By Topic

How to waste 2/3 of the throughput of a switch: a tight characterization of load balancing algorithms that do not split

Sign In

Cookies must be enabled to login.After enabling cookies , please use refresh or reload or ctrl+f5 on the browser for the login options.

Formats Non-Member Member
$33 $13
Learn how you can qualify for the best price for this item!
Become an IEEE Member or Subscribe to
IEEE Xplore for exclusive pricing!
close button

puzzle piece

IEEE membership options for an individual and IEEE Xplore subscriptions for an organization offer the most affordable access to essential journal articles, conference papers, standards, eBooks, and eLearning courses.

Learn more about:

IEEE membership

IEEE Xplore subscriptions

1 Author(s)
Saad Mneimneh ; Visiting Professor, Computer Science, Hunter College of CUNY, 695 Park Avenue, New York, NY 10021. Tel: (214)707-1075, Fax: (212)772-5219, Email:

Load balancing has recently acquired increased interest among researchers in the switching community. The premise is to replace the single switch, running at a speed proportional to the number n of input and output ports (and to the line rate), by a load balanced switch consisting of k switches, each running at 1/k the speed. Indeed, with such an architecture, and for some classes of traffic patterns, simply spreading the traffic among the k switches achieves 100% throughput (thus the term load balancing). However, reordering among packets of the same flow may occur, which becomes a major concern. One way of avoiding this problem is to use a load balancing algorithm that does not split, i.e. that routes all packets of a given flow through exactly one of the k switches, without re-routing existing Hows. We shall call such a load balancing algorithm a non-splitting algorithm. While non-splitting algorithms definitely preserve the order of packets, it is intuitively well understood that these algorithms waste throughput. The focus of this paper is to exactly characterize this waste. We prove an asymptotic tight upper bound of 1/3 on the throughput of any non-splitting algorithm when n >> k.

Published in:

2007 Workshop on High Performance Switching and Routing

Date of Conference:

May 30 2007-June 1 2007