A Reinforcement Learning Approach for Reconfiguring Data Stream Processing Applications on Edge Computing

Alexandre da Silva Veith - University of Toronto

Nov. 4, 2021, 4:30 p.m. - Nov. 4, 2021, 5:30 p.m.

McConnell 103

Hosted by: Oana Balmau

Abstract: There is increasing demand for handling massive amounts of data in a timely manner via Data Stream Processing (DSP). A DSP application is often structured as a directed graph whose vertices are operators that perform transformations over the incoming data and edges representing the data streams between operators. DSP applications are traditionally deployed on the Cloud in order to explore the virtually unlimited number of resources. Edge computing has emerged as a suitable paradigm for executing parts of DSP applications by offloading certain operators from the Cloud and placing them close to where the data is generated, hence minimising the overall time required to process data events (i.e., the end-to-end latency). The operator reconfiguration consists of changing the initial placement by reassigning operators to different devices given target performance metrics. In this work, we model the operator reconfiguration as a Reinforcement Learning (RL) problem and define a multi-objective reward considering metrics regarding operator reconfiguration, and infrastructure and application improvement. Experimental results show that reconfiguration algorithms that minimise only end-to-end processing latency can have a substantial impact on WAN traffic and communication cost. The results also demonstrate that when reconfiguring operators, RL algorithms improve by over 50% the performance of the initial placement provided by state-of-the-art approaches.


Bio: Alexandre da Silva Veith is a postdoctoral researcher at the University of Toronto, Canada. He holds a Ph.D. in Computer Science from the Ecole Normale Superieure (ENS) of Lyon and the University of Lyon, France. His wide-ranging areas of interest include distributed systems, machine learning, reinforcement learning, federated learning, data stream processing systems, optimization problems queueing theory, fog/edge computing, big data analytics, and complex issues emerging from the Internet of Things. His research on edge computing systems has been widely published in prestigious journals and conferences in the areas of networking and distributed systems including but not limited to: the Journal of Network and Computer Applications, as one of their most cited papers, IEEE Transactions on Cloud Computing, as well as the International Conference on Service-Oriented Computing, the International Conference on Parallel Processing and IEEE/ACM International Symposium in Cluster, Cloud, and Grid Computing.


Please note that this is a hybrid seminar. To attend in person, please reserve a spot in advance by following this link (@mcgill.ca account login is required). You may also attend virtually through Zoom: https://mcgill.zoom.us/j/84265536263