Adaptive Capacity Provisioning for Carbon-Aware Data Centers: A Digital Twin-Based Approach
The DOI matches its Crossref record; 45 of the 45 references it cites were found in a registry or an index.
Abstract
This paper considers the carbon-aware data center (DC) capacity provisioning problem under uncertain green energy availability and computing demand. To address it, accurate carbon emissions estimation and robust capacity provisioning are necessary. Existing studies mainly consider the carbon footprint of the computing system and merely consider that of the physical facilities, which also contribute significant carbon emissions. Furthermore, their capacity provisioning is neither uncertainty-aware nor adaptive to the dynamic computing demand. To bridge these gaps, we propose an adaptive capacity provisioning framework based on the physics-informed digital twin. We design the digital twin to holistically capture a DC's operational carbon footprint, including both the computing system and the physical facilities. The digital twin is differentiable and established with a collection of physics-informed learnable models that are learned with online operational data. We further address the challenge of capacity provisioning under uncertainties by designing a shrinking horizon model predictive control. The designed capacity planner updates its estimation of future computing demand based on the observable computing system states. At each capacity provisioning round, we solve the capacity provisioning problem using a gradient-based optimization technique with the gradient provided by the digital twin. We extensively evaluate our approach usingrealoperational data from a large-scale production data center. First, our digital twin accurately predicts holistic DC energy usage with a relative absolute error of less than 5%, which is accurate according to the industrial rule of thumb. Second, we show that our solution is comparable to the oracle solution with perfect knowledge about all uncertainties, outperforming the state-of-the-art Predict-then-Plan approach significantly in terms of SLO violation reduction. Furthermore, our approach reduces carbon footprint by 27% compared with the over-provisioning scheme currently adopted by the industry.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · cited 1×
“Then, the output of the Poisson arrival model is combined with the temporal features to the resource model to predict the resource request of each workload with a long short-term memory (LSTM) model [38].”
PyTorch: An Imperative Style, High-Performance Deep Learning Library
Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. (2019) · arXiv · doi:10.48550/arxiv.1912.01703 · Cited by 31,473 · cited 2×
“Furthermore, we implement our digital twin using Pytorch [33] to make it differentiable with respect to its inputs and internal parameters.”
EnergyPlus: creating a new-generation building energy simulation program
Crawley, D. B., Lawrie, L. K., Winkelmann, F. C., Buhl, W. F., Huang, Y. J., Pedersen, C. O., Strand, R. K., Liesen, R. J., Fisher, D. E., Witte, M. J., et al. (2001) · Energy and Buildings · doi:10.1016/s0378-7788(00)00114-6 · Cited by 2,822 · cited 4×
“Some latest works adopt sophisticated physics-based energy simulators (e.g., EnergyPlus [7]) to evaluate DC energy footprint.”
InfluentialDeepAR: Probabilistic forecasting with autoregressive recurrent networks
Salinas, D., Flunkert, V., Gasthaus, J., & Januschowski, T. (2020) · International Journal of Forecasting · doi:10.1016/j.ijforecast.2019.07.001 · Cited by 2,661
Stochastic Network Optimization with Application to Communication and Queueing Systems
Neely, M. J. (2010) · Synthesis lectures on communication networks · doi:10.2200/s00271ed1v01y201006cnt007 · Cited by 2,218
Making scheduling cool: temperature-aware workload placement in data centers
Moore, J., Chase, J., Ranganathan, P., & Sharma, R. K. (2005) · USENIX Annual Technical Conference · Cited by 651
Cortez, E., Bonde, A., Muzio, A., Russinovich, M., Fontoura, M., & Bianchini, R. (2017) · Proceedings of the 26th Symposium on Operating Systems Principles · doi:10.1145/3132747.3132772 · Cited by 633 · cited 1×
“We use publicly available cloud workload trace from Azure [42] in our evaluation.”
Kouvaritakis, B., & Cannon, M. (2016) · Advanced Textbooks in Control and Signal Processing · doi:10.1007/978-3-319-24853-0 · Cited by 515
Renewable and cooling aware workload management for sustainable data centers
Liu, Z., Chen, Y., Bash, C., Wierman, A., Gmach, D., Wang, Z., Marwah, M., & Hyser, C. (2012) · Proceedings of the 12th ACM SIGMETRICS/PERFORMANCE joint international conference on Measurement and Modeling of Computer Systems · doi:10.1145/2254756.2254779 · Cited by 413
Self-citationCarbon-aware computing for datacenters
Radovanović, A., Koningstein, R., Schneider, I., Chen, B., Duarte, A., Roy, B., Xiao, D., Haridasan, M., Hung, P., Care, N., et al. (2023) · IEEE Transactions on Power Systems · doi:10.1109/tpwrs.2022.3173250 · Cited by 353 · cited 9×
“In this paper, we focus on the power modeling of the aggregated computing demand similar to the previous work [5], [12].”
InfluentialTirmazi, M., Barker, A., Deng, N., Haque, M. E., Qin, Z. G., Hand, S., Harchol-Balter, M., & Wilkes, J. (2020) · Proceedings of the Fifteenth European Conference on Computer Systems · doi:10.1145/3342195.3387517 · Cited by 325 · cited 1×
“Furthermore, as many cloud workloads can tolerate long waiting (even up to 24 hours [48]), cloud data centers can significantly reduce their carbon footprint by adaptive capacity provisioning without significantly compromising service quality.”
GreenSlot: scheduling energy consumption in green datacenters
Goiri, Í., Le, K., Haque, M. E., Beauchea, R., Nguyen, T. D., Guitart, J., Torres, J., & Bianchini, R. (2011) · Proceedings of 2011 International Conference for High Performance Computing, Networking, Storage and Analysis · doi:10.1145/2063384.2063411 · Cited by 309 · cited 4×
“GreenSlot [11] is a parallel batch job scheduler that uses the predicted renewable energy availability to schedule workload with a greedy heuristic to minimize the DC’s electricity cost.”
InfluentialLet's wait awhile: how temporal workload shifting can reduce carbon emissions in the cloud
Wiesner, P., Behnke, I., Scheinert, D., Gontarska, K., & Thamsen, L. (2021) · International Middleware Conference · doi:10.1145/3464298.3493399 · Cited by 232 · cited 1×
“In some recent empirical studies [15], [31], the DC capacity provisioning problem is addressed by simplifying or ignoring the uncertainty consideration.”
Toward a Systematic Survey for Carbon Neutral Data Centers
Cao, Z., Zhou, X., Hu, H., Wang, Z., & Wen, Y. (2022) · IEEE Communications Surveys & Tutorials · doi:10.1109/comst.2022.3161275 · Cited by 179
Self-citationYadav, R., Zhang, W., Kaiwartya, O., Singh, P. R., Elgendy, I. A., & Tian, Y.-C. (2018) · IEEE Access · doi:10.1109/access.2018.2872750 · Cited by 177 · cited 1×
“For example, cloud operators can judiciously manage the VM selection, migration, and consolidation to reduce energy footprint [22], [23], [24], [25], [26].”
Carbon Explorer: A Holistic Framework for Designing Carbon Aware Datacenters
Acun, B., Lee, B., Kazhamiaka, F., Maeng, K., Gupta, U., Chakkaravarthy, M., Brooks, D., & Wu, C.-J. (2023) · Proceedings of the 28th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2 · doi:10.1145/3575693.3575754 · Cited by 171 · cited 2×
“6 C according to the statistics in [15], and the capacity switch threshold σ as 0 .”
Monte Carlo Motion Planning for Robot Trajectory Optimization Under Uncertainty
Janson, L., Schmerling, E., & Pavone, M. (2017) · Springer proceedings in advanced robotics · doi:10.1007/978-3-319-60916-4_20 · Cited by 132
An adaptive heuristic for managing energy consumption and overloaded hosts in a cloud data center
Yadav, R., Zhang, W., Li, K., Liu, C., Shafiq, M., & Karn, N. K. (2018) · Wireless Networks · doi:10.1007/s11276-018-1874-1 · Cited by 97
Online Convex Optimization With Time-Varying Constraints and Bandit Feedback
Cao, X., & Liu, K. J. R. (2018) · IEEE Transactions on Automatic Control · doi:10.1109/tac.2018.2884653 · Cited by 97 · cited 1×
“(cid:2) OnlineOpt [45]: This is a primal-dual optimization approach to solving the original problem with adjustable Fig.”
Managing overloaded hosts for energy-efficiency in cloud data centers
Yadav, R., Zhang, W., Li, K., Liu, C., & Laghari, A. A. (2021) · Cluster Computing · doi:10.1007/s10586-020-03182-3 · Cited by 84 · cited 1×
“For example, cloud operators can judiciously manage the VM selection, migration, and consolidation to reduce energy footprint [22], [23], [24], [25], [26].”
Carbon-Aware Electricity Cost Minimization for Sustainable Data Centers
Dou, H., Qi, Y., Wei, W., & Song, H. H. (2017) · IEEE Transactions on Sustainable Computing · doi:10.1109/tsusc.2017.2711925 · Cited by 78 · cited 3×
“In [8], [9], the authors adopt the Lyapunov optimization framework [10] to dynamically adjust the DC capacity provisioning with the time-varying renewable energy availability or electricity price.”
MeReg: Managing Energy-SLA Tradeoff for Green Mobile Cloud Computing
Yadav, R., & Zhang, W. (2017) · Wireless Communications and Mobile Computing · doi:10.1155/2017/6741972 · Cited by 71 · cited 1×
“For example, cloud operators can judiciously manage the VM selection, migration, and consolidation to reduce energy footprint [22], [23], [24], [25], [26].”
MuMs: Energy-Aware VM Selection Scheme for Cloud Data Center
Yadav, R., Zhang, W., Chen, H., & Guo, T. (2017) · International Workshop on Database and Expert Systems Applications (DEXA) · doi:10.1109/dexa.2017.43 · Cited by 69 · cited 1×
“For example, cloud operators can judiciously manage the VM selection, migration, and consolidation to reduce energy footprint [22], [23], [24], [25], [26].”
3Sigma: distribution-based cluster scheduling for runtime uncertainty
Park, J. W., Tumanov, A., Jiang, A. H., Kozuch, M. A., & Ganger, G. R. (2018) · European Conference on Computer Systems · doi:10.1145/3190508.3190515 · Cited by 68 · cited 2×
“In this section, we evaluate our approach with different dead-line slack ratios to explore its impacts on SLO violation similar to the analysis in [43].”
Exploiting Renewable Sources: When Green SLA Becomes a Possible Reality in Cloud Computing
Hasan, M. S., Kouki, Y., Ledoux, T., & Pazat, J.-L. (2015) · IEEE Transactions on Cloud Computing · doi:10.1109/tcc.2015.2459710 · Cited by 67
Deng, W., Liu, F., Jin, H., Wu, C., & Liu, X. (2013) · ACM International Conference on Future Energy Systems (e-Energy) · doi:10.1145/2487166.2487183 · Cited by 67
Adapting Datacenter Capacity for Greener Datacenters and Grid
Lin, L., & Chien, A. A. (2023) · ACM International Conference on Future Energy Systems (e-Energy) · doi:10.1145/3575813.3595197 · Cited by 58 · cited 1×
“Due to the high variability of grid carbon intensity, renewable energy availability, and electricity price, the flexibility of DC workloads has been explored to decarbonize both the DC itself [5], [8], [12], [27] and the power grid [28] or reduce the monetary cost [29].”
Goiri, Í., Katsak, W., Le, K., Nguyen, T. D., & Bianchini, R. (2013) · ACM SIGPLAN Notices · doi:10.1145/2499368.2451123 · Cited by 55
Jaureguialzo, E. (2011) · International Telecommunications Energy Conference (INTELEC) · doi:10.1109/intlec.2011.6099718 · Cited by 49 · cited 1×
“To address the first challenge, a common industrial practice is using a fixed Power Usage Effectiveness (PUE) [6], which is the ratio between total DC energy consumption and the total IT energy consumption, to estimate the DC energy consumption from the IT energy consumption.”
Optimization of Data and Energy Migrations in Mini Data Centers for Carbon-Neutral Computing
da Silva, M. D. M., Gamatié, A., Sassatelli, G., Poss, M., & Robert, M. (2022) · IEEE Transactions on Sustainable Computing · doi:10.1109/tsusc.2022.3197090 · Cited by 48 · cited 1×
“Due to the high variability of grid carbon intensity, renewable energy availability, and electricity price, the flexibility of DC workloads has been explored to decarbonize both the DC itself [5], [8], [12], [27] and the power grid [28] or reduce the monetary cost [29].”
Minimizing Energy Costs for Geographically Distributed Heterogeneous Data Centers
Hogade, N., Pasricha, S., Siegel, H. J., Maciejewski, A. A., Oxley, M. A., & Jonardi, E. (2018) · IEEE Transactions on Sustainable Computing · doi:10.1109/tsusc.2018.2822674 · Cited by 47 · cited 1×
“Due to the high variability of grid carbon intensity, renewable energy availability, and electricity price, the flexibility of DC workloads has been explored to decarbonize both the DC itself [5], [8], [12], [27] and the power grid [28] or reduce the monetary cost [29].”
Data Center Sustainability: Revisits and Outlooks
Cao, Z., Zhou, X., Wu, X., Zhu, Z., Liu, T., Neng, J., & Wen, Y. (2023) · IEEE Transactions on Sustainable Computing · doi:10.1109/tsusc.2023.3281583 · Cited by 46 · cited 1×
“Furthermore, as the energy efficiency improvement is stagnant, DC decarboniza-tion becomes more challenging [3].”
Self-citationOptimizing Data Center Energy Efficiency via Event-Driven Deep Reinforcement Learning
Ran, Y., Zhou, X., Hu, H., & Wen, Y. (2022) · IEEE Transactions on Services Computing · doi:10.1109/tsc.2022.3157145 · Cited by 43 · cited 1×
“This topic has recently been explored under the framework of deep reinforcement learning (DRL) [50], which jointly performs job allocation and data hall CRAH unit control to reduce total DC energy consumption.”
Self-citationGenerating Complex, Realistic Cloud Workloads using Recurrent Neural Networks
Bergsma, S., Zeyl, T. J., Senderovich, A., & Beck, J. C. (2021) · ACM SIGOPS Symposium on Operating Systems Principles (SOSP) · doi:10.1145/3477132.3483590 · Cited by 36
Forsthoffer's Best Practice Handbook for Rotating Machinery
Forsthoffer, W. E. (2011) · Elsevier eBooks · doi:10.1016/c2009-0-64191-x · Cited by 36
Mahmud, A. H., & Ren, S. (2013) · ACM SIGMETRICS Performance Evaluation Review · doi:10.1145/2518025.2518029 · Cited by 32 · cited 4×
“Due to the high variability of grid carbon intensity, renewable energy availability, and electricity price, the flexibility of DC workloads has been explored to decarbonize both the DC itself [5], [8], [12], [27] and the power grid [28] or reduce the monetary cost [29].”
InfluentialManaging Green Datacenters Powered by Hybrid Renewable Energy Systems
Li, C., Wang, R., Li, T., Qian, D., & Yuan, J. (2014) · International Conference on Autonomic Computing · Cited by 31
Self-citationData Driven Chiller Plant Energy Optimization with Domain Knowledge
Vu, H. D., Chai, K. S., Keating, B. D., Tursynbek, N., Xu, B., Yang, K., Yang, X., & Zhang, Z. (2017) · ACM International Conference on Information and Knowledge Management (CIKM) · doi:10.1145/3132847.3132860 · Cited by 27 · cited 1×
“Some other studies, e.g., [21], adopt data-driven approaches, but their models’ extrapolation capability is unsatisfactory.”
Phyllis: Physics-Informed Lifelong Reinforcement Learning for Data Center Cooling Control
Wang, R., Cao, Z., Zhou, X., Wen, Y., & Tan, R. (2023) · ACM International Conference on Future Energy Systems (e-Energy) · doi:10.1145/3575813.3595189 · Cited by 23 · cited 2×
“To synergistically analyze a DC’s energy foot-print, some proposals adopt the physics-based energy simulator EnergyPlus [7] to model the holistic DC infrastructure [18], [19], [20].”
Self-citationCarbon Responder: Coordinating Demand Response for the Datacenter FleetPreprint
Xing, J., Acun, B., Sundarrajan, A., Brooks, D. M., Chakkaravarthy, M., Avila, N., Wu, C.-J., & Lee, B. C. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2311.08589 · Cited by 23 · cited 4×
“(cid:2) Predict-then-Plan [5], [12], [40]: This approach uses next-day renewable energy and IT workload forecasting to generate the next-day capacity provisioning plan and then follow them strictly.”
InfluentialGreen Data Center Cooling Control via Physics-guided Safe Reinforcement Learning
Wang, R., Cao, Z., Zhou, X., Wen, Y., & Tan, R. (2023) · ACM Transactions on Cyber-Physical Systems · doi:10.1145/3582577 · Cited by 18 · cited 2×
“Some works also consider the impact of inlet temperature on server energy consumption [18], [19].”
Self-citationGreenDRL: managing green datacenters using deep reinforcement learning
Zhang, K., Wang, P., Gu, N., & Nguyen, T. D. (2022) · Symposium on Cloud Computing · doi:10.1145/3542929.3563501 · Cited by 14 · cited 1×
“GreenDRL [14] is a DC manager based on Deep Reinforcement Learning (DRL).”
Cao, Z., Wang, R., Zhou, X., & Wen, Y. (2023) · ACM International Conference on Future Energy Systems (e-Energy) · doi:10.1145/3575813.3597343 · Cited by 12 · cited 2×
“Some works also consider the impact of inlet temperature on server energy consumption [18], [19].”
Self-citationScheduling Challenges for Variable Capacity Resources
Zhang, C., & Chien, A. A. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-88224-2_10 · Cited by 9 · cited 1×
“Such observation also coincides with the findings in [49], showing the necessity of workload duration-aware scheduler design when the DC capacity is time-varying.”
NCVX: A General-Purpose Optimization Solver for Constrained Machine and Deep LearningPreprint
Liang, B., Mitchell, T., & Sun, J. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2210.00973 · Cited by 5 · cited 1×
“…at the beginning of the τ -th capacity provisioning round, we observe the system and obtain the updated cumulative supply deficit s [ τ ] and then solve the following optimization: As the DC energy model is differentiable , we solve it ef-ficiently via the gradient-based optimization technique [41].”
Research works printed but not linked: 4
- Shafto, 2012, Modeling, simulation, information technology & processing roadmap, Nat. Aeronaut. Space Admin., 32, 1
- Bergman, 2011, Fundamentals of Heat and Mass Transfer
- 2022, Energy transformation (chapter 2)
- Guideline, 2014, Measurement of energy, demand, and water savings, ASHRAE Guideline, 4, 1
Cited by
Kim, J., Kim, J., Sovacool, B. K., Griffiths, S., Bazilian, M., Creutzig, F., Rio, D. D. F. D., Agarwala, M., Choi, M., & Debnath, R. (2026) · Renewable and Sustainable Energy Reviews · doi:10.1016/j.rser.2026.116861 · Cited by 29
Alex, M., Ojo, S. O., & Awuor, F. M. (2025) · Computers · doi:10.3390/computers14070280 · Cited by 22
Bytes for beaks? The green transformation of data center and bird biodiversity in China
Liang, Y., Han, X., & Kai, Z. (2025) · Finance research letters · doi:10.1016/j.frl.2025.108564 · Cited by 9
Wang, J., Cao, S., Li, R., & Xu, W. (2025) · Energies · doi:10.3390/en18215741 · Cited by 7
Chen, J., Al-Nussairi, A. K. J., Chyad, M. H., Azarinfar, H., Khosravi, M., Jin, K., & Zhang, J. (2025) · Energy Reports · doi:10.1016/j.egyr.2025.03.052 · Cited by 7
Metadata
- Authors
- Cao, Z., Wang, R., Zhou, X., Tan, R., Wen, Y., Yan, Y., & Wang, Z.
- Affiliations
- Nanyang Technological University (SG)
- Jiangxi Science and Technology Normal University (CN)
- Alibaba Group (China) (CN)
- Venue
- IEEE Transactions on Sustainable Computing
- Published
- 2025-07
- Volume
- 10
- Issue
- 4
- Pages
- 730-743
- ISSN
- 2377-3782 (ISSN-L)
- 2377-3790
- OpenAlex
- W4406321541
- Cited by
- 16
- References
- 45
- Citations by year
- 2026: 9 · 2025: 7
- FWCI
- 16.77
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- IEEE Transactions on Sustainable Computing (publishedVersion)
- DR-NTU (Nanyang Technological University) (submittedVersion)
- Funders
- RIE2025 Industry Alignment Fund
- Industry Collaboration Projects (I2301E0026)
- Alibaba Group
- Nanyang Technological University
- National Natural Science Foundation of China (62262026)
- Natural Science Foundation of Jiangxi Province (20232BAB202020)
- Topics
- Physical Sciences › Computer Science › Information Systems › Cloud Computing and Resource Management
- Physical Sciences › Computer Science › Hardware and Architecture › Energy Efficiency in Computing
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Green IT and Sustainability
- Keywords
- capacity provisioning, digital twin, physics-informed neural networks, carbon footprint calculation, model predictive control, energy usage prediction
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.