pith. sign in

arxiv: 2605.16260 · v1 · pith:3SAWSVKUnew · submitted 2026-02-14 · 💻 cs.DC

The Impact of Process Competition on Energy Consumption: Analysis and Modeling

Eduardo Gomes Campos , Rafaela Sousa de Alencar Lacerda , Adnei Willian Donatti , Joberto S. B. Martins , Charles C. Miers , Tereza C. M. B. Carvalho This is my paper

Pith reviewed 2026-05-21 13:20 UTC · model grok-4.3

classification 💻 cs.DC
keywords energy consumptionprocess competitioncloud computingprocessor coresresource sharingenergy efficiencyvirtual machinestask scheduling
0
0 comments X

The pith

A process consumes energy in a linear way with rising competition when few cores are available, but the relationship becomes a root function once the machine has many cores.

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

This paper measures how much energy a single process draws when other processes compete with it for the same hardware resources. Experiments show the shape of that energy curve depends directly on the total number of processor cores present on the physical machine. With few cores the energy cost grows roughly in proportion to the amount of competition; with many cores the cost grows much more slowly, following a root shape. The finding matters for cloud operators who must decide how to price shared machines and how to place workloads so that total power draw stays low.

Core claim

Experiments establish that a process's energy consumption as a function of competition for computational resources transitions from linear to a root function as the number of processor cores on the hosting physical machine increases.

What carries the argument

The number of processor cores on the physical machine, which governs whether energy consumption scales linearly or as a root function with the level of resource competition faced by the process.

If this is right

  • Cloud pricing models can be made more accurate by charging differently on low-core versus high-core machines according to the measured energy curves.
  • Task schedulers gain a concrete rule for placing processes: on machines with many cores the marginal energy cost of added competition drops, allowing denser packing without proportional power increase.
  • Load-balancing algorithms can use the core count as a direct input to decide when moving a process will actually reduce total system energy.
  • Energy-efficiency analyses of container or virtual-machine deployments become quantitative once the linear-to-root transition point is known for each hardware generation.

Where Pith is reading between the lines

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

  • Data-center operators might prefer machines with core counts above the observed transition point when the goal is to maximize concurrent users per watt.
  • The same measurement technique could be applied to other contended resources such as last-level cache or memory bandwidth to test whether similar functional transitions appear.
  • If the root-function regime holds across many workloads, it supplies a simple rule of thumb for estimating the energy penalty of over-subscription on modern many-core servers.

Load-bearing premise

The experiments isolate competition for processor cores as the only variable that changes the measured energy consumption, without meaningful interference from memory bandwidth, I/O activity, or thermal throttling.

What would settle it

Repeating the competition measurements on a high-core machine and still obtaining a strictly linear energy curve, or on a low-core machine and obtaining a root curve, would show the claimed core-dependent transition does not hold.

Figures

Figures reproduced from arXiv: 2605.16260 by Adnei Willian Donatti, Charles C. Miers, Eduardo Gomes Campos, Joberto S. B. Martins, Rafaela Sousa de Alencar Lacerda, Tereza C. M. B. Carvalho.

Figure 1
Figure 1. Figure 1: Diagram representing a topology for the NSs [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Diagram representing the test environment and its [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Power consumption of a constant process 0 1000 2000 3000 4000 5000 6000 Time (s) 9.0 9.5 10.0 10.5 11.0 11.5 12.0 12.5 13.0 Power consumption (W) Energy consumption for process in competition context Sample data [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Power consumption with and without competi [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Worker 1 results 0 20 40 60 80 CPU being used by competition (%) 9 10 11 12 13 14 Power consumed by process (W) Sample data Regression = 8.14 + 0.70 x [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Worker 2 results 0 20 40 60 80 CPU being used by competition (%) 8 9 10 11 12 13 14 15 Power consumed by process (W) Sample data Regression = 8.69 + 0.71 x [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Worker 3 results pictured by an n-root function. Besides, it is known from [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 12
Figure 12. Figure 12: Worker 3 results with 4 cores 0 20 40 60 80 CPU being used by competition (%) 6 8 10 12 14 Power consumed by process (W) Sample data Regression = 6.04 + 0.11x [PITH_FULL_IMAGE:figures/full_fig_p008_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Worker 5 results with 4 cores 0 20 40 60 80 CPU being used by competition (%) 4 5 6 7 8 9 Power consumed by process (W) Sample data Regression = 4.38 + 0.06x [PITH_FULL_IMAGE:figures/full_fig_p008_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Worker 2 results with 6 cores and linear [PITH_FULL_IMAGE:figures/full_fig_p008_14.png] view at source ↗
Figure 11
Figure 11. Figure 11: Worker 2 results with 4 cores 0 20 40 60 80 CPU being used by competition (%) 6 8 10 12 14 Power consumed by process (W) Sample data Regression = 6.22 + 0.11x [PITH_FULL_IMAGE:figures/full_fig_p008_11.png] view at source ↗
Figure 15
Figure 15. Figure 15: Worker 2 results with 6 cores and n-root [PITH_FULL_IMAGE:figures/full_fig_p009_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Worker 3 results with 6 cores and n-root [PITH_FULL_IMAGE:figures/full_fig_p009_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Worker 3 results with 6 cores and linear [PITH_FULL_IMAGE:figures/full_fig_p009_17.png] view at source ↗
read the original abstract

With the development of distributed systems, the need to manage the sharing of machines among multiple simultaneous users arises. In the cloud computing context, the instantiation of virtual machines and containers by different users utilizing the same infrastructure leads to a dispute for physical computational resources. In this regard, this paper analyses a process's energy consumption as a function of the competition for computational resources it encounters. Investigating this behavior is fundamental for many applications, such as pricing in cloud computing services, and for task scheduling and load balancing, while increasing energy efficiency. To determine this behavior, experiments were conducted and resulted in a dependency on the number of processor cores of the physical machine hosting the process. As the number of cores increases, the process's energy consumption as a function of the competition it faces transitions from linear to a root function.

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 manuscript analyzes a process's energy consumption as a function of competition for computational resources (e.g., cores) in shared environments such as cloud computing. Experiments are reported to show that this functional dependence transitions from linear to a root function as the number of processor cores on the physical host increases.

Significance. A validated transition in energy scaling with core count could support improved energy-aware scheduling and pricing in distributed systems. The manuscript provides no machine-checked derivations, reproducible code, or parameter-free predictions, so its contribution rests entirely on the experimental observation.

major comments (1)
  1. [Abstract / Experimental Section] Abstract and experimental description: the central claim requires that measured energy depends only on core-sharing competition and changes functional form with core count, yet no details are supplied on workloads, measurement method (RAPL or equivalent), core-pinning, number of trials, error bars, or controls for memory-bandwidth saturation, I/O contention, or DVFS/thermal effects. If any of these covary with the number of competing processes, the reported linear-to-root transition could be an artifact.
minor comments (2)
  1. Define 'competition' quantitatively (e.g., number of co-scheduled processes or threads) and specify the exact root function (square root, etc.) with fitting procedure.
  2. Add a table or figure showing raw data points, fitted curves, and goodness-of-fit metrics for different core counts.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful review and constructive feedback on our manuscript. We address the major comment below and will revise the manuscript to strengthen the experimental description.

read point-by-point responses

  1. Referee: [Abstract / Experimental Section] Abstract and experimental description: the central claim requires that measured energy depends only on core-sharing competition and changes functional form with core count, yet no details are supplied on workloads, measurement method (RAPL or equivalent), core-pinning, number of trials, error bars, or controls for memory-bandwidth saturation, I/O contention, or DVFS/thermal effects. If any of these covary with the number of competing processes, the reported linear-to-root transition could be an artifact.

    Authors: We agree that the experimental section requires substantially more detail to support the central claim. The current manuscript provides only a high-level description of the experiments. In the revised version we will expand the Experimental Section with: (1) explicit workload descriptions (CPU-bound benchmarks and their parameters), (2) the energy measurement method (RAPL interface on the specific platform), (3) core-pinning policy and how competing processes were scheduled, (4) number of independent trials and statistical reporting including error bars, and (5) explicit controls or measurements confirming that memory bandwidth, I/O, DVFS, and thermal throttling did not covary with the number of competing processes. These additions will demonstrate that the observed linear-to-root transition is attributable to core-sharing competition rather than confounding factors. revision: yes

Circularity Check

0 steps flagged

No circularity: central claim is direct experimental observation

full rationale

The paper states that experiments were conducted to determine the behavior of a process's energy consumption as a function of competition for resources, resulting in a dependency on the number of processor cores where the functional form transitions from linear to root as core count increases. No derivation chain, mathematical model, fitted parameters presented as predictions, self-citations, or ansatz is described in the provided text that would reduce the reported transition to an input by construction. The result is presented as an empirical finding from measurements rather than a self-referential or fitted-output claim.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim is empirical and therefore rests primarily on measurement assumptions rather than mathematical axioms or new entities.

axioms (1)
  • domain assumption Energy consumption of a process can be isolated and measured as a function of CPU-core competition alone.
    This premise is required for the experimental results to support the stated functional transition.

pith-pipeline@v0.9.0 · 5696 in / 1124 out tokens · 50632 ms · 2026-05-21T13:20:29.355052+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

40 extracted references · 40 canonical work pages · 1 internal anchor

  1. [1]

    [n. d.]. Amazon pricing for VMs. https://aws.amazon.com/pt/ec2/pricing/on- demand/. Accessed: 2025-12-16

    work page 2025

  2. [2]

    [n. d.]. CpuLimit repository. https://github.com/opsengine/cpulimit. Accessed: 2024-07-11

    work page 2024

  3. [3]

    [n. d.]. Stress tool description. https://man.archlinux.org/man/stress.1. Accessed: 2024-07-11

    work page 2024

  4. [4]

    Thangaraju

    Vaibhav Aggarwal and B. Thangaraju. 2020. Performance Analysis of Virtualisa- tion Technologies in NFV and Edge Deployments. In2020 IEEE International Con- ference on Electronics, Computing and Communication Technologies (CONECCT). 1–5. doi:10.1109/CONECCT50063.2020.9198367

    work page doi:10.1109/conecct50063.2020.9198367 2020

  5. [5]

    Patrick Kwadwo Agyapong, Mikio Iwamura, Dirk Staehle, Wolfgang Kiess, and Anass Benjebbour. 2014. Design considerations for a 5G network architecture. IEEE Communications Magazine52, 11 (2014), 65–75. doi:10.1109/MCOM.2014. 6957145

    work page doi:10.1109/mcom.2014 2014

  6. [6]

    2024.Test results for Competition for CPU resources experiments

    Campos. 2024.Test results for Competition for CPU resources experiments. doi:10. 5281/zenodo.13221177

    work page 2024

  7. [7]

    Jeffrey Chase, Darrell Anderson, Prachi Thakar, Amin Vahdat, and Ronald Doyle

    work page

  8. [8]

    doi:10.1145/502034.502045

    Managing Energy and Server Resources in Hosting Centres.Operating Systems Review - SIGOPS35, 103–116. doi:10.1145/502034.502045

    work page doi:10.1145/502034.502045

  9. [9]

    Hui Chen, Youhuizi Li, and Weisong Shi. 2012. Fine-grained power management using process-level profiling.Sustainable Computing: Informatics and Systems2, 1 (2012), 33–42. doi:10.1016/j.suscom.2012.01.002

    work page doi:10.1016/j.suscom.2012.01.002 2012

  10. [10]

    Miyuru Dayarathna, Yonggang Wen, and Rui Fan. 2016. Data Center Energy Consumption Modeling: A Survey.IEEE Communications Surveys & Tutorials18, 1 (2016), 732–794. doi:10.1109/COMST.2015.2481183

    work page doi:10.1109/comst.2015.2481183 2016

  11. [11]

    Zhijun Ding, Song Wang, and Changjun Jiang. 2023. Kubernetes-Oriented Mi- croservice Placement With Dynamic Resource Allocation.IEEE Transactions on Cloud Computing11, 2 (2023), 1777–1793. doi:10.1109/TCC.2022.3161900

    work page doi:10.1109/tcc.2022.3161900 2023

  12. [12]

    Thanh Do, Suhib Rawshdeh, and Weisong Shi. 2009. pTop: A Process-level Power Profiling Tool

    work page 2009

  13. [13]

    Antunes, Eli Carlos Figueiredo Souza, Sand Correa, Tiago Ferreto, José Augusto Suruagy, Joberto S

    Adnei Willian Donatti, Marcia Cristina Machado, Marvin Alexander Lopez Mar- tinez, Sabino Rogério S. Antunes, Eli Carlos Figueiredo Souza, Sand Correa, Tiago Ferreto, José Augusto Suruagy, Joberto S. B. Martins, and Tereza Cristina Car- valho. 2025. Energy Efficiency in Network Slicing: Survey and Taxonomy.IEEE Access(2025), 1–20. doi:10.1109/ACCESS.2025.3590365

    work page doi:10.1109/access.2025.3590365 2025

  14. [14]

    Meryeme El Yadari, Ali Yahyaouy, Stéphane Le Masson, Khalid El Fazazy, and Hamid Gualous. 2022. Study of the correlation between server resources utiliza- tion and energy consumption. In2022 10th International Conference on Systems and Control (ICSC). 125–130. doi:10.1109/ICSC57768.2022.9993950

    work page doi:10.1109/icsc57768.2022.9993950 2022

  15. [15]

    Xiaobo Fan, Wolf-Dietrich Weber, and Luiz Andre Barroso. 2007. Power provi- sioning for a warehouse-sized computer.SIGARCH Comput. Archit. News35, 2 (jun 2007), 13–23. doi:10.1145/1273440.1250665

    work page doi:10.1145/1273440.1250665 2007

  16. [16]

    Xiaobo Fan, Wolf-Dietrich Weber, and Luiz Andre Barroso. 2007. Power pro- visioning for a warehouse-sized computer. InProceedings of the 34th Annual International Symposium on Computer Architecture(San Diego, California, USA) (ISCA ’07). Association for Computing Machinery, New York, NY, USA, 13–23. doi:10.1145/1250662.1250665

    work page doi:10.1145/1250662.1250665 2007

  17. [17]

    Vishal Gupta, Ripal Nathuji, and Karsten Schwan. 2011. An analysis of power reduction in datacenters using heterogeneous chip multiprocessors.SIGMETRICS Perform. Eval. Rev.39, 3 (dec 2011), 87–91. doi:10.1145/2160803.2160867

    work page doi:10.1145/2160803.2160867 2011

  18. [18]

    Nancy Jain and Sakshi Choudhary. 2016. Overview of virtualization in cloud computing. In2016 Symposium on Colossal Data Analysis and Networking (CDAN). 1–4. doi:10.1109/CDAN.2016.7570950

    work page doi:10.1109/cdan.2016.7570950 2016

  19. [19]

    Hiroki Kataoka, Dilawaer Duolikun, Tomoya Enokido, and Makoto Takizawa

    work page

  20. [20]

    In2015 IEEE 29th International Conference on Advanced Information Networking and Applications Workshops

    Power Consumption and Computation Models of a Server with a Multi-core CPU and Experiments. In2015 IEEE 29th International Conference on Advanced Information Networking and Applications Workshops. 217–222. doi:10.1109/WAINA.2015.127

    work page doi:10.1109/waina.2015.127 2015

  21. [21]

    Command line tool (kubectl)

    Kubernetes 2024. Command line tool (kubectl). https://kubernetes.io/docs/ reference/kubectl/. [Online; accessed 29-October-2024]

    work page 2024

  22. [22]

    Production-grade container orchestration

    Kubernetes 2024. Production-grade container orchestration. https://kubernetes. io/. [Online; accessed 23-July-2024]

    work page 2024

  23. [23]

    Adam Lewis, Soumik Ghosh, and Nian Feng Tzeng. 2008. Run-time Energy Consumption Estimation Based on Workload in Server Systems.HotPower

    work page 2008

  24. [24]

    Hui li, Giuliano Casale, and Tariq Ellahi. 2010. SLA-driven planning and op- timization of enterprise applications.WOSP/SIPEW’10 - Proceedings of the 1st Joint WOSP/SIPEW International Conference on Performance Engineering, 117–128. doi:10.1145/1712605.1712625

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1145/1712605.1712625 2010

  25. [25]

    Youhuizi Li, Jiancheng Zhang, Congfeng Jiang, Jian Wan, and Zujie Ren. 2019. PINE: Optimizing Performance Isolation in Container Environments.IEEE Access 7 (2019), 30410–30422. doi:10.1109/ACCESS.2019.2900451

    work page doi:10.1109/access.2019.2900451 2019

  26. [26]

    Unai Lopez-Novoa. 2019. Exploring Performance and Energy Consumption Differences between Recent Intel Processors. In2019 IEEE SmartWorld, Ubiquitous Intelligence & Computing, Advanced & Trusted Computing, Scalable Computing & Communications, Cloud & Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI...

    work page doi:10.1109/smartworld-uic-atc-scalcom-iop-sci.2019.00088 2019

  27. [27]

    Xiaoyan Ma, Changgeng Zhang, Shurong Li, and Xinsheng Yang. 2021. A Data- Driven Based Energy Consumption Modeling for Heterogeneous Servers in Data Centers. In2021 IEEE 5th Conference on Energy Internet and Energy System Integration (EI2). 3109–3114. doi:10.1109/EI252483.2021.9713131

    work page doi:10.1109/ei252483.2021.9713131 2021

  28. [28]

    Joberto S. B. Martins, Tereza C. Carvalho, Rodrigo Moreira, Cristiano Bonato Both, Adnei Donatti, João H. Corrêa, José A. Suruagy, Sand L. Corrêa, Antonio J. G. Abelem, Moisés R. N. Ribeiro, José-marcos S. Nogueira, Luiz C. S. Magal- hães, Juliano Wickboldt, Tiago C. Ferreto, Ricardo Mello, Rafael Pasquini, Marcos Schwarz, Leobino N. Sampaio, Daniel F. Ma...

    work page doi:10.1109/access.2023.3292788 2023

  29. [29]

    Alexander Nowak, Tobias Binz, Frank Leymann, and Nicolas Urbach. 2013. Deter- mining Power Consumption of Business Processes and Their Activities to Enable Green Business Process Reengineering. In2013 17th IEEE International Enterprise Distributed Object Computing Conference. 259–266. doi:10.1109/EDOC.2013.36

    work page doi:10.1109/edoc.2013.36 2013

  30. [30]

    2023.Scaphandre

    Benoit Petit. 2023.Scaphandre. https://github.com/hubblo-org/scaphandre

    work page 2023

  31. [31]

    Tran, Shaolei Ren, Walid Saad, and Choong Seon Hong

    Chuan Pham, Nguyen H. Tran, Shaolei Ren, Walid Saad, and Choong Seon Hong

    work page

  32. [32]

    doi:10.1109/TSC.2017.2671867

    Traffic-Aware and Energy-Efficient vNF Placement for Service Chaining: Joint Sampling and Matching Approach.IEEE Transactions on Services Computing 13, 1 (2020), 172–185. doi:10.1109/TSC.2017.2671867

    work page doi:10.1109/tsc.2017.2671867 2020

  33. [33]

    Meikel Poess and Raghunath Othayoth Nambiar. 2008. Energy cost, the key challenge of today’s data centers: a power consumption analysis of TPC-C results. Proc. VLDB Endow.1, 2 (Aug. 2008), 1229–1240. doi:10.14778/1454159.1454162

    work page doi:10.14778/1454159.1454162 2008

  34. [34]

    Peter Rost, Ignacio Berberana, Andreas Maeder, Henning Paul, Vinay Suryaprakash, Matthew Valenti, Dirk Wübben, Armin Dekorsy, and Gerhard Fet- tweis. 2015. Benefits and challenges of virtualization in 5G radio access networks. IEEE Communications Magazine53, 12 (2015), 75–82. doi:10.1109/MCOM.2015. 7355588

    work page doi:10.1109/mcom.2015 2015

  35. [35]

    Cheng-Jen Tang and Miau-Ru Dai. 2011. Dynamic computing resource adjustment for enhancing energy efficiency of cloud service data centers. In2011 IEEE/SICE The Impact of Process Competition on Energy Consumption: Analysis and Modeling ADVANCE, March 25–28, 2026, Brazil International Symposium on System Integration (SII). 1159–1164. doi:10.1109/SII. 2011.6147613

    work page doi:10.1109/sii 2011

  36. [36]

    Amir Varasteh, Basavaraj Madiwalar, Amaury Van Bemten, Wolfgang Kellerer, and Carmen Mas-Machuca. 2021. Holu: Power-Aware and Delay-Constrained VNF Placement and Chaining.IEEE Transactions on Network and Service Manage- ment18, 2 (2021), 1524–1539. doi:10.1109/TNSM.2021.3055693

    work page doi:10.1109/tnsm.2021.3055693 2021

  37. [37]

    Zihao Wang, Lei Zhuang, Feijie Zhou, and Ruimin Wang. 2023. Energy Efficient VNF Placement Algorithm Using Reinforcement Learning in NFV-Enabled Net- work. In2023 IEEE International Conference on Control, Electronics and Computer Technology (ICCECT). 625–629. doi:10.1109/ICCECT57938.2023.10141342

    work page doi:10.1109/iccect57938.2023.10141342 2023

  38. [38]

    Mehul Warade, Kevin Lee, Chathurika Ranaweera, and Jean-Guy Schneider. 2023. Monitoring the Energy Consumption of Docker Containers. In2023 IEEE 47th Annual Computers, Software, and Applications Conference (COMPSAC). 1703–1710. doi:10.1109/COMPSAC57700.2023.00263

    work page doi:10.1109/compsac57700.2023.00263 2023

  39. [39]

    Junzo Watada, Arunava Roy, Ruturaj Kadikar, Hoang Pham, and Bing Xu. 2019. Emerging Trends, Techniques and Open Issues of Containerization: A Review. IEEE Access7 (2019), 152443–152472. doi:10.1109/ACCESS.2019.2945930

    work page doi:10.1109/access.2019.2945930 2019

  40. [40]

    Chi Xu, Ziyang Zhao, Haiyang Wang, Ryan Shea, and Jiangchuan Liu. 2017. Energy Efficiency of Cloud Virtual Machines: From Traffic Pattern and CPU Affinity Perspectives.IEEE Systems Journal11, 2 (2017), 835–845. doi:10.1109/ JSYST.2015.2429731

    work page arXiv 2017