Micro load balancing with delayed queue lengths

DRILL is a micro load balancing algorithm designed to efficiently utilize the path redundancy in modern data centers. It uses egress port queue lengths to make fast packet routing decisions to reduce upstream congestion and queueing delays. However, high performance switches with multiple forwarding engines making routing decisions in parallel, do not have direct access to these queue lengths. We explore and evaluate different ways of obtaining this information in data center settings, specifically using incoming traffic and specially generated update packets to piggyback this information. We find that staleness of this data does not have a huge impact on flow completion times compared to DRILL (6% increase) and still achieves a considerable advantage over ECMP (28% decrease).

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2019

Author dc:creator

• Tariq, Fatima

Contributors dc:contributor

• Godfrey, Philip B.

• Load balancing
• Data centers

Statement dc:rights

• Copyright 2019 Fatima Tariq

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/105070

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/105070