Sign up for our newsletter and get the latest HPC news and analysis.

Grid as a managed service

Grid Blog points to an article about grid computing as a managed service offered by Sun, IBM, HP, and others.

The article is fair and balanced in addressing most of these issues and others (standards and OGSA, security, practical obstacles and risks). It’s a good read.

I’m on record as thinking that these concepts matter a lot, but not necessarily in the way the we are talking about now , and I’m also on record as a contrarian voice during most of the Grid (capital G) hype in recent years.

The article defines grid computing as

The ability to perform high-throughput computing via the Internet by leveraging distributed computers linked by networks; on-demand provisioning; resource sharing among organizations; making use of unused computing resources and the use of open standards.

I think there might be some traction in resources sharing brokered by an intermediate organization; a marketplace for cycles (we also talked about this in our CPUShare coverage). I also think there’s a good argument for satisfying workloads that aren’t tightly coupled using this type of approach.

The concept is important at another level. You can think of grid services (lower case, forget about a particular solution) as the next layer of abstraction after MPI. Write your apps for a grid environment and you gain the ability to move and adapt as resources change.

Right now this costs you performance, and there aren’t many places to run anyway, so portability isn’t necessary. But separating the presentation from the data is almost always a good thing, and a sign of technology maturity.

See the full article for more details.

Resource Links: