Cloud application adaptation using machine learning
Cloud applications operate in highly dynamic environments, characterised by fluctuating workloads, frequent failures, and varying levels of performance supported by underlying infrastructures [1]. Coping with such environments requires application adaptation, which may involve increasing/decreasing the number of instances of application services (e.g., web servers, databases), resizing instances (e.g., increasing the CPU allocation of the database) or even migrating instances across clouds. However, deciding when and how to adapt cloud applications remains challenging. Modern cloud platforms typically support only threshold-based adaptation policies (e.g., Amazon Auto Scaling [2]), leaving the burden of setting appropriate thresholds to application owners.
The aim of this internship is to investigate cloud application adaptation using machine learning, thus enabling automatically learning appropriate adaptation policies based on experience. The internship will be divided in two main parts. In the first part, the intern will study the state of the art in using machine learning techniques for adapting cloud systems [3,4,5]. In the second part, the intern will propose an adaptation solution based on a selected technique and evaluate the solution using multi-tier web applications and high-performance computing applications deployed across multiple clouds. The solution will be integrated in PaaSage, an open-source platform that supports the design and deployment of cloud applications independently of underlying cloud infrastructures [6].
Contact: Nikos.Parlavantzas@irisa.fr, Christine.Morin@inria.fr
References
[1] Stefania Costache, Djawida Dib, Nikos Parlavantzas, Christine Morin, “Resource management in cloud platform as a service systems: Analysis and opportunities”, In Journal of Systems and Software, Volume 132, 2017, Pages 98-118
[2] http://aws.amazon.com/autoscaling/
[3] Tania Lorido-Botran, Jose Miguel-Alonso, Jose A. Lozano, “A Review of Auto-scaling Techniques for Elastic Applications in Cloud Environments”, Journal of Grid Computing, December 2014, Volume 12, Issue 4, pp 559-592
[4] Muhammad Wajahat, Alexei Karve, Andrzej Kochut, Anshul Gandhi, “Using machine learning for black-box autoscaling,” 2016 Seventh International Green and Sustainable Computing Conference (IGSC), Hangzhou, 2016, pp. 1-8
[5] Hongzi Mao, Mohammad Alizadeh, Ishai Menache, Srikanth Kandula, “Resource Management with Deep Reinforcement Learning”, In Proceedings of the 15th ACM Workshop on Hot Topics in Networks (HotNets ’16). ACM, New York, NY, USA, 50-56
[6] http://www.paasage.eu
