Exploring the Challenges of Performance, High Availability, and Privacy in Distributed Systems
Invited talk: Guillaume Rosinosky presented "Exploring the Challenges of Performance, High Availability, and Privacy in Distributed Systems" at 4A312 the 13/1/2023 at 14h30.
Abstract
Nowadays, distributed systems support applications we use in every aspect of our lives. These systems have to be able to withstand a large number of users with sufficient performance and minimal cost, and support potential crashes while permitting users to keep their privacy. In this talk, I will present my past and current research on four key topics related the subjects of performance, high availability, and privacy of distributed systems: BPM systems elasticity, highly available stream processing engines, privacy-enabling proxies for recommendation systems, and usage of service meshes to support services sharding. For the first topic, we will discuss our efforts to improve the elasticity of BPM systems, which involves tooling, models and heuristics for resource cost optimization with constraints on QoS. We will then focus on our work on highly available stream processing engines, which aims to ensure the reliability and uptime of stream processing applications. We will follow with our development of a privacy-enabling proxy for recommendation systems, which allows users to receive personalized recommendations while preserving their privacy in a fast and non-destructive way by using SGX enclaves. Finally, we will discuss our current line of research concerning service meshes: the transparent coupling of proxies with a control plane permitting to enact transversal features on service communications without codebase modification. With our approach, we add the possibility to enact non-invasive sharding at runtime on support services such as databases or pub-sub systems. This final approach can be extended to ensure scalable privacy-enabling resilient systems.
Bio
Guillaume Rosinosky is a post-doc in the Cloud and Large Scale computing group at UCLouvain, Belgium. He obtained his PhD in 2019 from Université de Lorraine for his work on the elasticity of BPM engines in the cloud. Before that, he worked for a period of 10 years as a developer and lead developer in the industry. His research interests include distributed systems, optimization, and machine learning, and he is widely interested in the high availability, privacy, and elasticity of large scale systems.
