Optimizing Worker Communication in Serverless Environments with Worker Co-Location

Identifier:  TFG:8291

Handle: https://hdl.handle.net/20.500.11797/TFG8291

Authors:  Klymonchuk, Stepan

Abstract:
The thesis introduces Burst Computing, a serverless solution for burst-parallel jobs, enhancing Function-as-a-Service (FaaS) by addressing its inefficiencies in handling simultaneous, coordinated function invocations. Central to this is the Burst Communication Middleware (BCM), which improves indirect communication among workers by leveraging group awareness and locality for efficient local communications. BCM supports message passing and group collectives with zero-copy data transfers, significantly reducing communication overheads. The thesis details BCM's design, architecture, and implementation, showing that it outperforms standard FaaS solutions in managing burst-parallel workloads in cloud computing.