ABSTRACT
The increasing usage of edge devices and stricter data privacy regulations motivate the use of federated learning (FL). At the same time, more and more stakeholders are concerned about the ecological impact of machine learning and its associate network traffic. The current research in FL does not investigate the impact of different network constraints and privacy-enhancing techniques, such as differential privacy, on the network traffic and energy consumption of the clients. Most experiments run either on virtual machines or on one machine with simulated clients. In such environments, it is challenging to measure each client’s network and energy usage. Therefore, we built our "Distributed Edge Device Testbed" (DEDT) and evaluate a convolutional neural network trained on the MNIST data set under different network constraints on DEDT, with differential privacy and with an increasing amount of participating clients. For each experiment, we quantify the network traffic, energy consumption, and training time. The results show the importance of experiments on physically separated nodes and the need to improve software-based power monitoring. The estimated energy consumption deviates by up to 35 % from the measured ones. The accuracy of the estimated network traffic depends on the monitored network interface and gives an error of 18 % for virtual machines in combination with monitoring the Ethernet interface. The training time also increases linearly with the number of participating clients.
- David Basin, Søren Debois, and Thomas Hildebrandt. 2018. On Purpose and by Necessity: Compliance Under the GDPR. In Financial Cryptography and Data Security, Sarah Meiklejohn and Kazue Sako (Eds.). Springer Berlin Heidelberg, Berlin, Heidelberg, 20–37.Google Scholar
- Keith Bonawitz, Hubert Eichner, Wolfgang Grieskamp, Dzmitry Huba, Alex Ingerman, Vladimir Ivanov, Chloe Kiddon, Jakub Konečný, Stefano Mazzocchi, H. Brendan McMahan, Timon Van Overveldt, David Petrou, Daniel Ramage, and Jason Roselander. 2019. Towards Federated Learning at Scale: System Design. arxiv:1902.01046 [cs.LG]Google Scholar
- Aurelien Bourdon, Adel Noureddine, Romain Rouvoy, and Lionel Seinturier. 2013. PowerAPI: A Software Library to Monitor the Energy Consumed at the Process-Level. ERCIM News 2013 (2013).Google Scholar
- Hafsa Bousbiat, Roumaysa Bousselidj, Yassine Himeur, Abbes Amira, Faycal Bensaali, Fodil Fadli, Wathiq Mansoor, and Wilfried Elmenreich. 2023. Crossing Roads of Federated Learning and Smart Grids: Overview, Challenges, and Perspectives. arxiv:2304.08602 [cs.LG]Google Scholar
- Christopher Briggs, Zhong Fan, and Peter Andras. 2021. Federated Learning for Short-term Residential Energy Demand Forecasting. arxiv:2105.13325 [cs.LG]Google Scholar
- Pietro Ferrara and Fausto Spoto. 2018. Static Analysis for GDPR Compliance. In ITASEC 2018 - Italian Conference on Cyber Security. CEUR Workshop Proceedings, Milan, Italy, 10. http://ceur-ws.org/Vol-2058/#paper-10Google Scholar
- Yunzhe Guo, Dan Wang, Arun Vishwanath, Cheng Xu, and Qi Li. 2020. Towards Federated Learning for HVAC Analytics: A Measurement Study. In Proceedings of the Eleventh ACM International Conference on Future Energy Systems (Virtual Event, Australia) (e-Energy ’20). Association for Computing Machinery, New York, NY, USA, 68–73. https://doi.org/10.1145/3396851.3397717Google ScholarDigital Library
- Tzu-Ming Harry Hsu, Hang Qi, and Matthew Brown. 2019. Measuring the Effects of Non-Identical Data Distribution for Federated Visual Classification. arxiv:1909.06335 [cs.LG]Google Scholar
- IEA. 2022. Data Centres and Data Transmission Networks. https://www.iea.org/reports/data-centres-and-data-transmission-networksGoogle Scholar
- Fabian Kaup, Philip Gottschling, and David Hausheer. 2014. PowerPi: Measuring and modeling the power consumption of the Raspberry Pi. In 39th Annual IEEE Conference on Local Computer Networks. IEEE, Edmonton, Canada, 236–243. https://doi.org/10.1109/LCN.2014.6925777Google ScholarCross Ref
- Kamar Kesrouani, Houssam Kanso, and Adel Noureddine. 2020. A Preliminary Study of the Energy Impact of Software in Raspberry Pi devices. In 2020 IEEE 29th International Conference on Enabling Technologies: Infrastructure for Collaborative Enterprises (WETICE). IEEE, Bayonne, France, 231–234. https://doi.org/10.1109/WETICE49692.2020.00052Google ScholarCross Ref
- A. S. Khatouni, M. Trevisan, and D. Giordano. 2019. Data-Driven Emulation of Mobile Access Networks. In 2019 15th International Conference on Network and Service Management (CNSM). IEEE, 1515 South Park Street, Halifax, Nova Scotia, Canada, 1–6. https://doi.org/10.23919/CNSM46954.2019.9012691Google Scholar
- Sangyoon Lee and Dae-Hyun Choi. 2022. Federated Reinforcement Learning for Energy Management of Multiple Smart Homes With Distributed Energy Resources. IEEE Transactions on Industrial Informatics 18, 1 (2022), 488–497. https://doi.org/10.1109/TII.2020.3035451Google ScholarCross Ref
- Tian Li, Maziar Sanjabi, Ahmad Beirami, and Virginia Smith. 2020. Fair Resource Allocation in Federated Learning. arxiv:1905.10497 [cs.LG]Google Scholar
- Y. Liu, J. J. Q. Yu, J. Kang, D. Niyato, and S. Zhang. 2020. Privacy-Preserving Traffic Flow Prediction: A Federated Learning Approach. IEEE Internet of Things Journal 7, 8 (2020), 7751–7763. https://doi.org/10.1109/JIOT.2020.2991401Google ScholarCross Ref
- H. Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y Arcas. 2023. Communication-Efficient Learning of Deep Networks from Decentralized Data. arxiv:1602.05629 [cs.LG]Google Scholar
- D. Moss. 2014. Modelling the network performance of DSL connections using netem.Google Scholar
- State of California Department of Justice. 2018. California Consumer Privacy Act of 2018 [1798.100 - 1798.199.100].Google Scholar
- D. E. O’Leary. 2013. Artificial Intelligence and Big Data. IEEE Intelligent Systems 28, 2 (2013), 96–99. https://doi.org/10.1109/MIS.2013.39Google ScholarDigital Library
- Aman Priyanshu, Rakshit Naidu, Fatemehsadat Mireshghallah, and Mohammad Malekzadeh. 2021. Efficient Hyperparameter Optimization for Differentially Private Deep Learning.Google Scholar
- Xinchi Qiu, Titouan Parcollet, Daniel J. Beutel, Taner Topal, Akhil Mathur, and Nicholas D. Lane. 2020. Can Federated Learning Save The Planet?https://doi.org/10.48550/ARXIV.2010.06537Google Scholar
- Xinchi Qiu, Titouan Parcollet, Javier Fernandez-Marques, Pedro Porto Buarque de Gusmao, Yan Gao, Daniel J. Beutel, Taner Topal, Akhil Mathur, and Nicholas D. Lane. 2022. A first look into the carbon footprint of federated learning. arxiv:2102.07627 [cs.LG]Google Scholar
- Xidi Qu, Shengling Wang, Qin Hu, and Xiuzhen Cheng. 2021. Proof of Federated Learning: A Novel Energy-Recycling Consensus Algorithm. IEEE Transactions on Parallel and Distributed Systems 32, 8 (2021), 2074–2085. https://doi.org/10.1109/TPDS.2021.3056773Google ScholarDigital Library
- Sashank Reddi, Zachary Charles, Manzil Zaheer, Zachary Garrett, Keith Rush, Jakub Konečný, Sanjiv Kumar, and H. Brendan McMahan. 2021. Adaptive Federated Optimization. arxiv:2003.00295 [cs.LG]Google Scholar
- Y. Roh, G. Heo, and S. E. Whang. 2021. A Survey on Data Collection for Machine Learning: A Big Data - AI Integration Perspective. IEEE Transactions on Knowledge and Data Engineering 33, 4 (2021), 1328–1347. https://doi.org/10.1109/TKDE.2019.2946162Google ScholarCross Ref
- Yuris Mulya Saputra, Dinh Thai Hoang, Diep N. Nguyen, Eryk Dutkiewicz, Markus Dominik Mueck, and Srikathyayani Srikanteswara. 2019. Energy Demand Prediction with Federated Learning for Electric Vehicle Networks. In 2019 IEEE Global Communications Conference (GLOBECOM). IEEE, Waikoloa, Hawaii, USA, 1–6. https://doi.org/10.1109/GLOBECOM38437.2019.9013587Google ScholarDigital Library
- Personal Data Protection Commission Singapore. 2014. Personal Data Protection Act.Google Scholar
- A. Taïk and S. Cherkaoui. 2020. Electrical Load Forecasting Using Edge Computing and Federated Learning. In ICC 2020 - 2020 IEEE International Conference on Communications (ICC). IEEE, Virtual, 1–6. https://doi.org/10.1109/ICC40277.2020.9148937Google Scholar
- Andrew Trask, Emma Bluemke, Ben Garfinkel, Claudia Ghezzou Cuervas-Mons, and Allan Dafoe. 2020. Beyond Privacy Trade-offs with Structured Transparency. https://doi.org/10.48550/ARXIV.2012.08347Google Scholar
- Ye Lin Tun, Kyi Thar, Chu Myaet Thwal, and Choong Seon Hong. 2021. Federated Learning based Energy Demand Prediction with Clustered Aggregation. In 2021 IEEE International Conference on Big Data and Smart Computing (BigComp). IEEE, Jeju Island, Korea, 164–167. https://doi.org/10.1109/BigComp51126.2021.00039Google ScholarCross Ref
- European Union. 2016. REGULATION (EU) 2016/679 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation).Google Scholar
- Haijin Wang, Caomingzhe Si, and Junhua Zhao. 2021. A Federated Learning Framework for Non-Intrusive Load Monitoring. arxiv:2104.01618 [eess.SP]Google Scholar
- Haijin Wang, Caomingzhe Si, Junhua Zhao, Guolong Liu, and Fushuan Wen. 2021. Fed-NILM: A Federated Learning-based Non-Intrusive Load Monitoring Method for Privacy-Protection. arxiv:2105.11085 [cs.LG]Google Scholar
- Sihua Wang, Mingzhe Chen, Walid Saad, and Changchuan Yin. 2020. Federated Learning for Energy-Efficient Task Computing in Wireless Networks. In ICC 2020 - 2020 IEEE International Conference on Communications (ICC). IEEE, Virtual, 1–6. https://doi.org/10.1109/ICC40277.2020.9148625Google Scholar
- Zhaohui Yang, Mingzhe Chen, Walid Saad, Choong Seon Hong, and Mohammad Shikh-Bahaei. 2021. Energy Efficient Federated Learning Over Wireless Communication Networks. IEEE Transactions on Wireless Communications 20, 3 (2021), 1935–1949. https://doi.org/10.1109/TWC.2020.3037554Google ScholarDigital Library
- Ashkan Yousefpour, Igor Shilov, Alexandre Sablayrolles, Davide Testuggine, Karthik Prasad, Mani Malek, John Nguyen, Sayan Ghosh, Akash Bharadwaj, Jessica Zhao, Graham Cormode, and Ilya Mironov. 2021. Opacus: User-Friendly Differential Privacy Library in PyTorch. arXiv preprint arXiv:2109.12298 (2021), 18.Google Scholar
- Xinyu Zhou, Jun Zhao, Huimei Han, and Claude Guet. 2022. Joint Optimization of Energy Consumption and Completion Time in Federated Learning. In 2022 IEEE 42nd International Conference on Distributed Computing Systems (ICDCS). IEEE, Bologna, Italy, 1005–1017. https://doi.org/10.1109/ICDCS54860.2022.00101Google Scholar
- X. Zhu, J. Wang, Z. Hong, T. Xia, and J. Xiao. 2019. Federated Learning of Unsegmented Chinese Text Recognition Model. In 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI). IEEE, Portland, OR, USA, 1341–1345. https://doi.org/10.1109/ICTAI.2019.00186Google Scholar
Index Terms
- Energy vs Privacy: Estimating the Ecological Impact of Federated Learning
Recommendations
Towards energy-aware federated learning on battery-powered clients
FedEdge '22: Proceedings of the 1st ACM Workshop on Data Privacy and Federated Learning Technologies for Mobile Edge NetworkFederated learning (FL) is a newly emerged branch of AI that facilitates edge devices to collaboratively train a global machine learning model without centralizing data and with privacy by default. However, despite the remarkable advancement, this ...
Estimating energy impact of software releases and deployment strategies: the KPMG case study
ESEM '17: Proceedings of the 11th ACM/IEEE International Symposium on Empirical Software Engineering and MeasurementBackground. Often motivated by optimization objectives, software products are characterized by different subsequent releases and deployed through different strategies. The impact of these two aspects of software on energy consumption has still to be ...
Client-centered energy savings for concurrent HTTP connections
NOSSDAV '04: Proceedings of the 14th international workshop on Network and operating systems support for digital audio and videoIn mobile devices, the wireless network interface card (WNIC) consumes a significant portion of overall system energy. One way to reduce energy consumed by a WNIC is to transition it to a lower-power sleep mode when data is not being received or ...
Comments