Balancing Performance and Sustainability for FaaS in the Fog

Publié le
Equipe
Date de début de thèse (si connue)
October 2023
Lieu
Rennes
Unité de recherche
IRISA - UMR 6074
Description du sujet de la thèse

Fog computing is an extension of the traditional cloud computing model in which compute, storage, and network capabilities are distributed closer to users [1]. Fog computing is motivated by the need to support new Internet of Things (IoT) applications, such as smart cities and AI-enabled surveillance systems, that have strict demands for bandwidth and low-latency computation. A compelling programming model for developing such applications is the Function-as-a-Service (FaaS) model [2], the core element of serverless computing. FaaS supports easy movement of functions along the cloud-to-thing continuum, allowing optimizing for diverse factors, such as latency and energy efficiency. Moreover, FaaS supports fine-grained, short-lived resource allocations, enabling increased infrastructure utilization.

Managing FaaS applications running in fog environments presents significant challenges [3]. First, fog resources are geo-distributed, heterogeneous (e.g., sensors, mobile devices, micro data centers) with diverse power sources (e.g., battery-powered, grid-powered) and subject to unpredictable changes (e.g., fog nodes joining, failing), making it difficult to make effective management decisions. Second, FaaS workloads are highly dynamic and have a high deployment density due to the short duration and small size of individual functions. This exacerbates interferences between workloads [4], making it difficult to predict the performance and energy impact of management actions, and complicating decision making. Third, FaaS platforms must carefully balance the Quality of Service (QoS) requirements of applications with the need to reduce the energy consumption and carbon footprint of fog infrastructures, which is becoming increasingly important as environmental concerns continue to rise [5].

This thesis will explore QoS-driven, energy-aware management of FaaS applications in fog environments. Specifically, the goal is to develop an automated management solution that can ensure QoS requirements for FaaS applications while also reducing energy and carbon usage. This solution will have the ability to evaluate the potential costs and benefits of different management actions [6] by predicting their impact on performance and energy consumption [7]. To achieve this, the solution will use performance interference analysis techniques that were initially developed for High Performance Computing (HPC) applications [8,9] and adapt them to the specific characteristics of FaaS workloads [4]. Energy and carbon will be considered as first-class resources, and management will be guided by QoS requirements as well as energy consumption requirements [10], formalized in Service Level Agreements (SLAs).

The developed techniques and management tools will be deployed and tested in an environmental monitoring project on the Beaulieu campus, in partnership with researchers from the Observatoire des Sciences de l'Univers de Rennes (OSUR). The project uses a fog infrastructure made up of sensors, actuators, resource-constrained edge nodes, and cloud nodes. This infrastructure supports hosting various FaaS workloads, such as applications for monitoring wildlife and water and air quality. The tools will build on the Kubernetes resource orchestration system and an existing open-source FaaS platform, such as OpenFaaS.

Bibliographie

[1] R. Mahmud, R. Kotagiri, and R. Buyya. “Fog Computing: A Taxonomy, Survey and Future Directions”. In: Internet of Everything: Algorithms, Methodologies, Technologies and Perspectives. Ed. by B. Di Martino, K.-C. Li, L. T. Yang, and A. Esposito. Springer Singapore, Singapore, 2018, pp. 103–130. isbn: 978-981-10-5861-5. doi: 10.1007/978- 981-10-5861-5_5

[2] J. Schleier-Smith, V. Sreekanti, A. Khandelwal, J. Carreira, N. J. Yadwadkar, R. A. Popa, J. E. Gonzalez, I. Stoica, and D. A. Patterson, “What serverless computing is and should become: The next phase of cloud computing,” Commun. ACM, vol. 64, no. 5, p. 76–84, Apr. 2021. doi: 10.1145/3406011

[3] R. Xie, Q. Tang, S. Qiao, H. Zhu, F. R. Yu and T. Huang, “When Serverless Computing Meets Edge Computing: Architecture, Challenges, and Open Issues,” in IEEE Wireless Communications, vol. 28, no. 5, pp. 126-133, October 2021, doi: 10.1109/MWC.001.2000466

[4] L. Zhao, Y. Yang, Y. Li, X. Zhou, and K. Li. 2021, “Understanding, predicting and scheduling serverless workloads under partial interference”, In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC '21). ACM, New York, NY, USA, Article 22, 1–15. doi: 10.1145/3458817.3476215

[5] P. Patros, J. Spillner, A. V. Papadopoulos, B. Varghese, O. Rana and S. Dustdar, “Toward Sustainable Serverless Computing,” in IEEE Internet Computing, vol. 25, no. 6, pp. 42-50, 1 Nov.-Dec. 2021, doi: 10.1109/MIC.2021.3093105.

[6] N. Parlavantzas, L. M. Pham, A. Sinha and C. Morin, “Cost-Effective Reconfiguration for Multi-Cloud Applications,”, 26th Euromicro International Conference on Parallel, Distributed and Network-based Processing (PDP), Cambridge, UK, 2018, pp. 521-528, doi: 10.1109/PDP2018.2018.00088.

[7] J. Flores-Contreras, H.A., Duran-Limon, A. Chavoya, et al. “Performance prediction of parallel applications: a systematic literature review”. J Supercomput 77, 4014–4055 (2021). doi: 10.1007/s11227-020-03417-5

[8] J. Weinberg and A. Snavely. “User-guided symbiotic space-sharing of real workloads”, In Proceedings of the 20th annual international conference on Supercomputing (ICS '06). Association for Computing Machinery, New York, NY, USA, 345–352. doi: 10.1145/1183401.1183450

[9] D. Yokoyama, B. Schulze, H. Kloh, M. Bandini, and V. Rebello, “Affinity aware scheduling model of cluster nodes in private clouds”, J. Netw. Comput. Appl. 95, C (October 2017), 94–104. doi: 10.1016/j.jnca.2017.08.001

[10] T. E. Anderson, A. Belay, M. Chowdhury, A. Cidon, and I. Zhang, “Treehouse: A Case For Carbon-Aware Datacenter Software,” ArXiv preprint arXiv:2201.02120, doi: 10.48550/arXiv.2201.02120

 

 

Liste des encadrants et encadrantes de thèse

Nom, Prénom
Parlavantzas, Nikos
Type d'encadrement
Directeur.trice de thèse
Unité de recherche
IRISA
Equipe

Nom, Prénom
Duran-Limon, Hector
Type d'encadrement
Co-encadrant.e
Unité de recherche
University of Guadalajara, Mexic
Contact·s
Mots-clés
FaaS, serverless, fog, edge, performance interference, energy, sustainability