Sweet spots in stream computing: understanding and exploiting the interaction between operating system and stream computing settings for improved energy efficiency and performance
OA Version
Citation
Abstract
Data center energy consumption is a growing problem today. Online data-intensive workloads and streams of data from edge devices is a non-negligible part of this problem as they demand low tail-latencies. An example of data center software that handles data streams is Flink, a stream processing software. Stream processing software is, by it nature, network intensive. Han Dong’s dissertation work, “A data driven study of Operating System Energy-performance trade-offs towards system selfoptimization." focuses on network-intensive workloads. The following thesis extends the aforementioned dissertation and aims to find, understand, and exploit the sweet spots that exist while running the stream processing software Flink on the operating system Linux. Both these works are related to the concept of pacing for improved performance and energy efficiency which has been discussed in prior literature.
Description
2024