ML-based Incast Performance Optimization in Software-Defined Data Centers

Abstract

Traffic optimization is fundamental to achieve both great application performance and resource efficiency in data centers with heterogeneous workloads, including incast. However, general performance models, providing insights on how various factors affect a certain performance metric used in the network optimization process, are missing. For the special case of incast, the existing models are analytical models, either tightly coupled with a particular protocol version or specific to certain empirical data. This paper proposes an SDN-enabled machine-learningbased optimization framework for incast performance optimization in data center networks that leverages learning-based performance modeling. Evaluations based on intensive NS-3 simulations show that we can achieve accurate performance predictions that enable finding the efficient switch buffer space to achieve optimal incast completion time in different configurations. We expect this framework to be a building block for autonomous data center network management.

Publication
2021 IEEE 22nd International Conference on High Performance Switching and Routing (HPSR)

Add the full text or supplementary notes for the publication here using Markdown formatting.