in the datacenter
Most of the grid work we see is happening in the secure/controlled datacenter, not "on unused cycles of PCs running a screensaver." These servers are harnessed for their compute capabilities, and relatively effective because of the high bandwidth available for internode communications. (Theoretically, the compute:bandwidth requirements ratio determines the scalable effectiveness of a grid.)
We have a handful of customers already using grids for real-world problems. Often they use a product like "Platform" or "Data Synapse" (etc.) to manage the grid, and a product like ours (Tangosol Coherence) to manage application state (data) and grid execution across the nodes.
Depending on the application (obviously) they can see significant performance improvements and/or cost savings, depending what it is they are replacing or comparing to. Grid computing doesn't have that many applications tailor-made to show its benefits, but the few that we have been involved with have been beautiful examples of how grid computing can provide a "black box" (high reliability / availability) that spits out end results much more quickly (10x, 100x, sometimes more) than the massive boxen (e.g. 32 CPUs, 64 CPUs) that they are replacing, and do so at a fraction of the cost.
Peace,
Cameron Purdy
Tangosol, Inc.
Posted by: cpurdy on September 15, 2003 at 09:54 AM