Bachelor Thesis: Reducing Networking Overhead for High Performance Distributed Data Processing

EST-APIs are an industry standard for a big part of interprocess communication. They have a clear and easy to implement paradigm and major support for tooling and libraries. However for high performance distributed workloads, the overhead of tcp handshakes, SSL handshakes, HTTP writing and parsing and the repeated opening and closing for connections between the same machines can be a real bottleneck, especially for slower or older machines in a network.

This thesis will focus on the implementation of a REST-API and a socket-based API and compare the overhead as well as the best-case and real-world performance impacts of such approaches. To accomplish this, the same workload will be implemented once as a REST-API client and once as a client that communicates over websockets. The server will be an already existing distributed monitoring solution that provides both endpoints.

(In collaboration with Wuerth Phoenix, Bolzano, Italy)

Contact: Ruth.Breu@uibk.ac.at