[Beowulf] How Can Microsoft's HPC Server Succeed?

[Greg Lindahl]

On Sat, Apr 19, 2008 at 12:26:28PM -0700, Donald Becker wrote:

> And it's why I consider full installation to be unworkable 
> for large clusters, especially when re-installation is considered to be 
> part of cluster administration.

There seem to be 3 main opinions in this area of cluster admin:

1) Install little or nothing on the nodes; reboot all the time

2) Heavy install on the nodes; re-image to ensure consistency

3) Heavy install on the nodes; other mechanism to ensure consistency

All of these have their pros and cons. You are correct that (2) needs
a fast re-image, since you're going to be doing it fairly frequently.
But (3) will only re-image once a year or two.

Some people in (2) reboot&reimage any time they change a single
rpm. That's a recipe for annoying your users, unless you have the
ability to do a rolling-reboot between jobs.

-- greg