7 Things You Should Know about Grid Computing

Created by Elisa Coghlan (EDUCAUSE) on January 30, 2006

EDUCAUSE Learning InitiativeA snapshot of what grid computing is and its value to teaching and learning is offered in a new ELI brief, 7 Things You Should Know About Grid Computing.

Submitted by hes8 on Sun, 2006/02/05 - 7:29pm.
Some links to resources can be found on my grid resource page. It's not as up to date as I'd like, but I hope it is still of use. Comments and suggestions for improvement will be appreciated.
Submitted by hes8 on Sun, 2006/02/05 - 6:38pm.
A nice non-technical overview - but perhaps a misleading one. Misleading, that is, until one reaches #5 of the 7 Things.

Sections #1-4 are written in present tense, not future tense. This is a common technique used in marketing vaporware. -) As an advocate of developing grid computing, I agree that #1-4 are all good ideas and possible results of the developing grid computing environment. But it is not the way most people are doing most of their computing today.

Section #5 very briefly covers many of the "downsides", but leaves out a few reasonably important ones. E.g. "latency" in communication between parallel processors. Parallel processing is of central importance in modern scientific computing, and in many instances processors have to share or exchange data frequently - many times per second. This has given rise to a small industry of "fast interconnect" solutions - such as those from Myrinet or InfiniBand - which have a latency described in microseconds. Shared memory parallel systems have latency in nanoseconds. The general Internet has a latency typically in tens or hundreds of milliseconds - which is not competitive. Amdahl's law estimates the extreme inefficiency of trying to use a parallel system distributed widely over the Internet when a latency of microseconds or less is appropriate.

A more thorough consideration of the downsides helps understand the serious obstacles which have to be overcome and why the development and adoption of grid middleware has taken time. I'd like to avoid irrational exhuberance, and the resulting "trough of disillusion."

By the way, the "grid" described in the opening Scenario, is not what most of the grid community considers to be a grid. What is described is currently in use, and can be extremely valuable, this approach has very significant limitations. In the Scenario, it is limited to one administrative domain.

While I'm strongly in favor of developing grid middleware and adopting it, I suggest that we'd be better off with a more balanced and realistic presentation.