variety of networks, including ATM networks [95,189,190], wireless networks [8,36,359], cellular networks [66,67,281,372,458], ad hoc networks [452], and next generation networks [311].”

• cites [95] Cheng RG, Chang CJ. Neural-network connection-admission control for atm networks. IEE Proc-Commun. 1997; 144(2):93–8. sourceNot checked yet

  Not checked in the free run.

• cites [189] Hiramatsu A. Atm communications network control by neural networks. IEEE Trans Neural Netw. 1990; 1(1):122–30. sourceNot checked yet

  Not checked in the free run.

• cites [190] Hiramatsu A. Integration of atm call admission control and link capacity control by distributed neural networks. IEEE J Sel Areas Commun. 1991; 9(7):1131–8. sourceNot checked yet

  Not checked in the free run.

• cites [8] Ahn CW, Ramakrishna RS. Qos provisioning dynamic connection-admission control for multimedia wireless networks using a hopfield neural network. IEEE Trans Veh Technol. 2004; 53(1):106–117. sourceNot checked yet

  Not checked in the free run.

• cites [36] Baldo N, Dini P, Nin-Guerrero J. User-driven call admission control for voip over wlan with a neural network based cognitive engine. In: Cognitive Information Processing (CIP), 2010 2nd International Workshop on. IEEE: 2010. p. 52–6. sourceNot checked yet

  Not checked in the free run.

• cites [359] Piamrat K, Ksentini A, Viho C, Bonnin JM. Qoe-aware admission control for multimedia applications in ieee 802.11 wireless networks. In: Vehicular Technology Conference, 2008. VTC 2008-Fall. IEEE 68th. IEEE: 2008. p. 1–5. sourceNot checked yet

  Not checked in the free run.

• cites [66] Bojovic B, Baldo N, Dini P. A cognitive scheme for radio admission control in lte systems. In: Cognitive Information Processing (CIP), 2012 3rd International Workshop on. Piscataway: IEEE: 2012. p. 1–3.Not checked yet

  Not checked in the free run.

• cites [67] Bojovic B, Quer G, Baldo N, Rao RR. Bayesian and neural network schemes for call admission control in lte systems. In: Global Communications Conference (GLOBECOM), 2013 IEEE. Piscataway: IEEE: 2013. p. 1246–52.Not checked yet

  Not checked in the free run.

• cites [281] Liu D, Zhang Y, Zhang H. A self-learning call admission control scheme for cdma cellular networks. IEEE trans neural netw. 2005; 16(5):1219–28. sourceNot checked yet

  Not checked in the free run.

• cites [372] Quer G, Baldo N, Zorzi M. Cognitive call admission control for voip over ieee 802.11 using bayesian networks. In: Global Telecommunications Conference (GLOBECOM 2011), 2011 IEEE. IEEE: 2011. p. 1–6. sourceNot checked yet

  Not checked in the free run.

The evaluation of ML-based estimators was mostly performed using synthetic data collected from simulations and laboratory testbeds.

“The evaluation was mostly performed using synthetic data collected from simulations and laboratory testbeds.”

Various challenges still prevail for different networks, including cellular networks, wireless networks, and ad hoc networks, despite widespread application of ML for load prediction and resource management in cloud data centers.

“Despite the widespread application of ML for load prediction and resource management in cloud data centers [367], various challenges still prevail for different networks, including cellular networks, wireless networks and ad hoc networks.”

• cites [367] Prevost JJ, Nagothu K, Kelley B, Jamshidi M. Prediction of cloud data center networks loads using stochastic and neural models. IEEE: 2011. p. 276–281. sourceNot checked yet

  Not checked in the free run.

Only the WMA-based estimators collected real data to test their approaches.

“Only the WMA-based estimators collected real data to test their approaches.”

All ML-based estimators outperform the accuracy of conventional estimation mechanisms such as EWMA and history-based.

“All these ML-based estimators outperform the accuracy of conventional estimation mechanisms (e.g.EWMA and history-based).”

The majority of ML-based solutions rely on synthetic data to conduct the evaluation.

“As final remarks, note that the majority of the ML-based solutions rely on synthetic data to conduct the evaluation.”

Application of unsupervised techniques in resource management is rather unexplored, with the exception of a few works.

“Application of unsupervised techniques in resource management is rather unexplored, with the exception of a few works.”

A common challenge faced by ML-based fault management techniques is the scarcity of fault data generated in a production network.

“However, a common challenge faced by these techniques is the scarcity of fault data generated in a production network.”

Most ML-based fault management approaches use different supervised learning techniques that depend on training data to predict/detect/locate faults in the network.

“As summarized in Tables 16, 17 and 18, most of the ML-based fault management approaches use different supervised learning techniques that depend on training data to predict/detect/locate faults in the network.”

Due to the lack of real-life traces, ML-based resource management approaches adopt different methods to simulate the network of interest and produce training and testing data.

“Due to the lack of real-life traces, these approaches adopt different methods to simulate the network of interest and produce training and testing data.”

ML-based resource management schemes studied in the paper can be broadly classified into two groups: supervised learning-based and RL-based.

“As evident from Tables 14 and 15, the ML-based resource management schemes studied in this paper can be broadly classified into two groups-supervised learning-based and RL-based.”

Some approaches leverage unsupervised techniques that rely on detecting changes in network states instead of using labeled fault data.

“As a remedy, some approaches leverage unsupervised techniques that rely on detecting changes in network states instead of using labeled fault data.”

The evaluation settings adopted by ML-based fault management approaches lack uniformity.

“In addition, the evaluation settings adopted by these approaches lack uniformity.”

The fault scenarios studied in ML-based fault management approaches vary greatly as they depend both on the layer and the type of the network.

“Consequently, the fault scenarios studied in these approaches vary greatly as they depend both on the layer (e.g.physical, link, or IP layer) and the type (e.g.cellular, wireless, local area network) of the network.”

The ML-based fault management approaches surveyed focus on a variety of networks.

“The ML-based fault management approaches surveyed in this paper focus on a variety of networks.”

The feature set and output of ML-based fault management schemes depend on the fault scenario of a particular network.

“The same holds for feature set and output of these schemes, as both features and outputs depend on the fault scenario of a particular network.”

PCA can detect 25 anomalies with 0 false positives in the packet time series.

“However, PCA can detect 25 anomalies with 0 FP.”

• cites [7] Ahmed T, Coates M, Lakhina A. Multivariate online anomaly detection using kernel recursive least squares.IEEE; 2007. pp. 625–33. sourceNot checked yet

  Not checked in the free run.

Detecting new attacks using RBF-SVM with limited and full features achieves comparable true positive rates with a high false positive rate of approximately 18% for normal traces.

“However, detecting new attacks using the RBF-SVM with limited and full features achieve comparable TP with a high FP of approximately 18% for normal traces.”

• cites [64] Boero L, Marchese M, Zappatore S. Support vector machine meets software defined networking in ids domain. In: Proceedings of the 29th International Teletraffic Congress (ITC), vol. 3. New York: IEEE: 2017. p. 25–30.Not checked yet

  Not checked in the free run.

On the KDD dataset, IMM outperforms three related works in terms of detection rate and false positives.

“The KDD dataset is used for evaluation and shows that IMM outperforms three related works [131, 139, 363] in terms of detection rate and FP.”

• cites [220] Jiang S, Song X, Wang H, Han JJ, Li QH. A clustering-based method for unsupervised intrusion detections. Pattern Recog Lett. 2006; 27(7):802–10. sourceNot checked yet

  Not checked in the free run.

There are various surveys on machine learning in networking, but they are specialized to specific areas or specific network technologies.

“Though there are various surveys on ML in networking [18,61,82,142,246,339], this survey is purposefully different.”

• cites [18] Alsheikh MA, Lin S, Niyato D, Tan HP. Machine learning in wireless sensor networks: Algorithms, strategies, and applications. IEEE Commun Surv Tutor. 2014; 16(4):1996–2018. sourceNot checked yet

  Not checked in the free run.

• cites [61] Bkassiny M, Li Y, Jayaweera SK. A survey on machine-learning techniques in cognitive radios. IEEE Commun Surv Tutor. 2013; 15(3):1136–59. sourceNot checked yet

  Not checked in the free run.

• cites [82] Buczak AL, Guven E. A survey of data mining and machine learning methods for cyber security intrusion detection. IEEE Commun Surv Tutor. 2016; 18(2):1153–76. sourceNot checked yet

  Not checked in the free run.

• cites [142] Fadlullah Z, Tang F, Mao B, Kato N, Akashi O, Inoue T, Mizutani K. State-of-the-art deep learning: Evolving machine intelligence toward tomorrow’s intelligent network traffic control systems. IEEE Commun Surv Tutor. 2017; PP(99):1.Not checked yet

  Not checked in the free run.

• cites [246] Klaine PV, Imran MA, Onireti O, Souza RD. A survey of machine learning techniques applied to self organizing cellular networks. IEEE Commun Surv Tutor. 2017; PP(99):1.Not checked yet

  Not checked in the free run.

• cites [339] Nguyen TTT, Armitage G. A survey of techniques for internet traffic classification using machine learning. IEEE Commun Surv Tutor. 2008b; 10(4):56–76. sourceNot checked yet

  Not checked in the free run.

The patterns proven to work in one network may not be feasible for another network of the same kind.

“Therefore, the patterns proven to work in one network may not be feasible for another network of the same kind.”

Fadlullah et al. and Buczak et al., both state-of-the-art surveys, have a specialized treatment of machine learning to specific problems in networking.

“Whereas, Fadlullah et al. [142] and Buczak et al. [82], both state-of-theart surveys, have a specialized treatment of ML to specific problems in networking.”

• cites [142] Fadlullah Z, Tang F, Mao B, Kato N, Akashi O, Inoue T, Mizutani K. State-of-the-art deep learning: Evolving machine intelligence toward tomorrow’s intelligent network traffic control systems. IEEE Commun Surv Tutor. 2017; PP(99):1.Not checked yet

  Not checked in the free run.

• cites [82] Buczak AL, Guven E. A survey of data mining and machine learning methods for cyber security intrusion detection. IEEE Commun Surv Tutor. 2016; 18(2):1153–76. sourceNot checked yet

  Not checked in the free run.

The ability to program the network by leveraging Software-Defined Networking alleviates the obstacles to deploying machine learning for network operations and management.

“The ability to program the network by leveraging SDN alleviates these obstacles.”

Features that are extracted from multi-modal application classes (e.g. WWW) tend to show a non-Gaussian behavior.

“For example, in traffic classification, features that are extracted from multi-modal application classes (e.g.WWW) tend to show a non-Gaussian behavior [321].”

• cites [321] Moore AW, Zuev D. Internet traffic classification using bayesian analysis techniques. In: ACM SIGMETRICS Performance Evaluation Review, ACM, vol 33.2005. p. 50–60. sourceNot checked yet

  Not checked in the free run.

Average packet size can be skewed by intermediate fragmentation and encryption.

“However, average packet size can be skewed by intermediate fragmentation and encryption, and packet inter-arrival times and their distributions are affected by queuing in routers [492].”

These relationships contradict widely held assumptions in traffic classification, such as feature distributions being independent and following a Gaussian distribution.

“These relationships not only become irrelevant and redundant, they contradict widely held assumptions in traffic classification, such as feature distributions being independent and following a Gaussian distribution.”

Average packet size and flow duration have been shown to be good discriminators.

“Average packet size and flow duration have been shown to be good discriminators [390].”

• cites [390] Roughan M, Sen S, Spatscheck O, Duffield N. Class-of-service mapping for qos: a statistical signature-based approach to ip traffic classification. In: Proceedings of the 4th ACM SIGCOMM conference on Internet measurement. ACM: 2004a. p. 135–148. sourceNot checked yet

  Not checked in the free run.

Streaming applications often behave similar to bulk data transfer applications.

“Furthermore, streaming applications often behave similar to bulk data transfer applications [390].”

• cites [390] Roughan M, Sen S, Spatscheck O, Duffield N. Class-of-service mapping for qos: a statistical signature-based approach to ip traffic classification. In: Proceedings of the 4th ACM SIGCOMM conference on Internet measurement. ACM: 2004a. p. 135–148. sourceNot checked yet

  Not checked in the free run.

Packet inter-arrival times are a good discriminator for bulk data transfer (e.g. FTP) and streaming applications.

“Average packet size is fairly constant in nature [492] and packet inter-arrival times are a good discriminator for bulk data transfer (e.g.FTP) and streaming applications [390].”

• cites [390] Roughan M, Sen S, Spatscheck O, Duffield N. Class-of-service mapping for qos: a statistical signature-based approach to ip traffic classification. In: Proceedings of the 4th ACM SIGCOMM conference on Internet measurement. ACM: 2004a. p. 135–148. sourceNot checked yet

  Not checked in the free run.

Careful feature extraction and selection is crucial for the performance of machine learning models.

“Therefore, careful feature extraction and selection is crucial for the performance of ML models [77].”

• cites [77] Brill E, Lin JJ, Banko M, Dumais ST, Ng AY, et al. Data-intensive question answering. In: TREC, vol. 56.2001. p. 90. sourceNot checked yet

  Not checked in the free run.

Packet inter-arrival times and their distributions are affected by queuing in routers.

“However, average packet size can be skewed by intermediate fragmentation and encryption, and packet inter-arrival times and their distributions are affected by queuing in routers [492].”

• cites [492] Zander S, Nguyen T, Armitage G. Automated traffic classification and application identification using machine learning.IEEE; 2005, pp. 250–7. sourceNot checked yet

  Not checked in the free run.

There is an inherent mutual dependency on the size of the training data of one class of interest on another, impacting model performance.

“There is also an inherent mutual dependency on the size of the training data of one class of interest on another, impacting model performance [417].”

• cites [417] Soysal M, Schmidt EG. Machine learning algorithms for accurate flow-based network traffic classification: Evaluation and comparison. Perform Eval. 2010; 67(6):451–67. sourceNot checked yet

  Not checked in the free run.

The accuracy metric is not reliable when the data is skewed with respect to the classes.

“However, the accuracy metric is not reliable when the data is skewed with respect to the classes.”

By the end of the decade, the application domains of ML started expanding to the operation and management of communication networks.

“By the end of the decade, the application domains of ML started expending to the operation and management of communication networks [57,217,289].”

• cites [57] Bernstein L, Yuhas CM. How technology shapes network management. IEEE Netw. 1989; 3(4):16–9. sourceNot checked yet

  Not checked in the free run.

• cites [217] Jennings A. A learning system for communications network configuration. Eng Appl Artif Intell. 1988; 1(3):151–60. sourceNot checked yet

  Not checked in the free run.

• cites [289] Macleish KJ. Mapping the integration of artificial intelligence into telecommunications. IEEE J Sel Areas Commun. 1988; 6(5):892–8. sourceNot checked yet

  Not checked in the free run.

As a result, the poor performance of a single predictor or classifier can be compensated with ensemble learning at the price of extra computation.

“As a result, the poor performance of a single predictor or classifier can be compensated with ensemble learning at the price of (significantly) extra computation.”

The comparison amongst the TSF methods shows that in general the NNE produces the lowest MAPE for both datasets.

“The comparison amongst the TSF methods show that in general the NNE produces the lowest MAPE for both datasets.”

• cites [104] Cortez P, Rio M, Rocha M, Sousa P. Internet traffic forecasting using neural networks. In: Proceedings of IEEE International Joint Conference on Neural Networks (IJCNN). IEEE: 2006. p. 2635–42. sourceNot checked yet

  Not checked in the free run.

In the last decade, different types of NNs (SLP, MLP, RNN, etc.) and other supervised techniques have been employed for TSF of network traffic.

“In the last decade, different types of NNs (SLP, MLP, RNN, etc.) and other supervised techniques have been employed for TSF of network traffic.”

These proofs showed that SLP-NN approximators, which employed a sufficient number of neurons of continuous sigmoidal activation type, were universal approximators, capable of approximating any continuous function to any desired accuracy.

“These proofs showed that SLP-NN approximators, which employed sufficient number of neurons of continuous sigmoidal activation type (a constraint introduced by Cybenko and relaxed by Hornik), were universal approximators, capable of approximating any continuous function to any desired accuracy.”

• cites [106] Cybenko G. Approximation by superpositions of a sigmoidal function. Mathematics of Control. Signals Syst (MCSS). 1989; 2(4):303–14. sourceNot checked yet

  Not checked in the free run.

• cites [196] Hornik K. Approximation capabilities of multilayer feedforward networks. Neural Netw. 1991; 4(2):251–7. sourceNot checked yet

  Not checked in the free run.

• cites [158] Funahashi KI. On the approximate realization of continuous mappings by neural networks. Neural netw. 1989; 2(3):183–92. sourceNot checked yet

  Not checked in the free run.

Yu et al. were the first to apply ML in traffic prediction using MLP-NN.

“To the best of our knowledge, Yu et al. [489] were the first to apply ML in traffic prediction using MLP-NN.”

• cites [489] Yu E, Chen CR. Traffic prediction using neural networks. In: Proceedings of IEEE GLOBECOM. IEEE: 1993. p. 991–5. sourceNot checked yet

  Not checked in the free run.

NN and GPR have an average TPR consistently above 80%.

“On the other hand, NN and GPR, have an average TPR consistently above 80%.”

• cites [365] Poupart P, Chen Z, Jaini P, Fung F, Susanto H, Geng Y, Chen L, Chen K, Jin H. Online flow size prediction for improved network routing.IEEE; 2016, pp. 1–6. sourceNot checked yet

  Not checked in the free run.

The combined NN and wavelet transform model reduces the peak inter-DC link utilization, i.e. the ISP's billed utilization, by about 9%.

“The combined NN and wavelet transform model reduces the peak inter-DC link utilization, i.e. the ISP's billed utilization, by about 9%.”

• cites [274] Li Y, Liu H, Yang W, Hu D, Xu W. Inter-data-center network traffic prediction with elephant flows: IEEE; 2016b, pp. 206–13. sourceNot checked yet

  Not checked in the free run.

oBMM outperforms all other approaches in terms of average TNR in all 3 datasets.

“However, it is worth noting that oBMM outperforms all other approaches in terms of average TNR in all 3 datasets.”

• cites [365] Poupart P, Chen Z, Jaini P, Fung F, Susanto H, Geng Y, Chen L, Chen K, Jin H. Online flow size prediction for improved network routing.IEEE; 2016, pp. 1–6. sourceNot checked yet

  Not checked in the free run.

The motive for flow size prediction in [365] is to discriminate elephant flows from mice flows in routing to speed up elephant flow completion time.

“The motive for flow size prediction in [365], is to discriminate elephant flows from mice flows in routing to speed up elephant flow completion time.”

• cites [365] Poupart P, Chen Z, Jaini P, Fung F, Susanto H, Geng Y, Chen L, Chen K, Jin H. Online flow size prediction for improved network routing.IEEE; 2016, pp. 1–6. sourceNot checked yet

  Not checked in the free run.

• AssumptionThe authors state this as an assumption: Mice flows are routed through Equal-cost multi-path routing, while elephant flows are routed through the least congested path.

A 14% improvement is observed when the size of the three first packets is used along with the header information.

“A 14% improvement is observed when the size of the three first packets is used along with the header information.”

• cites [365] Poupart P, Chen Z, Jaini P, Fung F, Susanto H, Geng Y, Chen L, Chen K, Jin H. Online flow size prediction for improved network routing.IEEE; 2016, pp. 1–6. sourceNot checked yet

  Not checked in the free run.

• AssumptionThe authors state this as an assumption: Mice flows are routed through Equal-cost multi-path routing, while elephant flows are routed through the least congested path.

k-step ahead predictions feeding k × 120 features into the NN show a relative RMSE (RRMSE) ranging from 4 to 10% for the NN-Wavelet transformation model as the prediction horizon varies from 30 s to 20 min.

“Thus, k-step ahead predictions, feed k × 120 features into the NN and show a relative RMSE (RRMSE) ranging from 4 to 10% for the NN-Wavelet transformation model as the prediction horizon varies from 30 s to 20 min.”

• cites [274] Li Y, Liu H, Yang W, Hu D, Xu W. Inter-data-center network traffic prediction with elephant flows: IEEE; 2016b, pp. 206–13. sourceNot checked yet

  Not checked in the free run.

Wavelet transformation reduces the average prediction errors for different prediction horizons by 5.4 and 2.9% for incoming and outgoing traffic, respectively.

“Evidently, wavelet transformation reduces the average prediction errors for different prediction horizons by 5.4 and 2.9% for incoming and outgoing traffic, respectively.”

• cites [274] Li Y, Liu H, Yang W, Hu D, Xu W. Inter-data-center network traffic prediction with elephant flows: IEEE; 2016b, pp. 206–13. sourceNot checked yet

  Not checked in the free run.

In the latter dataset, oBMM seems to suffer the most from class imbalance.

“In the latter dataset, oBMM seems to suffer the most from class imbalance.”

According to the authors, GPR improves the completion time by 6.6% in average for 99% of elephant flows when only the first packet header information is considered.

“According to the authors, GPR improves the completion time by 6.6% in average for 99% of elephant flows when only the first packet header information is considered.”

• cites [365] Poupart P, Chen Z, Jaini P, Fung F, Susanto H, Geng Y, Chen L, Chen K, Jin H. Online flow size prediction for improved network routing.IEEE; 2016, pp. 1–6. sourceNot checked yet

  Not checked in the free run.

• AssumptionThe authors state this as an assumption: Mice flows are routed through Equal-cost multi-path routing, while elephant flows are routed through the least congested path.

Therefore, though the motive was to promote flow count based traffic prediction to overcome the cost of monitoring traffic volume, the performance is compromised.

“Therefore, though the motive was to promote flow count based traffic prediction to overcome the cost of monitoring traffic volume, the performance is compromised.”

The experiments show noticeable discrepancies in the performance of the approaches with varying datasets.

“The experiments show noticeable discrepancies in the performance of the approaches with varying datasets.”

• cites [365] Poupart P, Chen Z, Jaini P, Fung F, Susanto H, Geng Y, Chen L, Chen K, Jin H. Online flow size prediction for improved network routing.IEEE; 2016, pp. 1–6. sourceNot checked yet

  Not checked in the free run.

Supervised NNs (including MLP and RNN) have been successfully applied to traffic prediction, as shown in Table 9.

“Supervised NNs (including MLP and RNN) have been successfully applied to traffic prediction, as shown in Table 9.”

The linear ARIMA model has prediction error of approximately 8.5 and 6.9% for incoming and outgoing traffic, respectively.

“In contrast, the linear ARIMA model has prediction error of approximately 8.5 and 6.9% for incoming and outgoing traffic, respectively.”

• cites [274] Li Y, Liu H, Yang W, Hu D, Xu W. Inter-data-center network traffic prediction with elephant flows: IEEE; 2016b, pp. 206–13. sourceNot checked yet

  Not checked in the free run.

Furthermore, the hybrid PSO-ABC has a faster training time than ABC or PSO.

“Furthermore, the hybrid PSO-ABC has a faster training time than ABC or PSO.”

• cites [500] Zhu Y, Zhang G, Qiu J. Network traffic prediction based on particle swarm bp neural network. JNW. 2013; 8(11):2685–91.Not checked yet

  Not checked in the free run.

Various classifiers leverage port numbers in conjunction with other techniques to improve the performance of traffic classifiers.

“Nevertheless, various classifiers leverage port numbers in conjunction with other techniques [31,56,244,417] to improve the performance of the traffic classifiers.”

• cites [31] Bakhshi T, Ghita B. On internet traffic classification: A two-phased machine learning approach. J Comput Netw Commun. 2016; 2016:2016.Not checked yet

  Not checked in the free run.

• cites [56] Bernaille L, Teixeira R, Salamatian K. Early application identification. In: Proceedings of the 2006 ACM CoNEXT Conference. ACM: 2006b. p. 61–6:12. sourceNot checked yet

  Not checked in the free run.

• cites [244] Kim H, Claffy KC, Fomenkov M, Barman D, Faloutsos M, Lee K. Internet traffic classification demystified: myths, caveats, and the best practices.ACM; 2008, p. 11. sourceNot checked yet

  Not checked in the free run.

• cites [417] Soysal M, Schmidt EG. Machine learning algorithms for accurate flow-based network traffic classification: Evaluation and comparison. Perform Eval. 2010; 67(6):451–67. sourceNot checked yet

  Not checked in the free run.

The SVM classifier using Pearson's Chi-square signatures distinguishes between the classes of interest with an average true positive rate of 99.6%.

“The signatures are then used to train a SVM classifier that distinguishes between the classes of interest with an average TP of 99.6%.”

• cites [146] Finamore A, Mellia M, Meo M, Rossi D. Kiss: Stochastic packet inspection classifier for udp traffic. IEEE/ACM Trans Netw. 2010; 18(5):1505–15. sourceNot checked yet

  Not checked in the free run.

TSF approaches, such as [52, 86, 104, 141, 500], where NNs are used to infer traffic volumes from past measured volumes, show high long-term and short-term prediction accuracy at low complexity with limited number of features and limited number of layers and neurons.

“TSF approaches, such as [52, 86, 104, 141, 500], where NNs are used to infer traffic volumes from past measured volumes, show high long-term and short-term prediction accuracy at low complexity with limited number of features and limited number of layers and neurons.”

• cites [52] Bermolen P, Rossi D. Support vector regression for link load prediction. Comput Netw. 2009; 53(2):191–201. sourceNot checked yet

  Not checked in the free run.

• cites [86] Chabaa S, Zeroual A, Antari J. Identification and prediction of internet traffic using artificial neural networks. J Int Learn Syst Appl. 2010; 2(03):147.Not checked yet

  Not checked in the free run.

• cites [104] Cortez P, Rio M, Rocha M, Sousa P. Internet traffic forecasting using neural networks. In: Proceedings of IEEE International Joint Conference on Neural Networks (IJCNN). IEEE: 2006. p. 2635–42. sourceNot checked yet

  Not checked in the free run.

• cites [141] Eswaradass A, Sun XH, Wu M. Network bandwidth predictor (nbp): A system for online network performance forecasting. In: Proceedings of 6th IEEE International Symposium on Cluster Computing and the Grid (CCGRID). IEEE: 2006. p. 4–pp. sourceNot checked yet

  Not checked in the free run.

• cites [500] Zhu Y, Zhang G, Qiu J. Network traffic prediction based on particle swarm bp neural network. JNW. 2013; 8(11):2685–91.Not checked yet

  Not checked in the free run.

Non-TSF approaches were investigated in [94,274,365] to infer traffic volumes from flow count and packet header fields.

“Non-TSF approaches were investigated in [94,274,365] to infer traffic volumes from flow count and packet header fields.”

• cites [94] Chen Z, Wen J, Geng Y. Predicting future traffic using hidden markov models. In: Proceddings of 24th IEEE International Conference on Network Protocols (ICNP). IEEE: 2016. p. 1–6. sourceNot checked yet

  Not checked in the free run.

• cites [274] Li Y, Liu H, Yang W, Hu D, Xu W. Inter-data-center network traffic prediction with elephant flows: IEEE; 2016b, pp. 206–13. sourceNot checked yet

  Not checked in the free run.

• cites [365] Poupart P, Chen Z, Jaini P, Fung F, Susanto H, Geng Y, Chen L, Chen K, Jin H. Online flow size prediction for improved network routing.IEEE; 2016, pp. 1–6. sourceNot checked yet

  Not checked in the free run.

It is cumbersome to manually maintain and adapt the signatures to the ever growing number of applications and their dynamics.

“Also, it is cumbersome to manually maintain and adapt the signatures to the ever growing number of applications and their dynamics [138].”

• cites [138] Erman J, Mahanti A, Arlitt M, Williamson C. Identifying and discriminating between web and peer-to-peer traffic in the network core. In: Proceedings of the 16th international conference on World Wide Web. ACM: 2007b. p. 883–92. sourceNot checked yet

  Not checked in the free run.

Generally, network traffic classification methodologies can be decomposed into four broad categories that leverage port number, packet payload, host behavior or flow features.

“Generally, network traffic classification methodologies can be decomposed into four broad categories that leverage port number, packet payload, host behavior or flow features [31,244].”

• cites [31] Bakhshi T, Ghita B. On internet traffic classification: A two-phased machine learning approach. J Comput Netw Commun. 2016; 2016:2016.Not checked yet

  Not checked in the free run.

• cites [244] Kim H, Claffy KC, Fomenkov M, Barman D, Faloutsos M, Lee K. Internet traffic classification demystified: myths, caveats, and the best practices.ACM; 2008, p. 11. sourceNot checked yet

  Not checked in the free run.

For instance, to predict the traffic for a particular flow f on link l, there must be a counter on link l actively measuring the traffic for that particular flow f, which can be challenging on very high speed links.

“For instance, in order to predict the traffic for a particular flow f on link l, there must be a counter on link l actively measuring the traffic for that particular flow f, which can be challenging on very high speed links.”

Payload-based traffic classification can be performed without any a priori knowledge of the application classes using unsupervised clustering.

“Ma et al. [286] show that payload-based traffic classification can be performed without any a priori knowledge of the application classes using unsupervised clustering.”

• cites [286] Ma J, Levchenko K, Kreibich C, Savage S, Voelker GM. Unexpected means of protocol inference. In: Proceedings of the 6th ACM SIGCOMM conference on Internet measurement.2006. p. 313–26. sourceNot checked yet

  Not checked in the free run.

• AssumptionThe authors state this as an assumption: For tractability, sessions are assumed to be finite and a protocol model is derived as a distribution on n byte flows rather than a pair of flows.

Relying solely on port numbers has been shown to be ineffective, largely due to the use of dynamic port negotiation, tunneling and misuse of port numbers assigned to well-known applications for obfuscating traffic and avoiding firewalls.

“Furthermore, relying solely on port numbers has been shown to be ineffective [125,228,320], largely due to the use of dynamic port negotiation, tunneling and misuse of port numbers assigned to well-known applications for obfuscating traffic and avoiding firewalls [54,109,176,286].”

• cites [125] Dreger H, Feldmann A, Mai M, Paxson V, Sommer R. Dynamic application-layer protocol analysis for network intrusion detection. In: USENIX Security Symposium. Berkeley: USENIX Security Symposium: 2006. p. 257–72.Not checked yet

  Not checked in the free run.

• cites [228] Karagiannis T, Broido A, Brownlee N, Claffy KC, Faloutsos M. Is p2p dying or just hiding?[p2p traffic measurement]. In: IEEE Global Telecommunications Conference (GLOBECOM), vol. 3.2004. p. 1532–8. sourceNot checked yet

  Not checked in the free run.

• cites [320] Moore AW, Papagiannaki K. Toward the accurate identification of network applications. In: PAM. Springer: 2005. p. 41–54. sourceNot checked yet

  Not checked in the free run.

• cites [54] Bernaille L, Teixeira R. Implementation issues of early application identification. Lect Notes Compu Sci. 2007; 4866:156. sourceNot checked yet

  Not checked in the free run.

• cites [109] Dainotti A, Pescape A, Claffy KC. Issues and future directions in traffic classification. IEEE Netw. 2012; 26(1):35–40. sourceNot checked yet

  Not checked in the free run.

• cites [176] Haffner P, Sen S, Spatscheck O, Wang D. Acas: automated construction of application signatures. In: Proceedings of the 2005 ACM SIGCOMM workshop on Mining network data. New York: ACM: 2005. p. 197–202.Not checked yet

  Not checked in the free run.

• cites [286] Ma J, Levchenko K, Kreibich C, Savage S, Voelker GM. Unexpected means of protocol inference. In: Proceedings of the 6th ACM SIGCOMM conference on Internet measurement.2006. p. 313–26. sourceNot checked yet

  Not checked in the free run.

Temporal and spatial portability of Bermolan et al.'s signatures is validated with marginal degradation in performance.

“Also, temporal and spatial portability of signatures is validated with marginal degradation in performance.”

• cites [53] Bermolen P, Mellia M, Meo M, Rossi D, Valenti S. Abacus: Accurate behavioral classification of P2P-tv traffic. Comput Netw. 2011; 55(6):1394–411. sourceNot checked yet

  Not checked in the free run.

NBKE incurs increased inference time, especially for classifying unknown flows.

“However, it incurs increased inference time, especially for classifying unknown flows [347].”

• cites [347] Park J, Tyan HR, Kuo CCJ. Internet traffic classification for scalable qos provision. In: Multimedia and Expo, vol 2006 IEEE International Conference on. IEEE: 2006. p. 1221–4. sourceNot checked yet

  Not checked in the free run.

NBKE outperforms NB with respect to training time.

“Moreover, it also outperforms NB with respect to training time.”

• cites [321] Moore AW, Zuev D. Internet traffic classification using bayesian analysis techniques. In: ACM SIGMETRICS Performance Evaluation Review, ACM, vol 33.2005. p. 50–60. sourceNot checked yet

  Not checked in the free run.

Service proximity, activity profiles, session duration, and periodicity act as good discriminators for a SVM classifier to differentiate between webmail and non-webmail traffic.

“The authors show that these features act as good discriminators for a SVM classifier to differentiate between webmail and nonwebmail traffic.”

• cites [404] Schatzmann D, Mühlbauer W, Spyropoulos T, Dimitropoulos X. Digging into https: Flow-based classification of webmail traffic. In: Proceedings of the 10th ACM SIGCOMM Conference on Internet Measurement: 2010. p. 322–27. sourceNot checked yet

  Not checked in the free run.

Host behavior-based traffic classification overcomes the limitations of unregistered or misused port numbers and encrypted packet payload by moving the observation point to the edge of the network and examining traffic between hosts.

“It overcomes the limitations of unregistered or misused port numbers and encrypted packet payload, by moving the observation point to the edge of the network and examining traffic between hosts (e.g.how many hosts are contacted, by which transport protocol, how many different ports are involved).”

It is not always possible to collect bidirectional flows due to routing asymmetries.

“It is not always possible to collect bidirectional flows due to routing asymmetries [138,347].”

• cites [138] Erman J, Mahanti A, Arlitt M, Williamson C. Identifying and discriminating between web and peer-to-peer traffic in the network core. In: Proceedings of the 16th international conference on World Wide Web. ACM: 2007b. p. 883–92. sourceNot checked yet

  Not checked in the free run.

• cites [347] Park J, Tyan HR, Kuo CCJ. Internet traffic classification for scalable qos provision. In: Multimedia and Expo, vol 2006 IEEE International Conference on. IEEE: 2006. p. 1221–4. sourceNot checked yet

  Not checked in the free run.

The higher false negative rate in Schatzmann et al.'s classifier is attributed to the inability of the classifier to distinguish between VPN and webmail servers.

“The higher FN is attributed to the inability of the classifier to distinguish between VPN and webmail servers.”

• cites [404] Schatzmann D, Mühlbauer W, Spyropoulos T, Dimitropoulos X. Digging into https: Flow-based classification of webmail traffic. In: Proceedings of the 10th ACM SIGCOMM Conference on Internet Measurement: 2010. p. 322–27. sourceNot checked yet

  Not checked in the free run.

DT suffers from overfitting with noisy data, which deteriorates performance.

“Unfortunately, DT suffers from overfitting with noisy data, which deteriorates performance [347].”

• cites [347] Park J, Tyan HR, Kuo CCJ. Internet traffic classification for scalable qos provision. In: Multimedia and Expo, vol 2006 IEEE International Conference on. IEEE: 2006. p. 1221–4. sourceNot checked yet

  Not checked in the free run.

The data exchanged amongst P2P applications is highly discriminative.

“The data exchanged amongst P2P applications is highly discriminative [53].”

• cites [53] Bermolen P, Mellia M, Meo M, Rossi D, Valenti S. Abacus: Accurate behavioral classification of P2P-tv traffic. Comput Netw. 2011; 55(6):1394–411. sourceNot checked yet

  Not checked in the free run.

Flow feature-based traffic classification has the potential to overcome numerous limitations of other techniques, such as unregistered port numbers, encrypted packet payload, routing asymmetries, and high storage and computational overhead.

“It has the potential to overcome numerous limitations of other techniques, such as unregistered port numbers, encrypted packet payload, routing asymmetries, high storage and computational overhead [55,138,176,347].”

• cites [55] Bernaille L, Teixeira R, Akodkenou I, Soule A, Salamatian K. Traffic classification on the fly. ACM SIGCOMM Comput Commun Rev. 2006a; 36(2):23–6. sourceNot checked yet

  Not checked in the free run.

• cites [138] Erman J, Mahanti A, Arlitt M, Williamson C. Identifying and discriminating between web and peer-to-peer traffic in the network core. In: Proceedings of the 16th international conference on World Wide Web. ACM: 2007b. p. 883–92. sourceNot checked yet

  Not checked in the free run.

• cites [176] Haffner P, Sen S, Spatscheck O, Wang D. Acas: automated construction of application signatures. In: Proceedings of the 2005 ACM SIGCOMM workshop on Mining network data. New York: ACM: 2005. p. 197–202.Not checked yet

  Not checked in the free run.

• cites [347] Park J, Tyan HR, Kuo CCJ. Internet traffic classification for scalable qos provision. In: Multimedia and Expo, vol 2006 IEEE International Conference on. IEEE: 2006. p. 1221–4. sourceNot checked yet

  Not checked in the free run.

NBKE classification accuracy deteriorates dramatically with increasing number of features.

“This is in contrast to NBKE, where the classification accuracy deteriorates dramatically with increasing number of features.”

• cites [347] Park J, Tyan HR, Kuo CCJ. Internet traffic classification for scalable qos provision. In: Multimedia and Expo, vol 2006 IEEE International Conference on. IEEE: 2006. p. 1221–4. sourceNot checked yet

  Not checked in the free run.

Bulk data transfer applications exhibit higher burst than streaming applications.

“Evidently, bulk data transfer applications exhibit higher burst than streaming applications.”

• cites [347] Park J, Tyan HR, Kuo CCJ. Internet traffic classification for scalable qos provision. In: Multimedia and Expo, vol 2006 IEEE International Conference on. IEEE: 2006. p. 1221–4. sourceNot checked yet

  Not checked in the free run.

NBKE with FCBF achieves temporal stability by classifying new flows collected twelve months later with an accuracy of 93% and a true positive rate of 98%.

“It also achieves temporal stability by classifying new flows collected twelve months later with an accuracy of 93% and TPR of 98%.”

• cites [321] Moore AW, Zuev D. Internet traffic classification using bayesian analysis techniques. In: ACM SIGMETRICS Performance Evaluation Review, ACM, vol 33.2005. p. 50–60. sourceNot checked yet

  Not checked in the free run.

Streaming applications often behave very similar to bulk data transfer applications.

“They notice that often streaming applications behave very similar to bulk data transfer applications.”

• cites [390] Roughan M, Sen S, Spatscheck O, Duffield N. Class-of-service mapping for qos: a statistical signature-based approach to ip traffic classification. In: Proceedings of the 4th ACM SIGCOMM conference on Internet measurement. ACM: 2004a. p. 135–148. sourceNot checked yet

  Not checked in the free run.

Flow feature-based traffic classification exploits the diversity and distinguishable characteristics of the traffic footprint generated by different applications.

“In essence, flow feature-based traffic classification exploits the diversity and distinguishable characteristics of the traffic footprint generated by different applications [31,467].”

• cites [31] Bakhshi T, Ghita B. On internet traffic classification: A two-phased machine learning approach. J Comput Netw Commun. 2016; 2016:2016.Not checked yet

  Not checked in the free run.

• cites [467] Wang Y, Xiang Y, Yu S. Internet traffic classification using machine learning: a token-based approach. IEEE: 2011. p. 285–9. sourceNot checked yet

  Not checked in the free run.

The accuracy of host behavior-based traffic classification strongly depends on the location of the monitoring system, especially because the observed communication pattern may be affected by routing asymmetries in the network core.

“However, the accuracy of the host behavior-based traffic classification strongly depends on the location of the monitoring system, especially since the observed communication pattern may be affected by routing asymmetries in the network core [229].”

• cites [229] Karagiannis T, Papagiannaki K, Faloutsos M. BLINC: multilevel traffic classification in the dark. ACM SIGCOMM Comput Commun Rev. 2005; 35(4):229–40. sourceNot checked yet

  Not checked in the free run.

The F-measure of BoF-NB outperforms NB irrespective of the aggregation rule.

“The F-measure, used to evaluate per class performance, of BoF-NB outperforms NB irrespective of the aggregation rule.”

• cites [496] Zhang J, Chen C, Xiang Y, Zhou W, Xiang Y. Internet traffic classification by aggregating correlated naive bayes predictions. IEEE Trans Inf Forensic Secur. 2013; 8(1):5–15. sourceNot checked yet

  Not checked in the free run.

Erman et al.'s classifier remains fairly robust over time for different traces.

“Furthermore, the classifier remains fairly robust over time for different traces.”

• cites [137] Erman J, Mahanti A, Arlitt M, Cohen I, Williamson C. Offline/realtime traffic classification using semi-supervised learning. Perform Eval. 2007a; 64(9):1194–213. sourceNot checked yet

  Not checked in the free run.

Erman et al. [137] propose a semi-supervised TCP traffic classification technique that partially overcomes the limitation of a priori knowledge of application class or protocol.

“[137] propose a semi-supervised TCP traffic classification technique that partially overcomes a limitation of [55] and [337]-a priori knowledge of application class or protocol.”

• cites [137] Erman J, Mahanti A, Arlitt M, Cohen I, Williamson C. Offline/realtime traffic classification using semi-supervised learning. Perform Eval. 2007a; 64(9):1194–213. sourceNot checked yet

  Not checked in the free run.

• cites [55] Bernaille L, Teixeira R, Akodkenou I, Soule A, Salamatian K. Traffic classification on the fly. ACM SIGCOMM Comput Commun Rev. 2006a; 36(2):23–6. sourceNot checked yet

  Not checked in the free run.

• cites [337] Nguyen TT, Armitage G, Branch P, Zander S. Timely and continuous machine-learning-based classification for interactive ip traffic. IEEE/ACM Trans Netw (TON). 2012; 20(6):1880–94. sourceNot checked yet

  Not checked in the free run.

Li et al. [270] classify TCP traffic into application classes, including unknown application classes, using a few packets in a flow.

“Similar to [137, 337], Li et al. [270] classify TCP traffic into application classes, including unknown application classes using a few packets in a flow.”

• cites [270] Li W, Moore AW. A machine learning approach for efficient traffic classification. In: Modeling, Analysis, and Simulation of Computer and Telecommunication Systems, 2007. MASCOTS’07. 15th International, Symposium on. IEEE: 2007. p. 310–7. sourceNot checked yet

  Not checked in the free run.

For traffic classification using NetMate features, both supervised NB and C4.5 DT classifiers perform well when evaluated with missing flow start, missing directionality, or 5% independent random packet loss.

“Both classifiers perform well when evaluated with missing flow start, missing directionality, or 5% independent random packet loss.”

• cites [337] Nguyen TT, Armitage G, Branch P, Zander S. Timely and continuous machine-learning-based classification for interactive ip traffic. IEEE/ACM Trans Netw (TON). 2012; 20(6):1880–94. sourceNot checked yet

  Not checked in the free run.

In offline classification, Erman et al. [137] achieve over 94% flow accuracy with just two randomly labeled flows per cluster, amongst a mix of 64,000 unlabeled flows and k = 400.

“In offline classification, the authors in [137] achieve over 94% flow accuracy with just two randomly labeled flows per cluster, amongst a mix of 64,000 unlabeled flows and k = 400.”

• cites [137] Erman J, Mahanti A, Arlitt M, Cohen I, Williamson C. Offline/realtime traffic classification using semi-supervised learning. Perform Eval. 2007a; 64(9):1194–213. sourceNot checked yet

  Not checked in the free run.

It is infeasible to expect complete a priori information for applications, since network operators often do not know all the applications that are running in the network.

“However, it is infeasible to expect complete a priori information for applications, since often network operators do not even know all the applications that are running in the network.”

The DIFFUSE technique lacks flexibility because its classifier can only recognize application classes that were known a priori.

“Despite the high accuracy, this technique lacks in flexibility, since the classifier can only recognize the application classes that were known a priori.”

• cites [337] Nguyen TT, Armitage G, Branch P, Zander S. Timely and continuous machine-learning-based classification for interactive ip traffic. IEEE/ACM Trans Netw (TON). 2012; 20(6):1880–94. sourceNot checked yet

  Not checked in the free run.

RTC can achieve higher flow and byte accuracy of up to 15% and 10%, respectively, compared to the second best technique, even with small labeled training datasets.

“Therefore, even with small labeled training datasets, RTC can achieve a higher flow and byte accuracy of up to 15% and 10%, respectively, in comparison to the second best technique.”

• cites [497] Zhang J, Chen X, Xiang Y, Zhou W, Wu J. Robust network traffic classification. IEEE/ACM Trans Netw (TON). 2015; 23(4):1257–70. sourceNot checked yet

  Not checked in the free run.

Erman et al. integrate their layered online classification in Bro and achieve byte accuracies in the 70-90% range.

“The authors integrate their layered online classification in Bro and achieve byte accuracies in the 70-90% range.”

• cites [137] Erman J, Mahanti A, Arlitt M, Cohen I, Williamson C. Offline/realtime traffic classification using semi-supervised learning. Perform Eval. 2007a; 64(9):1194–213. sourceNot checked yet

  Not checked in the free run.

DIFFUSE achieves a high accuracy of 98-99% for Enemy Territory and VoIP traffic replayed across the network while monitoring one or more 1 Gb/s links.

“DIFFUSE achieves a high accuracy of 98-99% for Enemy Territory (online game) and VoIP traffic replayed across the network, while monitoring one or more 1 Gb/s links.”

• cites [337] Nguyen TT, Armitage G, Branch P, Zander S. Timely and continuous machine-learning-based classification for interactive ip traffic. IEEE/ACM Trans Netw (TON). 2012; 20(6):1880–94. sourceNot checked yet

  Not checked in the free run.

The accuracy of service classification is between 95-100% for the majority of providers.

“Furthermore, the accuracy of service classification is between 95-100% for majority of the providers.”

• cites [409] Shbair WM, Cholez T, Francois J, Chrisment I. A multi-level framework to identify https services. In: IEEE/IFIP Network Operations and Management Symposium (NOMS)2016. p. 240–8. sourceNot checked yet

  Not checked in the free run.

When trained and tested across datasets, C4.5 DT achieves 83.7% DR and 1.5% FPR.

“Furthermore, when trained and tested across datasets (i.e.across networks), it achieves 83.7% DR and 1.5% FPR.”

• cites [17] Alshammari R, Zincir-Heywood AN. Machine learning based encrypted traffic classification: Identifying ssh and skype. In: Computational Intelligence for Security and Defense Applications, 2009. CISDA 2009. IEEE Symposium on. IEEE: 2009. p. 1–8. sourceNot checked yet

  Not checked in the free run.

With a larger time difference between training and testing data collection, error rates increase to 5.5% for TCP and 1.2% for UDP traffic.

“However, with a larger time difference between training and testing data collection, the error rates increase to 5.5 and 1.2% for TCP and UDP traffic, respectively.”

• cites [222] Jin Y, Duffield N, Erman J, Haffner P, Sen S, Zhang ZL. A modular machine learning system for flow-level traffic classification in large networks. ACM Trans Knowl Discov Data (TKDD). 2012; 6(1):4.Not checked yet

  Not checked in the free run.

Alshammari et al. [17] do not detect the precise applications within a secure session.

“Unfortunately, Alshammari et al. [17] do not detect the precise applications within a secure session.”

• cites [17] Alshammari R, Zincir-Heywood AN. Machine learning based encrypted traffic classification: Identifying ssh and skype. In: Computational Intelligence for Security and Defense Applications, 2009. CISDA 2009. IEEE Symposium on. IEEE: 2009. p. 1–8. sourceNot checked yet

  Not checked in the free run.

C4.5 DT generalizes well from one network to another for SSH traffic.

“Hence, it generalizes well from one network to another.”

• cites [17] Alshammari R, Zincir-Heywood AN. Machine learning based encrypted traffic classification: Identifying ssh and skype. In: Computational Intelligence for Security and Defense Applications, 2009. CISDA 2009. IEEE Symposium on. IEEE: 2009. p. 1–8. sourceNot checked yet

  Not checked in the free run.

RF performs better than C4.5 DT with a precision of 92.6%, recall of 92.8%, and F-measure of 92.6%, to classify service providers with selected features.

“In the evaluation, RF performs better in comparison to C4.5 DT with a precision of 92.6%, recall of 92.8%, and F-measure of 92.6%, to classify service providers with selected features.”

• cites [409] Shbair WM, Cholez T, Francois J, Chrisment I. A multi-level framework to identify https services. In: IEEE/IFIP Network Operations and Management Symposium (NOMS)2016. p. 240–8. sourceNot checked yet

  Not checked in the free run.

Jointly, NB and χ2 classifiers outperform the classifiers in isolation by detecting Skype voice traffic over UDP and TCP with nearly zero false positives.

“Jointly, NB and χ 2 classifier outperform the classifiers in isolation by detecting Skype voice traffic over UDP and TCP with nearly zero FP.”

• cites [69] Bonfiglio D, Mellia M, Meo M, Rossi D, Tofanelli P. Revealing skype traffic: when randomness plays with you. In: ACM SIGCOMM Computer Communication Review. ACM: 2007. p. 37–48. sourceNot checked yet

  Not checked in the free run.

C4.5 DT performed well for Skype traffic with 98.4% DR and 7.8% FPR in the Dalhousie dataset.

“The C4.5 DT classifier also performed well for Skype traffic with 98.4% DR and 7.8% FPR in the Dalhousie dataset.”

• cites [17] Alshammari R, Zincir-Heywood AN. Machine learning based encrypted traffic classification: Identifying ssh and skype. In: Computational Intelligence for Security and Defense Applications, 2009. CISDA 2009. IEEE Symposium on. IEEE: 2009. p. 1–8. sourceNot checked yet

  Not checked in the free run.

C4.5 DT outperforms the other classifiers for the majority of datasets in Alshammari et al.'s evaluation.

“In the evaluation, C4.5 DT outperform the other classifiers for the majority of datasets.”

• cites [17] Alshammari R, Zincir-Heywood AN. Machine learning based encrypted traffic classification: Identifying ssh and skype. In: Computational Intelligence for Security and Defense Applications, 2009. CISDA 2009. IEEE Symposium on. IEEE: 2009. p. 1–8. sourceNot checked yet

  Not checked in the free run.

Unsupervised ML techniques have been explored for practical traffic classification using flow features.

“Therefore, unsupervised ML techniques have been explored for practical traffic classification using flow features.”

• AssumptionRests on an assumption: Network operators rarely have complete information about all the applications in their network.

Since flow features from applications can exhibit high similarity, it is unrealistic to apply hard clustering for fine-grained traffic classification.

“Since flow features from applications can exhibit high similarity, it is unrealistic to apply hard clustering for fine-grained traffic classification.”

For traffic classification with unsupervised ML, both hard and soft clustering techniques have been investigated.

“For traffic classification with unsupervised ML, both hard and soft clustering techniques have been investigated.”

Flow feature-based traffic classification can be achieved using only a small number of packets in a flow, rather than the complete flow.

“Flow feature-based traffic classification can be achieved using only a small of number of packets in a flow, rather than the complete flow.”

Long-lived UDP traffic lends itself to supervised payload-based traffic classification, where payload is inspected randomly in an observation window.

“On the other hand, long-lived UDP traffic lends itself to supervised payload-based traffic classification, where payload is inspected randomly in an observation window [146].”

• cites [146] Finamore A, Mellia M, Meo M, Rossi D. Kiss: Stochastic packet inspection classifier for udp traffic. IEEE/ACM Trans Netw. 2010; 18(5):1505–15. sourceNot checked yet

  Not checked in the free run.

A hierarchical approach to traffic classification reduces the effort required to retrain classifiers when a new service is added.

“This hierarchical approach reduces the effort required to retrain the classifiers in the event of an addition of a new service.”

• cites [409] Shbair WM, Cholez T, Francois J, Chrisment I. A multi-level framework to identify https services. In: IEEE/IFIP Network Operations and Management Symposium (NOMS)2016. p. 240–8. sourceNot checked yet

  Not checked in the free run.

Traditional and multi-class SVM have been applied jointly to increase the accuracy of traffic classification and its applicability to large datasets.

“Similarly, traditional and multi-class SVM have been applied jointly to increase the accuracy of traffic classification and its applicability to large datasets [464].”

• cites [464] Wang R, Liu Y, Yang Y, Zhou X. Solving the app-level classification problem of p2p traffic via optimized support vector machines. In: Intelligent Systems Design and Applications, 2006. ISDA’06. Sixth International Conference on, IEEE, vol 2: 2006. p. 534–9. sourceNot checked yet

  Not checked in the free run.

It is impractical to expect complete a priori knowledge about all applications for traffic classification.

“Therefore, it is impractical to expect complete a priori knowledge about all applications for traffic classification.”

• AssumptionRests on an assumption: Network operators rarely have complete information about all the applications in their network.

Port-based techniques are unreliable and antiquated, largely due to dynamic port negotiation, tunneling and misuse of port numbers assigned to well-known applications for obfuscating traffic and avoiding firewalls.

“Portbased techniques are unreliable and antiquated, largely due to the use of dynamic port negotiation, tunneling and misuse of port numbers assigned to well-known applications for obfuscating traffic and avoiding firewalls [54,109,176,286].”

• cites [54] Bernaille L, Teixeira R. Implementation issues of early application identification. Lect Notes Compu Sci. 2007; 4866:156. sourceNot checked yet

  Not checked in the free run.

• cites [109] Dainotti A, Pescape A, Claffy KC. Issues and future directions in traffic classification. IEEE Netw. 2012; 26(1):35–40. sourceNot checked yet

  Not checked in the free run.

• cites [176] Haffner P, Sen S, Spatscheck O, Wang D. Acas: automated construction of application signatures. In: Proceedings of the 2005 ACM SIGCOMM workshop on Mining network data. New York: ACM: 2005. p. 197–202.Not checked yet

  Not checked in the free run.

Q-routing does not require any prior knowledge of the network topology or traffic patterns.

“Q-routing does not require any prior knowledge of the network topology or traffic patterns.”

• cites [71] Boyan JA, Littman ML. Packet routing in dynamically changing networks: A reinforcement learning approach. In: Advances in neural information processing systems: 1994. p. 671–8.Not checked yet

  Not checked in the free run.

• cites [280] Littman M, Boyan J. A distributed reinforcement learning scheme for network routing. In: Proceedings of the international workshop on applications of neural networks to telecommunications. Psychology Press: 1993. p. 45–51.Not checked yet

  Not checked in the free run.

The original Q-routing shows better stability and robustness to topology changes under higher loads.

“On the contrary, the original Q-routing shows better stability and robustness to topology changes under higher loads.”

• cites [71] Boyan JA, Littman ML. Packet routing in dynamically changing networks: A reinforcement learning approach. In: Advances in neural information processing systems: 1994. p. 671–8.Not checked yet

  Not checked in the free run.

• cites [280] Littman M, Boyan J. A distributed reinforcement learning scheme for network routing. In: Proceedings of the international workshop on applications of neural networks to telecommunications. Psychology Press: 1993. p. 45–51.Not checked yet

  Not checked in the free run.

It is possible to circumvent the challenges that encryption poses to flow feature-based traffic classification.

“However, it is possible to circumvent these challenges.”

DRQ-Routing integrates dual reinforcement learning with Q-routing, so that nodes along the route between the source and the destination receive feedbacks in both directions.

“DRQ-Routing integrates dual reinforcement learning [167] with Q-routing, so that nodes along the route between the source and the destination receive feedbacks in both directions (i.e. from both the up-stream and down-stream nodes).”

• cites [254] Kumar S, Miikkulainen R. Dual reinforcement q-routing: An on-line adaptive routing algorithm. In: Proceedings of the artificial neural networks in engineering Conference.1997. p. 231–8.Not checked yet

  Not checked in the free run.

• cites [167] Goetz P, Kumar S, Miikkulainen R. On-line adaptation of a signal predistorter through dual reinforcement learning. In: ICML.1996. p. 175–81.Not checked yet

  Not checked in the free run.

A hierarchical method that identifies service provider followed by type of service, using statistical features from encrypted payload, has been highly accurate and temporally stable.

“For instance, a hierarchical method that identifies service provider followed by type of service, using statistical features from encrypted payload has been highly accurate and temporally stable [409].”

• cites [409] Shbair WM, Cholez T, Francois J, Chrisment I. A multi-level framework to identify https services. In: IEEE/IFIP Network Operations and Management Symposium (NOMS)2016. p. 240–8. sourceNot checked yet

  Not checked in the free run.

Encryption can complicate flow feature-based traffic classification.

“Similar to payload-based traffic classification, encryption can also complicate flow feature-based traffic classification.”

Experiments on a 36-node network demonstrated that Q-routing outperforms the shortest path first routing algorithm in terms of average packet delivery time.

“However, experiments on a 36-node network demonstrated the Q-routing outperforms the shortest path first routing algorithm in terms of average packet delivery time.”

• cites [71] Boyan JA, Littman ML. Packet routing in dynamically changing networks: A reinforcement learning approach. In: Advances in neural information processing systems: 1994. p. 671–8.Not checked yet

  Not checked in the free run.

• cites [280] Littman M, Boyan J. A distributed reinforcement learning scheme for network routing. In: Proceedings of the international workshop on applications of neural networks to telecommunications. Psychology Press: 1993. p. 45–51.Not checked yet

  Not checked in the free run.

Joint application of supervised and unsupervised ML for traffic classification has demonstrated success.

“Consequentially, joint application of supervised and unsupervised ML for traffic classification [137, 497] has demonstrated success.”

• cites [137] Erman J, Mahanti A, Arlitt M, Cohen I, Williamson C. Offline/realtime traffic classification using semi-supervised learning. Perform Eval. 2007a; 64(9):1194–213. sourceNot checked yet

  Not checked in the free run.

• cites [497] Zhang J, Chen X, Xiang Y, Zhou W, Wu J. Robust network traffic classification. IEEE/ACM Trans Netw (TON). 2015; 23(4):1257–70. sourceNot checked yet

  Not checked in the free run.

Sub-flow-based traffic classification achieves high accuracy using the fast and efficient C4.5 DT classifier with correlation-based filtering.

“Sub-flow-based traffic classification achieves high accuracy using fast and efficient C4.5 DT classifier with correlation-based filtering.”

QELAR is 25% more energy efficient than VBF.

“Results show that QELAR is 25% more energy efficient than VBF.”

• cites [197] Hu T, Fei Y. Qelar: a machine-learning-based adaptive routing protocol for energy-efficient and lifetime-extended underwater sensor networks. IEEE Trans Mob Comput. 2010; 9(6):796–809. sourceNot checked yet

  Not checked in the free run.

Q-PR preserves the lifetime of the WSN better, 3× and 4× more than GPSR and EFE respectively.

“Thus the Q-PR algorithm preserves better the lifetime of the WSN (3× and 4× more than GPSR and EFE respectively).”

• cites [24] Arroyo-Valles R, Alaiz-Rodriguez R, Guerrero-Curieses A, Cid-Sueiro J. Q-probabilistic routing in wireless sensor networks. In: Intelligent Sensors, Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on. IEEE: 2007. p. 1–6. sourceNot checked yet

  Not checked in the free run.

The problem of ML-based multicast routing was first addressed by Sun et al. in the context of MANETs.

“The problem of ML-based multicast routing was first addressed by Sun et al. [430] in the context of MANETs.”

• cites [430] Sun R, Tatsumi S, Zhao G. Q-map: A novel multicast routing method in wireless ad hoc networks with multiagent reinforcement learning. In: TENCON’02. Proceedings. 2002 IEEE Region 10 Conference on Computers, Communications, Control and Power Engineering, vol 1. IEEE: 2002. p. 667–670.Not checked yet

  Not checked in the free run.

The lifetime of the network is 25%∼30% higher with QELAR in the presence of failures and network partition compared with VBF with comparable transmission range.

“The lifetime of the network is 25% ∼ 30% higher with QELAR in the presence of failures and network partition compared with VBF with comparable transmission range, which makes QELAR more robust to faults.”

• cites [197] Hu T, Fei Y. Qelar: a machine-learning-based adaptive routing protocol for energy-efficient and lifetime-extended underwater sensor networks. IEEE Trans Mob Comput. 2010; 9(6):796–809. sourceNot checked yet

  Not checked in the free run.

Arroyo-Valles et al. propose Q-PR, a localization-aware routing scheme for WSNs that applies Q-learning to achieve a trade-off between packet delivery rate, expected transmission count, and network lifetime.

“Arroyo-Valles et al. [24] propose Q-probabilistic routing (Q-PR), a localization-aware routing scheme for WSNs that applies Q-learning to achieve a trade-off between packet delivery rate, expected transmission count (ETX), and network lifetime.”

• cites [24] Arroyo-Valles R, Alaiz-Rodriguez R, Guerrero-Curieses A, Cid-Sueiro J. Q-probabilistic routing in wireless sensor networks. In: Intelligent Sensors, Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on. IEEE: 2007. p. 1–6. sourceNot checked yet

  Not checked in the free run.

Q-PR requires each node to maintain locally a number of information regarding each of its neighbors, which hampers the scalability of the approach.

“However the algorithm requires that each node maintains locally a number of information regarding each of its neighbors.These include the distance between the nodes, the distance of the neighbor to the sink, the delivery probability between nodes, the estimated residual energy at the neighboor, and the 2-hop neighboors.This hampers the scalability of the approach.”

• cites [24] Arroyo-Valles R, Alaiz-Rodriguez R, Guerrero-Curieses A, Cid-Sueiro J. Q-probabilistic routing in wireless sensor networks. In: Intelligent Sensors, Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on. IEEE: 2007. p. 1–6. sourceNot checked yet

  Not checked in the free run.

Taking into account the residual energy helps achieve a balanced energy distribution among nodes by avoiding highly utilized routes (hotspots).

“Taking into account the residual energy helps achieve a balanced energy distribution among nodes by avoiding highly utilized routes (hotspots).”

• cites [197] Hu T, Fei Y. Qelar: a machine-learning-based adaptive routing protocol for energy-efficient and lifetime-extended underwater sensor networks. IEEE Trans Mob Comput. 2010; 9(6):796–809. sourceNot checked yet

  Not checked in the free run.

TPOT-RL, proposed by Stone and Veloso for RoboCup-1998, is the first fully collaborative MARL technique to be applied to packet routing.

“Team-partitioned opaque-transition reinforcement learning (TPOT-RL), proposed by Stone and Veloso for the RoboCup-1998 (Robot Soccer World Cup II) [425], is the first fully collaborative MARL technique to be applied to packet routing [424].”

• cites [425] Stone P, Veloso M. Team-partitioned, opaque-transition reinforcement learning. In: Proceedings of the third annual conference on Autonomous Agents. ACM: 1999. p. 206–12. sourceNot checked yet

  Not checked in the free run.

• cites [424] Stone P. Tpot-rl applied to network routing. In: ICML.2000. p. 935–42. sourceNot checked yet

  Not checked in the free run.

Q-PR shows a lower number of retransmission retries and acknowledgements, on average over 50% and 40% less than GPSR and EFE respectively.

“Moreover, Q-PR shows lower number of retransmission retries and acknowledgements (on average over 50% and 40% less than GPSR and EFE respectively).”

• cites [24] Arroyo-Valles R, Alaiz-Rodriguez R, Guerrero-Curieses A, Cid-Sueiro J. Q-probabilistic routing in wireless sensor networks. In: Intelligent Sensors, Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on. IEEE: 2007. p. 1–6. sourceNot checked yet

  Not checked in the free run.

The traditional single-agent RL model, which is greedy in nature, provides local optimizations regardless of global performance.

“The traditional single-agent RL model, which is greedy in nature, provides local optimizations regardless of the global performance.”

Q-MAP, a Q-learning-based algorithm, was proposed to find and build the optimal multicast tree in MANETs.

“Q-MAP, a Q-learning-based algorithm, was proposed to find and build the optimal multicast tree in MANETs.”

• cites [430] Sun R, Tatsumi S, Zhao G. Q-map: A novel multicast routing method in wireless ad hoc networks with multiagent reinforcement learning. In: TENCON’02. Proceedings. 2002 IEEE Region 10 Conference on Computers, Communications, Control and Power Engineering, vol 1. IEEE: 2002. p. 667–670.Not checked yet

  Not checked in the free run.

QELAR is more robust to faults than VBF.

“which makes QELAR more robust to faults”

• cites [197] Hu T, Fei Y. Qelar: a machine-learning-based adaptive routing protocol for energy-efficient and lifetime-extended underwater sensor networks. IEEE Trans Mob Comput. 2010; 9(6):796–809. sourceNot checked yet

  Not checked in the free run.

The traditional single-agent RL model is not sufficient to achieve global optimizations such as network lifetime maximization or network-wide QoS provisioning.

“Therefore, it is not sufficient to achieve global optimizations such as network lifetime maximization or network-wide QoS provisioning.”

Q-routing and Q-routing-inspired approaches like PR-routing and DRQ-Routing use a form of MARL, as Q-values are exchanged between neighboring nodes.

“Q-routing and Q-routing-inspired approaches like PR-routing and DRQ-Routing do use a form of MARL, as Q-values are exchanged between neighboring nodes.”

d-AdaptOR consistently outperforms existing adaptive routing algorithms in terms of number of retransmissions per packet.

“Simulations show that d-AdaptOR consistently outperforms existing adaptive routing algorithms, in terms of number of retransmissions per packet.”

• cites [59] Bhorkar AA, Naghshvar M, Javidi T, Rao BD. Adaptive opportunistic routing for wireless ad hoc networks. IEEE/ACM Trans Netw. 2012; 20(1):243–56. sourceNot checked yet

  Not checked in the free run.

Q-PR achieves approximately 20% lower overhead compared to directed diffusion.

“Compared to directed diffusion: • up to 5× higher delivery rate • ≈ 20% lower overhead Q-PR [24]”

• cites [24] Arroyo-Valles R, Alaiz-Rodriguez R, Guerrero-Curieses A, Cid-Sueiro J. Q-probabilistic routing in wireless sensor networks. In: Intelligent Sensors, Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on. IEEE: 2007. p. 1–6. sourceNot checked yet

  Not checked in the free run.

After convergence, DRQ-routing minimizes end-to-end delay.

“Simulation results show that after convergence, DRQ-routing minimizes end-to-end delay”

• cites [482] Xia B, Wahab MH, Yang Y, Fan Z, Sooriyabandara M. Reinforcement learning based spectrum-aware routing in multi-hop cognitive radio networks. In: Cognitive Radio Oriented Wireless Networks and Communications, 2009. CROWNCOM’09. 4th International, Conference on. IEEE: 2009. p. 1–5. sourceNot checked yet

  Not checked in the free run.

Q-learning is more sensitive to the initial choice of Q-values than AdaR.

“the authors show that Q-learning is more sensitive to the initial choice of Q-values than AdaR.”

• cites [461] Wang P, Wang T. Adaptive routing for sensor networks using reinforcement learning. In: Computer and Information Technology, 2006. CIT’06. The Sixth IEEE International Conference on. IEEE: 2006. p. 219. sourceNot checked yet

  Not checked in the free run.

AdaR converges faster than Q-learning.

“Results show that the algorithm converges faster than Q-learning”

• cites [461] Wang P, Wang T. Adaptive routing for sensor networks using reinforcement learning. In: Computer and Information Technology, 2006. CIT’06. The Sixth IEEE International Conference on. IEEE: 2006. p. 219. sourceNot checked yet

  Not checked in the free run.

Reducing the frequency of message exchange by 95% only slightly affects convergence time (1∼2sec).

“The convergence time seems to be only slightly affected (1 ∼ 2sec).”

• cites [24] Arroyo-Valles R, Alaiz-Rodriguez R, Guerrero-Curieses A, Cid-Sueiro J. Q-probabilistic routing in wireless sensor networks. In: Intelligent Sensors, Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on. IEEE: 2007. p. 1–6. sourceNot checked yet

  Not checked in the free run.

DRQ-routing is faster to converge than Q-routing (50% faster at lower activity level).

“is faster to converge than Q-routing (50% faster at lower activity level)”

• cites [482] Xia B, Wahab MH, Yang Y, Fan Z, Sooriyabandara M. Reinforcement learning based spectrum-aware routing in multi-hop cognitive radio networks. In: Cognitive Radio Oriented Wireless Networks and Communications, 2009. CROWNCOM’09. 4th International, Conference on. IEEE: 2009. p. 1–5. sourceNot checked yet

  Not checked in the free run.

Q-learning took 40 episodes to reach comparable success rates to AdaR.

“whereas, it took 40 episodes for Q-learning to reach comparable success rates.”

• cites [461] Wang P, Wang T. Adaptive routing for sensor networks using reinforcement learning. In: Computer and Information Technology, 2006. CIT’06. The Sixth IEEE International Conference on. IEEE: 2006. p. 219. sourceNot checked yet

  Not checked in the free run.

In a simulated multi-hop network with 2 sources transmitting videos to the same destination, reducing the frequency of message exchange by 95% (from every 1ms to every 20ms) increases goodput and effective data rate by over 40%.

“Results show that by reducing the frequency of message exchange by 95% (from every 1ms to every 20ms), the goodput and effective data rate are increased by over 40%”

• cites [24] Arroyo-Valles R, Alaiz-Rodriguez R, Guerrero-Curieses A, Cid-Sueiro J. Q-probabilistic routing in wireless sensor networks. In: Intelligent Sensors, Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on. IEEE: 2007. p. 1–6. sourceNot checked yet

  Not checked in the free run.

AdaR is more sensitive to link failures and node mobility.

“which also makes it more sensitive to link failures and node mobility.”

• cites [461] Wang P, Wang T. Adaptive routing for sensor networks using reinforcement learning. In: Computer and Information Technology, 2006. CIT’06. The Sixth IEEE International Conference on. IEEE: 2006. p. 219. sourceNot checked yet

  Not checked in the free run.

QAR is a centralized SARSA with softmax policy selection applied to achieve QoS-aware adaptive routing in SDN.

“a centralized SARSA with a softmax policy selection algorithm has been applied by Lin et al. [276] to achieve QoS-aware adaptive routing (QAR) in SDN.”

• cites [276] Lin SC, Akyildiz IF, Wang P, Luo M. Qos-aware adaptive routing in multi-layer hierarchical software defined networks: a reinforcement learning approach. In: Services Computing (SCC) 2016, IEEE International Conference on. IEEE: 2016. p. 25–33. sourceNot checked yet

  Not checked in the free run.

DRQ-routing significantly reduces end-to-end delay compared to SP-routing at higher activity levels.

“it significantly reduces end-to-end delay compared to SP-routing at higher activity levels”

• cites [482] Xia B, Wahab MH, Yang Y, Fan Z, Sooriyabandara M. Reinforcement learning based spectrum-aware routing in multi-hop cognitive radio networks. In: Cognitive Radio Oriented Wireless Networks and Communications, 2009. CROWNCOM’09. 4th International, Conference on. IEEE: 2009. p. 1–5. sourceNot checked yet

  Not checked in the free run.

AdaR reaches a routing success rate of ∼95% with low deviation before the 5th learning episode.

“a routing success rate of ∼ 95% with a low deviation was reached even before the 5th learning episode”

• cites [461] Wang P, Wang T. Adaptive routing for sensor networks using reinforcement learning. In: Computer and Information Technology, 2006. CIT’06. The Sixth IEEE International Conference on. IEEE: 2006. p. 219. sourceNot checked yet

  Not checked in the free run.

Routing in wireless ad hoc networks calls for decentralized RL where the learning capability is distributed among the routing nodes.

“routing in wireless ad hoc networks calls for decentralized RL [59,277] where the learning capability is distributed among the routing nodes.”

• cites [59] Bhorkar AA, Naghshvar M, Javidi T, Rao BD. Adaptive opportunistic routing for wireless ad hoc networks. IEEE/ACM Trans Netw. 2012; 20(1):243–56. sourceNot checked yet

  Not checked in the free run.

• cites [277] Lin Z, van der Schaar M. Autonomic and distributed joint routing and power control for delay-sensitive applications in multi-hop wireless networks. IEEE Trans Wirel Commun. 2011; 10(1):102–13. sourceNot checked yet

  Not checked in the free run.

The assumption that RTT values never change during measurement is unrealistic because a modification in the return path changes RTT values without affecting the cause of packet loss.

“This is an unrealistic assumption since a modification in the return path changes the RTT values without affecting the cause of packet loss.”

In SDN and WSN, the existence of a central node allows for centralized learning.

“in SDN [276] as well as WSN, the existence of a central node-the controller in SDN and the sink in WSN, respectively-allows for centralized learning.”

• cites [276] Lin SC, Akyildiz IF, Wang P, Luo M. Qos-aware adaptive routing in multi-layer hierarchical software defined networks: a reinforcement learning approach. In: Services Computing (SCC) 2016, IEEE International Conference on. IEEE: 2016. p. 25–33. sourceNot checked yet

  Not checked in the free run.

Hiding error losses may violate the TCP end-to-end principle as it may require splitting the TCP connection by sending an ACK before the packet arrives at the receiver.

“hiding error losses may violate TCP end-to-end principle as it may require splitting the TCP connection by sending an ACK to the sender before the packet arrives at the receiver [129].”

• cites [129] El Khayat I, Geurts P, Leduc G. Improving TCP in Wireless Networks with an Adaptive Machine-Learnt Classifier of Packet Loss Causes. Berlin, Heidelberg: Springer Berlin Heidelberg; 2005, pp. 549–60. sourceNot checked yet

  Not checked in the free run.

A major problem in TCP is that it recognizes and handles all packet losses as network congestion, i.e., buffer overflow.

“the major problem in TCP is that it recognizes and handles all packet losses as network congestion, that is buffer overflow.”

Utility functions such as QoS provisioning, load balancing and network lifetime maximization require more information to be disseminated at the cost of increased complexity and communication overhead.

“utility functions such as QoS provisioning, load balancing and network lifetime maximization, as in Q-PR [24], QELAR [197,277], require more information to be disseminated at the cost of an increased complexity and communication overhead.”

• cites [24] Arroyo-Valles R, Alaiz-Rodriguez R, Guerrero-Curieses A, Cid-Sueiro J. Q-probabilistic routing in wireless sensor networks. In: Intelligent Sensors, Sensor Networks and Information, 2007. ISSNIP 2007. 3rd International Conference on. IEEE: 2007. p. 1–6. sourceNot checked yet

  Not checked in the free run.

• cites [197] Hu T, Fei Y. Qelar: a machine-learning-based adaptive routing protocol for energy-efficient and lifetime-extended underwater sensor networks. IEEE Trans Mob Comput. 2010; 9(6):796–809. sourceNot checked yet

  Not checked in the free run.

• cites [277] Lin Z, van der Schaar M. Autonomic and distributed joint routing and power control for delay-sensitive applications in multi-hop wireless networks. IEEE Trans Wirel Commun. 2011; 10(1):102–13. sourceNot checked yet

  Not checked in the free run.

TCP unnecessarily reduces its transmission rate at each detected packet loss, lowering end-to-end throughput.

“As a consequence, TCP unnecessarily reduces its transmission rate at each detected packet loss, lowering the end-to-end throughput.”

It is difficult to perform a good classification using simple tests, such as those implemented by TCP-Veno and TCP-Westwood, on metrics, since they lack correlation to the cause for packet loss.

“it has been shown that it is difficult to perform a good classification using simple tests, such as the ones implemented by TCP-Veno and TCP-Westwood, on these metrics, since they lack correlation to the cause for packet loss [60].”

• cites [60] Biaz S, Vaidya NH. Distinguishing congestion losses from wireless transmission losses: a negative result. In: Proceedings 7th International Conference on Computer Communications and Networks (Cat. No.98EX226). Piscataway: IEEE: 1998. p. 722–31.Not checked yet

  Not checked in the free run.

Different approaches lend themselves more naturally to different network topologies and utility functions.

“Clearly, different approaches lend themselves more naturally to different network topologies and utility functions.”

The first ML-based approaches for congestion control were proposed in the context of ATM networks.

“the first ML-based approaches for congestion control were proposed in the context of asynchronous t ransfer mode (ATM) networks [175, 264, 284, 437].”

• cites [175] Habib I, Tarraf A, Saadawi T. A neural network controller for congestion control in atm multiplexers. Comput Netw ISDN Syst. 1997; 29(3):325–34. sourceNot checked yet

  Not checked in the free run.

• cites [264] Lee SJ, Hou CL. A neural-fuzzy system for congestion control in atm networks. IEEE Transactions on Systems, Man, and Cybernetics. Part B (Cybernetics). 2000; 30(1):2–9. sourceNot checked yet

  Not checked in the free run.

• cites [284] Liu YC, Douligeris C. Static vs. adaptive feedback congestion controller for atm networks. In: Global Telecommunications Conference, 1995. GLOBECOM ’95., vol 1. IEEE: 1995. p. 291–5. sourceNot checked yet

  Not checked in the free run.

• cites [437] Tarraf AA, Habib IW, Saadawi TN. Congestion control mechanism for atm networks using neural networks. In: Communications 1995. ICC ’95 Seattle, ’Gateway to Globalization’ 1995 IEEE International Conference on, vol 1.1995. p. 206–10. sourceNot checked yet

  Not checked in the free run.

The majority of ML techniques for congestion control have been applied to TCP/IP networks.

“Majority of the techniques have been applied to TCP/IP networks.”

Liu et al. proposed the first approach using ML for inferring the cause of packet loss in hybrid wired-wireless networks.

“Liu et al. [282] proposed, to the best of our knowledge, the first approach using ML for inferring the cause of packet loss in hybrid wired-wireless networks.”

• cites [282] Liu J, Matta I, Crovella M. End-to-end inference of loss nature in a hybrid wired/wireless environment.2003.Not checked yet

  Not checked in the free run.

TCP-Vegas has been shown to outperform non-ML-based packet loss classifiers.

“TCP-Vegas has been shown to outperform non-ML-based packet loss classifiers [60]”

• cites [60] Biaz S, Vaidya NH. Distinguishing congestion losses from wireless transmission losses: a negative result. In: Proceedings 7th International Conference on Computer Communications and Networks (Cat. No.98EX226). Piscataway: IEEE: 1998. p. 722–31.Not checked yet

  Not checked in the free run.

TCP congestion control has no mechanism for identifying the cause of packet loss.

“However, TCP congestion control has no mechanism for identifying the cause of packet loss.”

The combination of Drop-Tail with TCP congestion avoidance leads to TCP synchronization that may cause inefficient link utilization and excessive packet loss due to a simultaneous decrease in TCP rate.

“the combination of Drop-Tail with the TCP congestion avoidance mechanism leads to TCP synchronization that may cause serious problems [68,72]: (i) inefficient link utilization and excessive packet loss due to a simultaneous decrease in TCP rate”

• cites [68] Bonald T, May M, Bolot JC. Analytic evaluation of red performance. In: Proceedings IEEE INFOCOM 2000. Conference on, Computer Communications. Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies (Cat. No.00CH37064), vol 3.2000. p. 1415–24. sourceNot checked yet

  Not checked in the free run.

• cites [72] Braden B, Clark D, Crowcroft J, Davie B, Deering S, Estrin D, Floyd S, Jacobson V, Minshall G, Partridge C, Peterson L, Ramakrishnan K, Shenker S, Wroclawski J, Zhang L. Recommendations on queue management and congestion avoidance in the internet. RFC 2309, Internet Engineering Task Force. 1998. https://tools.ietf.org/html/rfc2309 . sourceNot checked yet

  Not checked in the free run.

The combination of Drop-Tail with TCP congestion avoidance may cause unacceptable queuing delay due to a continuous full queue state.

“(ii) unacceptable queuing delay due to a continuous full queue state”

• cites [68] Bonald T, May M, Bolot JC. Analytic evaluation of red performance. In: Proceedings IEEE INFOCOM 2000. Conference on, Computer Communications. Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies (Cat. No.00CH37064), vol 3.2000. p. 1415–24. sourceNot checked yet

  Not checked in the free run.

• cites [72] Braden B, Clark D, Crowcroft J, Davie B, Deering S, Estrin D, Floyd S, Jacobson V, Minshall G, Partridge C, Peterson L, Ramakrishnan K, Shenker S, Wroclawski J, Zhang L. Recommendations on queue management and congestion avoidance in the internet. RFC 2309, Internet Engineering Task Force. 1998. https://tools.ietf.org/html/rfc2309 . sourceNot checked yet

  Not checked in the free run.

In case of computational constraints, DT achieves the best balance between accuracy and training time.

“Therefore, in case of computational constraints, DT achieves the best balance between accuracy and training time.”

[38,282] evaluate their packet loss classifiers on simple linear topologies, which is far from realistic network topologies.

“Furthermore, [38,282] evaluate their packet loss classifiers on simple linear topologies, which is far from realistic network topologies.”

• cites [38] Barman D, Matta I. Model-based loss inference by tcp over heterogeneous networks. In: Proceedings of WiOpt 2004 Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks. Cambridge: 2004. p. 364–73.Not checked yet

  Not checked in the free run.

• cites [282] Liu J, Matta I, Crovella M. End-to-end inference of loss nature in a hybrid wired/wireless environment.2003.Not checked yet

  Not checked in the free run.

The combination of Drop-Tail with TCP congestion avoidance may cause TCP unfairness due to a few connections that monopolize the queue space.

“(iii) TCP unfairness due to a few connections that monopolize the queue space (i.e.lock-out phenomenon).”

• cites [68] Bonald T, May M, Bolot JC. Analytic evaluation of red performance. In: Proceedings IEEE INFOCOM 2000. Conference on, Computer Communications. Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies (Cat. No.00CH37064), vol 3.2000. p. 1415–24. sourceNot checked yet

  Not checked in the free run.

• cites [72] Braden B, Clark D, Crowcroft J, Davie B, Deering S, Estrin D, Floyd S, Jacobson V, Minshall G, Partridge C, Peterson L, Ramakrishnan K, Shenker S, Wroclawski J, Zhang L. Recommendations on queue management and congestion avoidance in the internet. RFC 2309, Internet Engineering Task Force. 1998. https://tools.ietf.org/html/rfc2309 . sourceNot checked yet

  Not checked in the free run.

Fonseca and Crovella's Bayesian packet loss classifier achieves up to 90% detection probability with a false alarm of 20% on real wired network datasets from BU and PMA.

“The authors construct a Bayesian packet loss classifier that achieves up to 90% detection probability with a false alarm of 20% on real wired network datasets from the Boston University (BU) and Passive Measure Analysis (PMA) [1].”

• cites [1] Aben E. NLANR PMA data. 2010. https://labs.ripe.net/datarepository/data-sets/nlanr-pma-data . Accessed 27 Dec 2017.Not checked yet

  Not checked in the free run.

Improved TCP can maintain a fair link share with legacy protocols (TCP-friendly).

“the improved TCP can maintain a fair link share with legacy protocols (i.e.TCPfriendly).”

• cites [129] El Khayat I, Geurts P, Leduc G. Improving TCP in Wireless Networks with an Adaptive Machine-Learnt Classifier of Packet Loss Causes. Berlin, Heidelberg: Springer Berlin Heidelberg; 2005, pp. 549–60. sourceNot checked yet

  Not checked in the free run.

• cites [130] El Khayat I, Geurts P, Leduc G. Enhancement of tcp over wired/wireless networks with packet loss classifiers inferred by supervised learning. Wirel Netw. 2010; 16(2):273–90. sourceNot checked yet

  Not checked in the free run.

• cites [163] Geurts P, Khayat IE, Leduc G. A machine learning approach to improve congestion control over wireless computer networks. In: Data Mining, 2004. ICDM ’04. Fourth IEEE International Conference on: 2004. p. 383–6. sourceNot checked yet

  Not checked in the free run.

Performance is superior for the BU dataset due to poor quality of RTT measurements in the PMA dataset.

“The performance is superior for the BU dataset due to the poor quality of RTT measurements in the PMA dataset.”

• cites [150] Fonseca N, Crovella M. Bayesian packet loss detection for tcp. In: Proceedings IEEE 24th Annual Joint Conference of the IEEE Computer and Communications Societies., vol 3.2005. p. 1826–37. sourceNot checked yet

  Not checked in the free run.

All classifiers by El Khayat et al. achieve much higher classification accuracy than non-ML solutions, such as TCP-Veno and TCP-Westwood.

“All the classifiers achieve a much higher classification accuracy than non-ML solutions, such as TCP-Veno and TCP-Westwood.”

• cites [129] El Khayat I, Geurts P, Leduc G. Improving TCP in Wireless Networks with an Adaptive Machine-Learnt Classifier of Packet Loss Causes. Berlin, Heidelberg: Springer Berlin Heidelberg; 2005, pp. 549–60. sourceNot checked yet

  Not checked in the free run.

• cites [130] El Khayat I, Geurts P, Leduc G. Enhancement of tcp over wired/wireless networks with packet loss classifiers inferred by supervised learning. Wirel Netw. 2010; 16(2):273–90. sourceNot checked yet

  Not checked in the free run.

• cites [163] Geurts P, Khayat IE, Leduc G. A machine learning approach to improve congestion control over wireless computer networks. In: Data Mining, 2004. ICDM ’04. Fourth IEEE International Conference on: 2004. p. 383–6. sourceNot checked yet

  Not checked in the free run.

Barman and Matta's classifier requires support from the network to obtain the estimated probability of wireless loss.

“this ML-based packet loss classifier, unlike others, requires support from the network to obtain one of its input features, the estimated probability of wireless loss.”

• cites [38] Barman D, Matta I. Model-based loss inference by tcp over heterogeneous networks. In: Proceedings of WiOpt 2004 Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks. Cambridge: 2004. p. 364–73.Not checked yet

  Not checked in the free run.

Improved TCP with Boosting DT classifier exhibits throughput gains over standard TCP-NewReno and TCP-Veno.

“The authors continue on to improve TCP with the Boosting DT classifier, which exhibit throughput gains over the standard TCP-NewReno [185] and TCP-Veno.”

• cites [185] Henderson T, Floyd S, Gurtov S, Nishida Y. sThe newreno modification to tcp’s fast recovery algorithm. RFC 6582, Internet Engineering Task Force. 2012. https://tools.ietf.org/html/rfc6582 . sourceNot checked yet

  Not checked in the free run.

Their classifiers integrate two TCP variants that keep low control overhead for better performance over standard TCP-NewReno and TCP-SACK, and Burst-TCP for OBS networks.

“These classifiers integrate two TCP variants that keep a low control overhead for providing better performance (e.g. higher throughput and fewer timeouts) over the standard TCP-NewReno [185] and TCP-SACK [299], and Burst-TCP [490] for OBS networks.”

• cites [185] Henderson T, Floyd S, Gurtov S, Nishida Y. sThe newreno modification to tcp’s fast recovery algorithm. RFC 6582, Internet Engineering Task Force. 2012. https://tools.ietf.org/html/rfc6582 . sourceNot checked yet

  Not checked in the free run.

• cites [299] Mathis M, Mahdavi J, Floyd S, Romanow A. Tcp selective acknowledgment options. RFC 2018, Internet Engineering Task Force.1996. https://tools.ietf.org/html/rfc2018 . sourceNot checked yet

  Not checked in the free run.

• cites [490] Yu X, Qiao C, Liu Y. Tcp implementations and false time out detection in obs networks. In: IEEE INFOCOM 2004, vol 2.2004. p. 774–84. sourceNot checked yet

  Not checked in the free run.

A TCP variant enhanced with the Bayesian packet loss classifier results in a throughput improvement of up to 25% over standard TCP-Reno.

“resulting in a throughput improvement of up to 25% over the standard TCP-Reno.”

• cites [150] Fonseca N, Crovella M. Bayesian packet loss detection for tcp. In: Proceedings IEEE 24th Annual Joint Conference of the IEEE Computer and Communications Societies., vol 3.2005. p. 1826–37. sourceNot checked yet

  Not checked in the free run.

RED suffers from poor responsiveness.

“RED suffers from poor responsiveness”

• cites [269] Li F, Sun J, Zukerman M, Liu Z, Xu Q, Chan S, Chen G, Ko KT. A comparative simulation study of tcp/aqm systems for evaluating the potential of neuron-based aqm schemes. J Netw Comput Appl. 2014; 41(Supplement C):274–99. sourceNot checked yet

  Not checked in the free run.

NN-based schemes result in higher packet drop than PI schemes.

“the NNbased schemes result in a higher packet drop than PI schemes.”

• cites [269] Li F, Sun J, Zukerman M, Liu Z, Xu Q, Chan S, Chen G, Ko KT. A comparative simulation study of tcp/aqm systems for evaluating the potential of neuron-based aqm schemes. J Netw Comput Appl. 2014; 41(Supplement C):274–99. sourceNot checked yet

  Not checked in the free run.

RED fails to stabilize the queue length to a target value.

“fails to stabilize the queue length to a target value”

• cites [269] Li F, Sun J, Zukerman M, Liu Z, Xu Q, Chan S, Chen G, Ko KT. A comparative simulation study of tcp/aqm systems for evaluating the potential of neuron-based aqm schemes. J Netw Comput Appl. 2014; 41(Supplement C):274–99. sourceNot checked yet

  Not checked in the free run.

Significant research has been conducted to apply ML for building an effective and reliable AQM scheme capable of intelligently managing the queue length and tuning its parameters based on network and traffic conditions.

“significant research has been conducted to apply ML for building an effective and reliable AQM scheme, which is capable of intelligently managing the queue length and tuning its parameters based on network and traffic conditions.”

NLMS-based predictors have a high computational overhead that is unjustified compared to a simpler predictor based on a low pass filter.

“these NLMS-based predictors have a high computational overhead that is unjustified in comparison to a simpler predictor based on, for instance, a low pass filter.”

For a single bottleneck link, NN-based schemes outperform RED/PI schemes in terms of queue length accuracy and queue length jitter under different network settings.

“For a single bottleneck link, the results demonstrate that the NN-based schemes outperform the RED/PI schemes in terms of queue length accuracy (QL Acc ) and queue length jitter (QL Jit ), under different network settings.”

• cites [269] Li F, Sun J, Zukerman M, Liu Z, Xu Q, Chan S, Chen G, Ko KT. A comparative simulation study of tcp/aqm systems for evaluating the potential of neuron-based aqm schemes. J Netw Comput Appl. 2014; 41(Supplement C):274–99. sourceNot checked yet

  Not checked in the free run.

AN-AQM modification improves the performance of the PIDNN in more realistic scenarios.

“This modification improves the performance of the PIDNN in more realistic scenarios.”

• cites [427] Sun J, Zukerman M. An adaptive neuron aqm for a stable internet. In: Proceedings of the 6th International IFIP-TC6 Conference on Ad Hoc and Sensor Networks, Wireless Networks, Next Generation Internet, Springer-Verlag, Berlin, Heidelberg, NETWORKING’07.2007. p. 844–54. sourceNot checked yet

  Not checked in the free run.

α_SNFAQM produces more packet drops in order to notify congestion earlier.

“α_SNFAQM produce more packet drops in order to notify the congestion earlier.”

• cites [498] Zhani MF, Elbiaze H, Kamoun F. α_snfaqm: an active queue management mechanism using neurofuzzy prediction. In: 2007 12th IEEE Symposium on Computers and Communications.2007. p. 381–6. sourceNot checked yet

  Not checked in the free run.

FAPIDNN's fuzzy controller adds computational complexity.

“the fuzzy controller of FAPIDNN also adds computational complexity.”

AN-AQM incurs higher computational overhead due to an additional PID transfer function and the increase in the number of input neurons.

“it incurs a higher computational overhead, due to an additional PID transfer function and the increase in the number of input neurons.”

PAQM is the first approach using ML for improving AQM.

“PAQM [160], to the best of our knowledge, is the first approach using ML for improving AQM.”

• cites [160] Gao Y, He G, Hou JC. On exploiting traffic predictability in active queue management. In: Proceedings. Twenty-First Annual Joint Conference of the IEEE Computer and Communications Societies, vol. 3. Piscataway: IEEE: 2002. p. 1630–9.Not checked yet

  Not checked in the free run.

TCP-FALA is the first RL-based TCP variant that focuses on CWND adaptation in wireless networks, particularly in WANETs.

“To the best of our knowledge, TCP-FALA [380] is the first RL-based TCP variant that focuses on CWND adaptation in wireless networks, particularly in WANETs.”

• cites [380] Ramana BV, Manoj BS, Murthy CSR. Learning-tcp: a novel learning automata based reliable transport protocol for ad hoc wireless networks. In: 2nd International Conference on Broadband Networks 2005.2005. p. 484–493. Vol. 1. sourceNot checked yet

  Not checked in the free run.

The limited size of TCP-FALA's action-set makes it difficult to map the range of network responses to appropriate actions.

“However, the limited size of the action-set makes difficult mapping the range of responses provided by the network to the appropriate actions.”

AN-AQM is superior in these performance metrics for realistic scenarios involving UDP traffic noise.

“AN-AQM is superior in these performance metrics for realistic scenarios involving UDP traffic noise.”

• cites [269] Li F, Sun J, Zukerman M, Liu Z, Xu Q, Chan S, Chen G, Ko KT. A comparative simulation study of tcp/aqm systems for evaluating the potential of neuron-based aqm schemes. J Netw Comput Appl. 2014; 41(Supplement C):274–99. sourceNot checked yet

  Not checked in the free run.

Methods for improving TCP-Vegas (e.g., Vegas-W) are still insufficient to account for such variability, as their operation relies on past network conditions rather than present or future.

“methods for improving TCP-Vegas (e.g.Vegas-W [119]) are still insufficient to account for such variability, as their operation relies on the past network conditions rather than present or future.”

• cites [119] Ding L, Wang X, Xu Y, Zhang W. Improve throughput of tcp-vegas in multihop ad hoc networks. Comput Commun. 2008; 31(10):2581–8. sourceNot checked yet

  Not checked in the free run.

The CWND update mechanism of TCP is not suitable for the diverse characteristics of different network technologies.

“the CWND update mechanism of TCP is not suitable for the diverse characteristics of different network technologies [30,122].”

• cites [30] Badarla V, Siva Ram Murthy C. A novel learning based solution for efficient data transport in heterogeneous wireless networks. Wirel Netw. 2010; 16(6):1777–98. sourceNot checked yet

  Not checked in the free run.

• cites [122] Dong M, Li Q, Zarchy D, Godfrey PB, Schapira M. Pcc: Re-architecting congestion control for consistent high performance. In: Proceedings of the 12th, USENIX Conference on Networked Systems Design and Implementation, USENIX Association, Berkeley, CA, USA, NSDI’15. Berkley: USENIX Association: 2015. p. 395–408.Not checked yet

  Not checked in the free run.

The evaluation of these RL-based approaches relies on synthetic data generated from multiple network topologies simulated in tools such as GloMoSim, ns-2, and ns-3.

“The evaluation of these RL-based approaches rely on synthetic data generated from multiple network topologies simulated in tools, such as GloMoSim, ns-2, and ns-3.”

The deterministic nature of TCP is more prone to cause higher contention losses and CWND synchronization problems in WANET, due to node mobility that continuously modifies the wireless multi-hop paths.

“the deterministic nature of TCP is more prone to cause higher contention losses and CWND synchronization problems in WANET, due to node mobility that continuously modifies the wireless multi-hop paths [29,379].”

• cites [29] Badarla V, Murthy CSR. Learning-tcp: A stochastic approach for efficient update in tcp congestion window in ad hoc wireless networks. J Parallel Distrib Comput. 2011; 71(6):863–78. sourceNot checked yet

  Not checked in the free run.

• cites [379] Ramana BV, Murthy CSR. Learning-tcp: A novel learning automata based congestion window updating mechanism for ad hoc wireless networks. In: Proceedings of the 12th International Conference on High Performance Computing, Springer-Verlag, Berlin, Heidelberg, HiPC’05.2005. p. 454–464. sourceNot checked yet

  Not checked in the free run.

A couple of the RL-based approaches also include experimental evaluation.

“A couple of these approaches [29,122] also include experimental evaluation.”

• cites [29] Badarla V, Murthy CSR. Learning-tcp: A stochastic approach for efficient update in tcp congestion window in ad hoc wireless networks. J Parallel Distrib Comput. 2011; 71(6):863–78. sourceNot checked yet

  Not checked in the free run.

• cites [122] Dong M, Li Q, Zarchy D, Godfrey PB, Schapira M. Pcc: Re-architecting congestion control for consistent high performance. In: Proceedings of the 12th, USENIX Conference on Networked Systems Design and Implementation, USENIX Association, Berkeley, CA, USA, NSDI’15. Berkley: USENIX Association: 2015. p. 395–408.Not checked yet

  Not checked in the free run.

Several simulations in ns-2 and GloMoSim show that both Learning-TCP and TCP-FALA outperform standard TCP-NewReno in terms of packet loss, goodput, and fairness.

“Several simulations in ns-2 and GloMoSim show that both Learning-TCP and TCP-FALA outperform standard TCP-NewReno in terms of packet loss, goodput, and fairness.”