Abstract: Many distributed algorithms are limited by the performance of the slowest machine in the system.  For instance, if an algorithm requires all processes to reach a barrier before it can continue, then the progress of the whole is determined by the last process to reach that barrier.  One possible solution to this problem would be to reduce the load on the slower machines by giving some of their work to the faster ones.  In practise, however, this is very difficult to implement.  Deciding how to distribute the load often requires global knowledge, direct supervision by an omniscient machine, and heavy calculation.  Furthermore, if we do decide to take a job from a slow machine and give it to a fast machine, we must first guarantee that the latter machine has all of the necessary resources (Data, information, communication ability, etc.) to complete the job.

This paper examines a specific setting where workload redistribution is desirable, describes a simple solution, and analyses its effectiveness.  The setting is a large parallel database in which 100 client machines work concurrently to read a file that is stored across 100 servers.  The solution requires only local communication (each server must communicate with two clients, and each client with two servers), and only very minor calculations.  It changes dynamically with time and can be shown to converge on the best possible solution.  We present it here in several variations, and discuss the strengths and weaknesses of each.
 

Available as Maple Worksheet   (This is more fun and interactive, but not easily printable.  Open it in xmaple if you get a chance.)

Also available as: Postscript