Beowulf & VMWare

[Donald Becker]

On Fri, 1 Nov 2002, Laurence Liew wrote:

> Vmware GSX/ESX is great for server consolidation. But consolidated 
> servers are not meant for Beowulf clusters and vis-versa.

Oh, I strongly disagree.  The Scyld system is being used for server
consolitation.

> Imagine your Oracle database processing a large transaction, and all of 
> a sudden you start up a vmware session to do some other stuff... both 
> your Oracle db and new vmware session will be severly impacted.

That's why it's important to make both status monitoring and explicit
node allocation part of the cluster system.  Status monitoring allows
CPU loads and memory pressure to influence scheduling.  Explicit node
allocation is an override to the automatic scheduling.

> Server consolidation requires careful studies into loads of various 
> applications that will be running on the server... you cannot just move 
> the VMs around (I am thinking of openmosix + vmware.. dynamic vmware 
> allocation in a cluster)...

As most list readers know, I consider Mosix to be an academically
interesting system and its transparent process migration an excellent
touchstone to compare process migration approachs.

Mosix is not a good system for scalable or performance-oriented
clustering, and I agree with your conclustion that Mosix+VMware
application serving is not a good match.

> we see customers consolidating 100+ servers into 8 - 10 large 4/8 ways 
> servers. And in these 8-10 large servers.. Vmware sessions are carefully 
> partitioned according to CPU loads, memory / disk space requirements.

Specifically, you almost never want to migrate an already running
VMWare instance.  Consider it something to be scheduled around instead
of something to schedule.  This approach does work reasonably well for
most server consolidation environments.

-- 
Donald Becker				becker at scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Scyld Beowulf cluster system
Annapolis MD 21403			410-990-9993