Data center traffic scheduling with hot-cold link detection capabilities

Author(s)

Publication date

2018

Series/Report no

RACS '18 Proceedings of the 2018 Conference on Research in Adaptive and Convergent Systems;

Publisher

ACM

Document type

Abstract

Software-Defined Networking (SDN) has been one of the most discussed areas in computer networking over the last years. The field has raised an extensive amount of research, and led to a transformation of traditional network architectures. The architecture of SDN enables the separation of the control and data planes and centralizes the network intelligence. Today's data center networks are clusters of thousands of machines. The most used routing protocol in Data centers is Equal-Cost Multi-Path Protocol (ECMP) which relies on a per-flow static hashing that is known to cause bandwidth loss because of long term collisions. In this paper, a traffic engineering approach built on the concept of SDN is presented that aims to enhance the least-loaded link routing mechanism with intelligent monitoring capabilities. In this perspective, we introduce Hot and Cold link detection (HCLD) mechanism. Our HCLD permits to dynamically re-route heavy flows from heavily utilized links (Hot links) while attracting more flows to lowly utilized links (Cold links). Comprehensive experimental results show that the devised flow scheduling solution outperforms the widely used ECMP. Results also demonstrate that dynamic monitoring of traffic statistics could be used to better utilize the total available bandwidth of the network in a reactive manner.

Keywords

Version

acceptedVersion

Permanent URL (for citation purposes)

  • https://hdl.handle.net/10642/7437