[Beowulf] distributions

[Greg M. Kurtzer]

On Mon, Feb 06, 2006 at 04:07:50PM -0800, Donald Becker wrote:
> Nor does "ramdisk root" give you the magic.  A ramdisk root is part of how 
> we implement the architecture, especially the part about not requiring local 
> storage or network file systems to work.  (Philosophy: You mount file 
> systems as needed for application data, not for the underlying system.)  
> But to be fast, efficient and effective the ramdisk can't just be a 
> stripped-down full distribution.  You need a small 'init' system and 

I think you may want to rephrase that (as it is a matter of opinion,
and considered by many to be efficient *and* effective).

> a dynamic, version-based caching mechanism.  Otherwise you end up with 
> lots of wasted memory, version skew and still have a crippled compute node 
> environment.

I have the same comment as above.

If memory is a concern, I have known people to run Warewulf based
clusters using less then then 30MB of RAM for the OS, and when nodes
are booted, there is no chance for version skew. There are simple ways
to make sure that the VNFS is consistent among all nodes.

There are also ways to utilize the RAM disk efficiently and via the use
of a network file system (simple NFS or even parallel file systems) to
have nodes with many different roles (eg. compute, web, development and
compile, interactive, DB, IO, etc...). A solution like this naively
emphasizes the flexibility inherent to Unix/Linux without getting in
the way and allows the system administrator to decide how things should
work and what technologies should be implemented.

I am not trying to say that one method is better then the other, they
both have their merits.
-- 
Greg Kurtzer
Berkeley Lab, Linux guy