pith. sign in

arxiv: 2604.09615 · v1 · submitted 2026-03-14 · 💻 cs.DC

Calibrating Microgrid Simulations for Energy-Aware Computing Systems

Marvin Steinke This is my paper

Pith reviewed 2026-05-15 11:56 UTC · model grok-4.3

classification 💻 cs.DC
keywords microgrid simulationenergy-aware computingpower approximationKepler frameworkVessimGPU workloadscarbon-aware computingself-calibration
0
0 comments X

The pith

A self-calibrating testbed integrates real computing nodes with Vessim microgrid simulations to achieve accurate power approximations using Kepler estimates refined by socket meter measurements.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes a software testbed that combines renewable energy simulators with actual GPU and CPU hardware to evaluate energy-aware strategies without full physical infrastructure. It approximates per-process power draw through the Kepler framework and then refines those values inside Vessim by treating external socket meter readings as ground truth. Evaluation across intensive workloads shows the uncalibrated approximations already track total node power closely, with further calibration delivering large gains on GPU tasks. This setup aims to make realistic testing of carbon-aware scheduling accessible by keeping simulations economical while incorporating live node fluctuations.

Core claim

The Kepler framework supplies a fairly accurate whole-node power approximation with an average regression coefficient of 1.01 and R-squared values of 0.95 across tested workloads, although machine learning tasks exhibit higher deviation; an average static y-intercept of 5.23 W reveals idle-power gaps. Dynamic per-process calibration against the external socket meter then raises accuracy by roughly 50 percent for GPU workloads and 3.5 percent for CPU workloads.

What carries the argument

Self-calibration loop that feeds Kepler per-process power estimates into Vessim microgrid simulation and corrects them in real time against external socket power meter readings treated as definitive system-level truth.

If this is right

  • Developers can test carbon-aware scheduling policies on simulated renewables while the hardware side reflects live node power draw.
  • GPU-heavy AI workloads become the primary targets for calibration gains, suggesting focused refinement on those workloads first.
  • Static idle-power offsets of roughly 5 W indicate a clear target for improving Kepler's baseline estimates.
  • The overall approach lowers the barrier to experimenting with renewable-aligned computing without maintaining costly physical microgrids.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same calibration pattern could be applied to other simulators or frameworks that currently lack hardware-in-the-loop grounding.
  • If idle-power inaccuracies persist, future work might combine the socket meter with additional sensors at the CPU and GPU level to isolate the source.
  • Real-time versions of this calibration might allow production systems to adapt power models on the fly as workload mixes change.

Load-bearing premise

The external socket power meter supplies an unbiased ground-truth measurement that transfers directly to calibrate internal per-process approximations without introducing new systematic errors from real-time node fluctuations.

What would settle it

Re-run the same GPU and CPU workloads while logging simulated versus measured power traces at one-second resolution; the central claim fails if the post-calibration regression coefficient deviates from 1.0 by more than 0.05 or if R-squared drops below 0.90 on previously high-accuracy workloads.

Figures

Figures reproduced from arXiv: 2604.09615 by Marvin Steinke.

Figure 1
Figure 1. Figure 1: Sustainable Cloud-Edge Infrastructure Vision [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Ecovisor Design⁶ Research and development of carbon-aware applications and sys￾tems, however, continue to face challenges due to the limited availabil￾ity of testing environments. Souza et al. [13] implemented an ecovi￾sor on a hardware testbed, including a few low-powered microservers, but even this small-scale physical prototype proved to be both very complex and costly, primarily due to the expensive co… view at source ↗
Figure 3
Figure 3. Figure 3: Vessim Simulation Scenario Architecture [22] An actor 𝑎 ∈ 𝐴 within a microgrid can be a power producer (𝑝 𝑎 𝑡 > 0) or consumer (𝑝 𝑎 𝑡 < 0), communicating their power production or consumption at simulation step 𝑡 ∈ 𝑁 to the grid simulator and op￾tionally sending additional state information 𝑠𝑡𝑎𝑡𝑒𝑎 𝑡 to controllers. The default grid simulator in Vessim aggregates net power from all actors to determine the g… view at source ↗
Figure 4
Figure 4. Figure 4: Power domains supported by RAPL⁷ NVIDIA GPUs have transitioned from using estimation-based to measurement-based methodologies for tracking power consumption. Older models, such as from the Fermi architecture from 2010 estimate power usage by monitoring activity signals. Newer GPUs from the Ke￾pler architecture from 2012 onward, in contrast, utilize shunt resistors 14 [PITH_FULL_IMAGE:figures/full_fig_p021… view at source ↗
Figure 5
Figure 5. Figure 5: Kepler Bare-Metal Deployment Architecture [PITH_FULL_IMAGE:figures/full_fig_p023_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: SmartWatts System Architecture [33] mations at the container level using Intel’s RAPL and Linux perf with hardware performance counters. It employs automatic event selection to navigate counter limitations and uses Ridge regression for adaptive power modeling with real-time workloads. Static power is isolated by capturing HWPC data during idle states, refining it through median values. The implementation i… view at source ↗
Figure 7
Figure 7. Figure 7: Testbed Architecture 4.1.1 Applications Carbon-aware applications are deployed as K8s pods which can inter￾act with the simulation environment’s HTTP API server. They are able to monitor their power consumption, power system related metrics or other information provided through the configurable API such as cur￾rent grid carbon intensity. Additionally, they can observe and manage their resource utilization,… view at source ↗
Figure 8
Figure 8. Figure 8: Application ↔ Vessim API 4.1.4 Resource Control Computing resources can be managed with K8s’ in-place vertical pod resizing of resource requests and limits for individual pods and their containers. For RAM and CPU resource caps, Linux cgroups are uti￾lized, while for controlling GPU resources, currently only NVIDIA of￾fers a sufficiently featured API for their GPUs. Depending on the spe￾cific use case and … view at source ↗
Figure 9
Figure 9. Figure 9: Testbed K8s Implementation 30 [PITH_FULL_IMAGE:figures/full_fig_p037_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Benchmark Architecture derived from Vessim’s Controller abstract class, which oversees the lifecycle of benchmarks and synchronizes data with Vessim’s Monitor. Each specific benchmark extends this custom controller class, imple￾menting the updateValues() method to adjust the configurations for subsequent runs. While resource control via the K8s API is technically feasible, it is inadvisable in the context… view at source ↗
Figure 11
Figure 11. Figure 11: Ray Benchmark Architecture In the context of evaluating batch workloads using Ray, the general benchmark design from [PITH_FULL_IMAGE:figures/full_fig_p051_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Approximations vs Measurements with Regression Line [PITH_FULL_IMAGE:figures/full_fig_p053_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Y-Intercept of Regression Line 47 [PITH_FULL_IMAGE:figures/full_fig_p054_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Deviation from Regression Line The results indicate a strong relationship between Kepler’s approx￾imations and the actual measured power consumption, as evidenced by the average regression coefficient of approximately 1.01 across all test cases. This suggests that the approximations serve as reliable pre￾dictors of power consumption. The y-intercept of the regression lines is employed to examine the devia… view at source ↗
Figure 15
Figure 15. Figure 15: System Processes’ Power This phenomenon can potentially be attributed to the special na￾ture of GPUs and their complex interactions with system components, as opposed to the more tightly integrated CPUs and memory mod￾ules. The GPU-specific kernel activities, such as scheduling, memory management, and data transfer, might elevate system process activ￾ity levels [90]. Additionally, the mandatory incorporat… view at source ↗
Figure 16
Figure 16. Figure 16: GPU Workload Calibration Offset further overhead, impacting power consumption dynamics. For the current state of art, the attributed power consumption of these sys￾tem processes is proportionally distributed among all active workload processes, weighted by their individual power consumption metrics, for calibration purposes. This procedure is depicted by the green plot p_delta, though the weighting effect… view at source ↗
Figure 17
Figure 17. Figure 17: Per-Process Power Time Series where 60% of requests targeted the search service, 39% the recommend service, and a mere 0.5% each to the user and reservation services, the reservation service exhibited the highest power consumption among all. Additionally, certain services, such as MongoDB and memcached, which exhibit minimal power consumption, were easily identifiable. In the representation, any service c… view at source ↗
Figure 18
Figure 18. Figure 18: Per-Process Total Energy Consumption 6.3.3 Restrictions Several constraints which interfere with the accuracy of validation and approximation procedures have been identified. Notably, current GPU drivers and integrations in K8s and computing systems in general, are suboptimal with frequent issues reported by users, with a heavy dependency on the specific GPU manufacturer. Although anticipated advancements… view at source ↗
Figure 19
Figure 19. Figure 19: Microservices Dependency Graph for the Hotel Reservation [PITH_FULL_IMAGE:figures/full_fig_p066_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Microservices Dependency Graph for the Media Service [PITH_FULL_IMAGE:figures/full_fig_p067_20.png] view at source ↗
read the original abstract

The surge for computing resource demand is increasing global electricity consumption in data centers which is expected to exceed 1000 TWh by 2026, mainly attributable to adoption of new AI technologies. Carbon-aware computing strategies can mitigate their environmental impact by aligning power consumption with the production of low-carbon renewable energy, but they face challenges due to the scarcity of development environments. Existing solutions either rely on costly and complex physical system architectures that are difficult to integrate and maintain or on full simulations that, while more economical, often lack realism by ignoring system overheads, and real-time node power consumption and resource fluctuations. This thesis remediates these issues by proposing a self-calibrating energy-aware software testbed that uses the Software-in-the-Loop co-simulation framework Vessim to integrate renewable energy production simulators, while including real computing nodes. The application-level power consumption of these are first approximated by the Kepler framework and then calibrated within Vessim's microgrid simulation using an external socket power meter as a definitive measurement source on the system-level. The evaluation of the testbed with GPU and CPU intensive workloads reveal fairly accurate power approximation of the whole computing node by the Kepler framework, with an average regression coefficient of 1.01 and R^2 values of 0.95, though certain machine learning workloads showed higher deviation. The average static y-intercept of the regression line of ~5.23 W indicate inaccuracies in the idle power approximation. Calibration of dynamic per-process power consumption improved accuracy for GPU workloads by ~50%, while CPU workloads saw a modest improvement of ~3.5%.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper proposes a self-calibrating energy-aware software testbed that integrates the Vessim microgrid co-simulation framework with real computing nodes. Application-level power is first approximated via the Kepler framework and then calibrated inside Vessim using an external socket power meter as ground truth. Evaluation on GPU- and CPU-intensive workloads reports an average Kepler regression coefficient of 1.01 and R² of 0.95, with dynamic per-process calibration yielding ~50% accuracy improvement for GPU workloads and ~3.5% for CPU workloads; a static y-intercept of ~5.23 W is noted as indicating idle-power inaccuracies.

Significance. If the calibration transfer from aggregate socket measurements to per-process dynamic terms is robust, the approach would offer a practical middle ground between expensive physical microgrid testbeds and purely simulated environments, enabling more realistic development of carbon-aware scheduling strategies that account for renewable variability and node-level fluctuations.

major comments (1)
  1. [Evaluation] Evaluation section: the headline accuracy claim (regression coefficient 1.01, R² 0.95, ~50% GPU improvement) rests on treating the socket-meter total as unbiased ground truth for scaling dynamic per-process terms, yet the reported static y-intercept of 5.23 W already signals idle-model error; no explicit additivity test or cross-validation on held-out mixed workloads is described to confirm that real-time node fluctuations (thermal, PSU efficiency, shared rails) do not introduce systematic offsets between calibration and deployment traces.
minor comments (2)
  1. [Abstract] Abstract: the statement that 'certain machine learning workloads showed higher deviation' is left unspecified; naming the workloads and reporting the magnitude of those deviations would strengthen the claim of overall accuracy.
  2. [Abstract] Abstract: no error bars, standard deviations, or confidence intervals accompany the reported regression coefficient, R² values, or percentage improvements, limiting assessment of statistical reliability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the single major comment point-by-point below and indicate the planned revisions.

read point-by-point responses

  1. Referee: [Evaluation] Evaluation section: the headline accuracy claim (regression coefficient 1.01, R² 0.95, ~50% GPU improvement) rests on treating the socket-meter total as unbiased ground truth for scaling dynamic per-process terms, yet the reported static y-intercept of 5.23 W already signals idle-model error; no explicit additivity test or cross-validation on held-out mixed workloads is described to confirm that real-time node fluctuations (thermal, PSU efficiency, shared rails) do not introduce systematic offsets between calibration and deployment traces.

    Authors: We agree that the reported static y-intercept of ~5.23 W signals inaccuracies in the idle-power component of the Kepler model, as already stated in the manuscript. The socket power meter is treated as ground truth for aggregate node power, and the linear regression (coefficient 1.01, R² 0.95) is used solely to scale the dynamic per-process terms; the intercept is not propagated into the dynamic calibration. Nevertheless, the manuscript does not include explicit additivity tests or cross-validation on held-out mixed workloads that would quantify possible systematic offsets arising from thermal drift, PSU efficiency curves, or shared power rails. We will revise the evaluation section to (1) explicitly discuss these assumptions and their potential impact on calibration transfer, (2) report the per-workload intercept values alongside the dynamic scaling factors, and (3) add a short paragraph on the limitations of the current validation. Because new mixed-workload experiments would require additional hardware time, we treat this as a partial revision focused on improved exposition rather than new empirical results. revision: partial

Circularity Check

0 steps flagged

No significant circularity; calibration anchored to independent external meter

full rationale

The paper calibrates Kepler per-process approximations inside Vessim by direct comparison to physical socket-meter readings treated as ground truth. Reported metrics (regression coefficient 1.01, R² 0.95, ~50% GPU improvement) are computed from these external measurements rather than from any internal fit or self-referential loop. No self-definitional equations, fitted-input predictions, or load-bearing self-citations appear in the derivation chain. The approach is therefore self-contained against an external benchmark.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the accuracy of two existing external tools and the transferability of socket-meter readings to simulated per-process dynamics; no new entities are postulated.

free parameters (1)
  • static y-intercept
    Average idle-power offset of ~5.23 W obtained from regression; treated as an empirical correction rather than a fitted model parameter.
axioms (2)
  • domain assumption Kepler framework supplies usable initial power estimates for CPU and GPU processes
    Invoked to justify starting the calibration from Kepler outputs.
  • domain assumption Socket power meter readings constitute an unbiased system-level ground truth
    Used as the definitive reference for all calibration steps.

pith-pipeline@v0.9.0 · 5581 in / 1311 out tokens · 46807 ms · 2026-05-15T11:56:23.328850+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

97 extracted references · 97 canonical work pages · 2 internal anchors

  1. [1]

    Exponential laws of computing growth,

    P. J. Denning and T. G. Lewis, “Exponential laws of computing growth, ” Communications of the ACM , vol. 60, no. 1, pp. 54–65, 2016

    work page 2016

  2. [2]

    Tracking Clean En- ergy Progress,

    International Energy Agency (IEA), “Tracking Clean En- ergy Progress, ” 2023. [Online]. Available: https://www.iea.org/ reports/tracking-clean-energy-progress-2023

    work page 2023

  3. [3]

    Language Models are Few-Shot Learners,

    T. Brown et al. , “Language Models are Few-Shot Learners, ” in Advances in Neural Information Processing Systems, Volume 33 (NeurIPS), Curran Associates, Inc., 2020, pp. 1877–1901

    work page 2020

  4. [4]

    arXiv preprint arXiv:2311.16863 , year=

    A. S. Luccioni, Y. Jernite, and E. Strubell, “Power hungry process- ing: Watts driving the cost of ai deployment?. ” arXiv preprint arXiv:2311.16863, 2023

    work page arXiv 2023

  5. [5]

    A. S. Luccioni, S. Viguier, and A.-L. Ligozat, Journal of Machine Learning Research, vol. 24, no. 253, pp. 1–15, 2023

    work page 2023

  6. [6]

    Reducing the Carbon Impact of Generative AI In- ference (today and in 2035),

    A. A. Chien, L. Lin, H. Nguyen, V. Rao, T. Sharma, and R. Wi- jayawardana, “Reducing the Carbon Impact of Generative AI In- ference (today and in 2035), ” in Proceedings of the 2nd Workshop on Sustainable Computer Systems (HotCarbon), Boston, MA, USA: Association for Computing Machinery, 2023

    work page 2035

  7. [7]

    Electricity 2024,

    International Energy Agency (IEA), “Electricity 2024, ” 2024. [On- line]. Available: https://www.iea.org/reports/electricity-2024

    work page 2024

  8. [8]

    ICT sector electricity consumption and greenhouse gas emis- 61 sions--2020 outcome,

    J. Malmodin, N. Lövehagen, P. Bergmark, and D. Lundén, “ICT sector electricity consumption and greenhouse gas emis- 61 sions--2020 outcome, ” Telecommunications Policy, vol. 48, no. 3, Art. no. 102701, 2024

    work page 2020

  9. [9]

    The real climate and transformative impact of ICT: A critique of estimates, trends, and regulations,

    C. Freitag, M. Berners-Lee, K. Widdicks, B. Knowles, G. S. Blair, and A. Friday, “The real climate and transformative impact of ICT: A critique of estimates, trends, and regulations, ” Patterns, vol. 2, no. 9, Art. no. 100576, 2021

    work page 2021

  10. [10]

    Uptime Insti- tute Global Data Center Survey 2021,

    D. Bizo, R. Ascierto, A. Lawrence, and J. Davis, “Uptime Insti- tute Global Data Center Survey 2021, ” Uptime Institute, New York, 2021. [Online]. Available: https://uptimeinstitute.com/ resources/asset/2021-data-center-industry-survey

    work page 2021

  11. [11]

    Carbon-Aware Computing for Datacen- ters,

    A. Radovanović et al., “Carbon-Aware Computing for Datacen- ters, ” IEEE Transactions on Power Systems, vol. 38, no. 2, pp. 1270– 1280, 2022

    work page 2022

  12. [12]

    Carbon Explorer: A Holistic Framework for De- signing Carbon Aware Datacenters,

    B. Acun et al., “Carbon Explorer: A Holistic Framework for De- signing Carbon Aware Datacenters, ” in Proceedings of the 28th ACM International Conference on Architectural Support for Pro- gramming Languages and Operating Systems, Volume 2 (ASP- LOS), Vancouver, Canada: Association for Computing Machin- ery, 2023, pp. 118–132

    work page 2023

  13. [13]

    Ecovisor: A Virtual Energy System for Carbon- Efficient Applications,

    A. Souza et al., “Ecovisor: A Virtual Energy System for Carbon- Efficient Applications, ” in Proceedings of the 28th ACM Interna- tional Conference on Architectural Support for Programming Lan- guages and Operating Systems, Volume 2 (ASPLOS) , Vancouver, Canada: Association for Computing Machinery, 2023, pp. 252– 265

    work page 2023

  14. [14]

    Treehouse: A case for carbon-aware datacenter software,

    T. Anderson, A. Belay, M. Chowdhury, A. Cidon, and I. Zhang, “Treehouse: A case for carbon-aware datacenter software, ” ACM SIGENERGY Energy Informatics Review , vol. 3, no. 3, pp. 64–70, 2023

    work page 2023

  15. [15]

    Let's wait awhile: how temporal workload shifting can re- duce carbon emissions in the cloud,

    P. Wiesner, I. Behnke, D. Scheinert, K. Gontarska, and L. Tham- sen, “Let's wait awhile: how temporal workload shifting can re- duce carbon emissions in the cloud, ” in Proceedings of the 22nd International Middleware Conference (Middleware) , Québec city, 62 Canada: Association for Computing Machinery, 2021, pp. 260– 272

    work page 2021

  16. [16]

    Enabling sustainable clouds: The case for virtu- alizing the energy system,

    N. Bashir et al., “Enabling sustainable clouds: The case for virtu- alizing the energy system, ” in Proceedings of the ACM Symposium on Cloud Computing (SoCC) , Seattle, WA, USA: Association for Computing Machinery, 2021, pp. 350–358

    work page 2021

  17. [17]

    Sustainable com- puting-without the hot air,

    N. Bashir, D. Irwin, P. Shenoy, and A. Souza, “Sustainable com- puting-without the hot air, ” ACM SIGENERGY Energy Informatics Review, vol. 3, no. 3, pp. 47–52, 2023

    work page 2023

  18. [18]

    FedZero: Leveraging Renewable Excess Energy in Fed- erated Learning,

    P. Wiesner, R. Khalili, D. Grinwald, P. Agrawal, L. Thamsen, and O. Kao, “FedZero: Leveraging Renewable Excess Energy in Fed- erated Learning, ” in Proceedings of the 15th ACM International Conference on Future and Sustainable Energy Systems (e-Energy) , Singapore: Association for Computing Machinery, 2024, pp. 37– 49

    work page 2024

  19. [19]

    Not all doom and gloom: How energy-intensive and temporally flexible data center applications may actually promote renewable energy sources,

    G. Fridgen, M.-F. Körner, S. Walters, and M. Weibelzahl, “Not all doom and gloom: How energy-intensive and temporally flexible data center applications may actually promote renewable energy sources, ” Business & Information Systems Engineering , vol. 63, 2021

    work page 2021

  20. [20]

    Mitigating curtailment and carbon emissions through load migration between data centers,

    J. Zheng, A. A. Chien, and S. Suh, “Mitigating curtailment and carbon emissions through load migration between data centers, ” Joule, vol. 4, no. 10, pp. 2208–2222, 2020

    work page 2020

  21. [21]

    LACS: Learning-Augmented Algorithms for Carbon-Aware Resource Scaling with Uncertain Demand

    R. Bostandoost, A. Lechowicz, W. A. Hanafy, N. Bashir, P. Shenoy, and M. Hajiesmaili, “LACS: Learning-Augmented Algorithms for Carbon-Aware Resource Scaling with Uncertain Demand. ” arXiv preprint arXiv:2404.15211, 2024

    work page arXiv 2024

  22. [22]

    Vessim: A Testbed for Carbon-Aware Applications and Systems,

    P. Wiesner, I. Behnke, P. Kilian, M. Steinke, and O. Kao, “Vessim: A Testbed for Carbon-Aware Applications and Systems, ” in Pro- ceedings of the 3rd Workshop on Sustainable Computer Systems (HotCarbon), New York, NY, USA: Association for Computing Machinery, 2024. 63

    work page 2024

  23. [23]

    Soft- ware-in-the-loop simulation for developing and testing carbon- aware applications,

    P. Wiesner, M. Steinke, H. Nickel, Y. Kitana, and O. Kao, “Soft- ware-in-the-loop simulation for developing and testing carbon- aware applications, ” Software: Practice and Experience, vol. 53, no. 12, pp. 2362–2376, 2023

    work page 2023

  24. [24]

    Kepler: A framework to calculate the energy consumption of containerized applications,

    M. Amaral et al., “Kepler: A framework to calculate the energy consumption of containerized applications, ” in IEEE 16th Interna- tional Conference on Cloud Computing (CLOUD), Chicago, Illinois USA: IEEE, 2023, pp. 69–71

    work page 2023

  25. [25]

    RAPL: Memory power estimation and capping,

    H. David, E. Gorbatov, U. R. Hanebutte, R. Khanna, and C. Le, “RAPL: Memory power estimation and capping, ” in Proceedings of the 16th ACM/IEEE international symposium on Low power elec- tronics and design, 2010, pp. 189–194

    work page 2010

  26. [26]

    Measuring energy consumption for short code paths using RAPL,

    M. Hähnel, B. Döbel, M. Völp, and H. Härtig, “Measuring energy consumption for short code paths using RAPL, ” ACM SIGMET- RICS Performance Evaluation Review , vol. 40, no. 3, pp. 13–17, 2012

    work page 2012

  27. [27]

    Power measurement techniques on standard compute nodes: A quantitative comparison,

    D. Hackenberg, T. Ilsche, R. Schöne, D. Molka, M. Schmidt, and W. E. Nagel, “Power measurement techniques on standard compute nodes: A quantitative comparison, ” in 2013 IEEE International Symposium on Performance Analysis of Systems and Software (IS- PASS), 2013, pp. 194–204

    work page 2013

  28. [28]

    An Energy Efficiency Feature Survey of the Intel Haswell Processor,

    D. Hackenberg, R. Schöne, T. Ilsche, D. Molka, J. Schuchart, and R. Geyer, “An Energy Efficiency Feature Survey of the Intel Haswell Processor, ” in 2015 IEEE International Parallel and Distributed Pro- cessing Symposium Workshop, 2015, pp. 896–904

    work page 2015

  29. [29]

    Measuring GPU power with the K20 built-in sensor,

    M. Burtscher, I. Zecena, and Z. Zong, “Measuring GPU power with the K20 built-in sensor, ” in Proceedings of Workshop on Gen- eral Purpose Processing Using GPUs, 2014, pp. 28–36

    work page 2014

  30. [30]

    Part-time Power Mea- surements: nvidia-smi's Lack of Attention,

    Z. Yang, K. Adamek, and W. Armour, “Part-time Power Mea- surements: nvidia-smi's Lack of Attention, ” arXiv preprint arXiv:2312.02741, 2023

    work page arXiv 2023

  31. [31]

    NVIDIA Management Library (NVML) API Reference,

    “NVIDIA Management Library (NVML) API Reference, ” NVIDIA Corporation, 2024. 64

    work page 2024

  32. [32]

    The benefits of event: driven energy accounting in power-sensitive systems,

    F. Bellosa, “The benefits of event: driven energy accounting in power-sensitive systems, ” in Proceedings of the 9th workshop on ACM SIGOPS European workshop: beyond the PC: new challenges for the operating system, 2000, pp. 37–42

    work page 2000

  33. [33]

    SmartWatts: Self-calibrat- ing software-defined power meter for containers,

    G. Fieni, R. Rouvoy, and L. Seinturier, “SmartWatts: Self-calibrat- ing software-defined power meter for containers, ” in 2020 20th IEEE/ACM International Symposium on Cluster, Cloud and Inter- net Computing (CCGRID), 2020, pp. 479–488

    work page 2020

  34. [34]

    WattScope: Non- intrusive application-level power disaggregation in datacenters,

    X. Guan, N. Bashir, D. Irwin, and P. Shenoy, “WattScope: Non- intrusive application-level power disaggregation in datacenters, ” Performance Evaluation, vol. 162, Art. no. 102369, 2023

    work page 2023

  35. [35]

    Process-level power estimation in vm-based systems,

    M. Colmant, M. Kurpicz, P. Felber, L. Huertas, R. Rouvoy, and A. Sobe, “Process-level power estimation in vm-based systems, ” in Proceedings of the Tenth European Conference on Computer Sys- tems, 2015

    work page 2015

  36. [36]

    How to build valid and credible simulation models,

    A. M. Law, “How to build valid and credible simulation models, ” in 2022 Winter Simulation Conference (WSC) , 2022, pp. 1283– 1295

    work page 2022

  37. [37]

    A. M. Law, W. D. Kelton, and W. D. Kelton, Simulation modeling and analysis, vol. 3. New York: Mcgraw-hill New York, 2007

    work page 2007

  38. [38]

    An introduction to regression analysis,

    A. O. Sykes, “An introduction to regression analysis, ” 1993

    work page 1993

  39. [39]

    Bayesian calibration of com- puter models,

    M. C. Kennedy and A. O'Hagan, “Bayesian calibration of com- puter models, ” Journal of the Royal Statistical Society: Series B (S- tatistical Methodology), vol. 63, no. 3, pp. 425–464, 2001

    work page 2001

  40. [40]

    Trends in microgrid control,

    D. E. Olivares et al., “Trends in microgrid control, ” IEEE Transac- tions on smart grid, vol. 5, no. 4, pp. 1905–1919, 2014

    work page 1905

  41. [41]

    En- hancing smart grid with microgrids: Challenges and opportuni- ties,

    Y. Yoldaş, A. Önen, S. Muyeen, A. V. Vasilakos, and I. Alan, “En- hancing smart grid with microgrids: Challenges and opportuni- ties, ” Renewable and Sustainable Energy Reviews, vol. 72, pp. 205– 214, 2017

    work page 2017

  42. [42]

    An overview on functional integration of hybrid renew- 65 able energy systems in multi-energy buildings,

    L. Canale, A. R. Di Fazio, M. Russo, A. Frattolillo, and M. Del- l’Isola, “An overview on functional integration of hybrid renew- 65 able energy systems in multi-energy buildings, ” Energies, vol. 14, no. 4, Art. no. 1078, 2021

    work page 2021

  43. [43]

    Distributed real-time energy man- agement in data center microgrids,

    L. Yu, T. Jiang, and Y. Zou, “Distributed real-time energy man- agement in data center microgrids, ” IEEE Transactions on Smart Grid, vol. 9, no. 4, pp. 3748–3762, 2016

    work page 2016

  44. [44]

    Cpes testing with mosaik: Co-simulation planning, execution and analysis,

    C. Steinbrink et al. , “Cpes testing with mosaik: Co-simulation planning, execution and analysis, ” Applied Sciences, vol. 9, no. 5, Art. no. 923, 2019

    work page 2019

  45. [45]

    RAPL in Action: Experiences in Using RAPL for Power Measurements,

    K. N. Khan, M. Hirki, T. Niemi, J. K. Nurminen, and Z. Ou, “RAPL in Action: Experiences in Using RAPL for Power Measurements, ” ACM Transactions on Modeling and Performance Evaluation of Computing Systems, vol. 3, no. 2, Art. no. 9, 2018

    work page 2018

  46. [46]

    Advanced Configuration and Power Interface (ACPI) Specifica- tion Release 6.5,

    “Advanced Configuration and Power Interface (ACPI) Specifica- tion Release 6.5, ” UEFI Forum Inc, 2022

    work page 2022

  47. [47]

    Intelligent Platform Management Interface Specification v2.0 rev. 1.1,

    “Intelligent Platform Management Interface Specification v2.0 rev. 1.1, ” Intel Corporation, Hewlett-Packard Company, NEC Corporation, Dell Inc, 2013

    work page 2013

  48. [48]

    Redfish Specification DSP0266 Whitepaper,

    “Redfish Specification DSP0266 Whitepaper, ” DMTF, 2024

    work page 2024

  49. [49]

    Virtual machine power metering and provisioning,

    A. Kansal, F. Zhao, J. Liu, N. Kothari, and A. A. Bhattacharya, “Virtual machine power metering and provisioning, ” in Proceed- ings of the 1st ACM symposium on Cloud computing , 2010, pp. 39–50

    work page 2010

  50. [50]

    HaPPy: Hy- perthread-aware Power Profiling Dynamically,

    Y. Zhai, X. Zhang, S. Eranian, L. Tang, and J. Mars, “HaPPy: Hy- perthread-aware Power Profiling Dynamically, ” in 2014 USENIX annual technical conference (USENIX ATC 14), 2014, pp. 211–217

    work page 2014

  51. [51]

    Bitwatts: A process-level power monitoring middleware,

    M. Colmant, M. Kurpicz, P. Felber, L. Huertas, R. Rouvoy, and A. Sobe, “Bitwatts: A process-level power monitoring middleware, ” in Proceedings of the Posters and Demos Session of the 15th Inter- national Middleware Conference, 2014, pp. 41–42

    work page 2014

  52. [52]

    Power containers: An OS facility for fine-grained power and 66 energy management on multicore servers,

    K. Shen, A. Shriraman, S. Dwarkadas, X. Zhang, and Z. Chen, “Power containers: An OS facility for fine-grained power and 66 energy management on multicore servers, ” ACM SIGARCH Com- puter Architecture News, vol. 41, no. 1, pp. 65–76, 2013

    work page 2013

  53. [53]

    Pow- erapi: A software library to monitor the energy consumed at the process-level,

    A. Bourdon, A. Noureddine, R. Rouvoy, and L. Seinturier, “Pow- erapi: A software library to monitor the energy consumed at the process-level, ” ERCIM News, vol. 92, 2013

    work page 2013

  54. [54]

    Process-Based Efficient Power Level Exporter,

    M. Amaral et al., “Process-Based Efficient Power Level Exporter, ” in 2024 IEEE 17th International Conference on Cloud Computing (CLOUD), 2024, pp. 456–467

    work page 2024

  55. [55]

    Exploring Kepler's potentials: unveil- ing cloud application power consumption,

    M. Amaral, S. Choochotkaew, E. K. Lee, Chen Huamin, and Eilam Tamar, “Exploring Kepler's potentials: unveil- ing cloud application power consumption, ” Cloud Native Computing Foundation (CNCF) Blog Post. Accessed: Aug. 17, 2024. [Online]. Available: https://www.cncf.io/blog/2023/ 10/11/exploring-keplers-potentials-unveiling-cloud-application- power-consumption/

    work page 2024

  56. [56]

    ICT Sector Guidance built on the GHG Protocol Product Life Cycle Accounting and Reporting Standard,

    G. Protocol, “ICT Sector Guidance built on the GHG Protocol Product Life Cycle Accounting and Reporting Standard, ” Global: GHG Protocol, 2017

    work page 2017

  57. [57]

    Car- bonscaler: Leveraging cloud workload elasticity for optimizing carbon-efficiency,

    W. A. Hanafy, Q. Liang, N. Bashir, D. Irwin, and P. Shenoy, “Car- bonscaler: Leveraging cloud workload elasticity for optimizing carbon-efficiency, ” Proceedings of the ACM on Measurement and Analysis of Computing Systems, vol. 7, no. 3, Art. no. 57, 2023

    work page 2023

  58. [58]

    Energy-Aware Process Schedul- ing in Linux,

    F. Qiao, Y. Fang, and A. Cidon, “Energy-Aware Process Schedul- ing in Linux, ” in Proceedings of the 3rd Workshop on Sustainable Computer Systems (HotCarbon), New York, NY, USA: Association for Computing Machinery, 2024

    work page 2024

  59. [59]

    ghost: Fast & flexible user-space delega- tion of linux scheduling,

    J. T. Humphries et al., “ghost: Fast & flexible user-space delega- tion of linux scheduling, ” in Proceedings of the ACM SIGOPS 28th Symposium on Operating Systems Principles, 2021, pp. 588–604

    work page 2021

  60. [60]

    Dynamic power budget redistribution under a power cap on multi-application environ- ments,

    L. Costero, F. D. Igual, and K. Olcoz, “Dynamic power budget redistribution under a power cap on multi-application environ- ments, ” Sustainable Computing: Informatics and Systems , vol. 38, Art. no. 100865, 2023. 67

    work page 2023

  61. [61]

    OpenMP: an industry standard API for shared-memory programming,

    L. Dagum and R. Menon, “OpenMP: an industry standard API for shared-memory programming, ” IEEE computational science and engineering, vol. 5, no. 1, 1998

    work page 1998

  62. [62]

    An Alloca- tion Model for Attributing Emissions in Multi-tenant Cloud Data Centers,

    R. Westerhof, R. Atherton, and V. Andrikopoulos, “An Alloca- tion Model for Attributing Emissions in Multi-tenant Cloud Data Centers, ” arXiv preprint arXiv:2305.10439, 2023

    work page arXiv 2023

  63. [63]

    carbond: An Operating-System Daemon for Carbon Awareness,

    A. Schmidt, G. Stock, R. Ohs, L. Gerhorst, B. Herzog, and T. Hönig, “carbond: An Operating-System Daemon for Carbon Awareness, ” in Proceedings of the 2nd Workshop on Sustainable Computer Systems (HotCarbon) , Boston, MA, USA: Association for Computing Machinery, 2023

    work page 2023

  64. [64]

    Towards Game-The- oretic Approaches to Attributing Carbon in Cloud Data Centers,

    L. Han, J. Kakadia, B. C. Lee, and U. Gupta, “Towards Game-The- oretic Approaches to Attributing Carbon in Cloud Data Centers, ” in Proceedings of the 3rd Workshop on Sustainable Computer Sys- tems (HotCarbon), New York, NY, USA: Association for Comput- ing Machinery, 2024

    work page 2024

  65. [65]

    EnergAt: Fine- Grained Energy Attribution for Multi-Tenancy,

    Hé Hongyu, M. Friedman, and T. Rekatsinas, “EnergAt: Fine- Grained Energy Attribution for Multi-Tenancy, ” in Proceedings of the 2nd Workshop on Sustainable Computer Systems (HotCarbon), Boston, MA, USA: Association for Computing Machinery, 2023

    work page 2023

  66. [66]

    Advancing Cloud Sustainability: A Ver- satile Framework for Container Power Model Training,

    S. Choochotkaew et al., “Advancing Cloud Sustainability: A Ver- satile Framework for Container Power Model Training, ” in 2023 31st International Symposium on Modeling, Analysis, and Simu- lation of Computer and Telecommunication Systems (MASCOTS) , 2023

    work page 2023

  67. [67]

    Open-source mqtt evaluation,

    M. Bender, E. Kirdan, M.-O. Pahl, and G. Carle, “Open-source mqtt evaluation, ” in 2021 IEEE 18th Annual Consumer Communi- cations & Networking Conference (CCNC), 2021, pp. 1–4

    work page 2021

  68. [68]

    Desbiens, “Mqtt, ” Building Enterprise IoT Solutions with Eclipse IoT Technologies: An Open Source Approach to Edge Computing

    F. Desbiens, “Mqtt, ” Building Enterprise IoT Solutions with Eclipse IoT Technologies: An Open Source Approach to Edge Computing . Springer, pp. 67–101, 2022

    work page 2022

  69. [69]

    NVIDIA A100 Tensor Core GPU: Performance and Inno- 68 vation,

    J. Choquette, W. Gandhi, O. Giroux, N. Stam, and R. Krashin- sky, “NVIDIA A100 Tensor Core GPU: Performance and Inno- 68 vation, ” IEEE Micro, vol. 41, no. 2, pp. 29–35, 2021, doi: 10.1109/ MM.2021.3061394

    work page arXiv 2021

  70. [70]

    Dockercap: A software-level power capping orchestrator for docker contain- ers,

    A. Asnaghi, M. Ferroni, and M. D. Santambrogio, “Dockercap: A software-level power capping orchestrator for docker contain- ers, ” in 2016 IEEE Intl Conference on Computational Science and Engineering (CSE) and IEEE Intl Conference on Embedded and Ubiquitous Computing (EUC) and 15th Intl Symposium on Dis- tributed Computing and Applications for Business E...

    work page 2016

  71. [71]

    PRESTO: a latency-aware power-capping orchestrator for cloud-native microservices,

    R. Brondolin and M. D. Santambrogio, “PRESTO: a latency-aware power-capping orchestrator for cloud-native microservices, ” in 2020 IEEE International Conference on Autonomic Computing and Self-Organizing Systems (ACSOS), 2020, pp. 11–20

    work page 2020

  72. [72]

    Thunderbolt:Throughput-Optimized, Quality-of-Ser- vice-Aware Power Capping at Scale,

    S. Li et al., “Thunderbolt:Throughput-Optimized, Quality-of-Ser- vice-Aware Power Capping at Scale, ” in 14th USENIX Symposium on Operating Systems Design and Implementation (OSDI) , 2020, pp. 1241–1255

    work page 2020

  73. [73]

    Multi-Process Service Release r550,

    “Multi-Process Service Release r550, ” NVIDIA Corporation, 2024

    work page 2024

  74. [74]

    NVIDIA Multi-Instance GPU and NVIDIA Virtual Compute Server,

    “NVIDIA Multi-Instance GPU and NVIDIA Virtual Compute Server, ” NVIDIA Corporation, 2020

    work page 2020

  75. [75]

    EPC 8041/8045 Series User Manual

    “EPC 8041/8045 Series User Manual. ” 2022. [Online]. Avail- able: https://gude-systems.com/app/uploads/2022/05/manual- epc8041-series.pdf

    work page 2022

  76. [76]

    A. R. Hambley, Electrical engineering: principles and applications, vol. 4. Prentice Hall, 2017

    work page 2017

  77. [77]

    Bermbach, E

    D. Bermbach, E. Wittern, and S. Tai, Cloud service benchmarking. Springer, 2017

    work page 2017

  78. [78]

    The UNIX time-sharing system,

    D. M. Ritchie and K. Thompson, “The UNIX time-sharing system, ” Communications of the ACM, vol. 17, no. 7, pp. 365–375, 1974

    work page 1974

  79. [79]

    Kernelet: High-throughput GPU kernel ex- ecutions with dynamic slicing and scheduling,

    J. Zhong and B. He, “Kernelet: High-throughput GPU kernel ex- ecutions with dynamic slicing and scheduling, ” IEEE Transactions 69 on Parallel and Distributed Systems, vol. 25, no. 6, pp. 1522–1532, 2013

    work page 2013

  80. [80]

    An open-source benchmark suite for microservices and their hardware-software implications for cloud & edge sys- tems,

    Y. Gan et al., “An open-source benchmark suite for microservices and their hardware-software implications for cloud & edge sys- tems, ” in Proceedings of the Twenty-Fourth International Confer- ence on Architectural Support for Programming Languages and Operating Systems (ASPLOS), 2019, pp. 3–18

    work page 2019

Showing first 80 references.