PE2950 and Linux virtualization

JACOB_LIBERMAN at Dell.com JACOB_LIBERMAN at Dell.com
Fri Jun 6 10:02:24 CDT 2008


Hi Scott,

This is really a complicated (and closesly scrutinzed) area for Dell
right now.

In general, the performance impact of virtualization on distributed
memory (clustered) scientific applications is dependant on the amount of
communication, interconnect type, and virtualization method.

If you are not clustering the application and your code runs in a
paravirtualized environment your performance will be near native if the
application is not IO intensive.  In some cases VM performance exceeds
native performance by reducing OS jitter caused by unstable drivers,
etc. VM performance suffers most when running IO intensive applications
-- particularly if multiple VMs are sharing the same underlying IO
subsystem.

If your application is clustered, you can achieve near native
performance and scalability provided there is very little IO and
inter-node communication.  You can also achieve near native performance
on communication intensive applications if you use an RDMA based cluster
interconnect (like iWARP or Infiniband) with a VMM-bypass layer.

Please see this excellent paper from OSU for more details:
http://nowlab.cse.ohio-state.edu/publications/conf-papers/2006/huangwei-
ics06.pdf . It includes native vs virtual performance benchmarking in
distributed memory application workloads.

In terms of use cases for virtualization with scientific computing --
job checkpointing, job migration, and environment customization are all
nice use cases.

I hope this is helpful.

Jacob 

 

> -----Original Message-----
> From: linux-poweredge-bounces at dell.com 
> [mailto:linux-poweredge-bounces at dell.com] On Behalf Of Scott 
> R. Ehrlich
> Sent: Thursday, June 05, 2008 7:55 PM
> To: linux-poweredge-Lists
> Subject: PE2950 and Linux virtualization
> 
> What are people's experiences with CentOS 5.0 64-bit 
> installed on dual 3 Ghz quad-core PE2950 systems with 32 GB 
> RAM each, high-performance computing (applications that tax 
> both the CPUs and RAM), not currently in a Beowolf cluster 
> but could adapt to that, and doing so with VMWare or other 
> vitualization software vs activity being done directly in the OS?
> 
> How much of a performance hit, or gain (I'd presume hit), 
> does virtualization cause an application, resulting in what 
> percentage poorer or better (I'd presume poorer) performance 
> vs dealing directly with the OS?
> 
> It would be nice to have a VM perform some work, and if a 
> person's code or application breaks, have it take down a VM 
> while keeping a machine up, and not affecting other people's work.
> 
> It may also depend on if an application or code is written 
> directly with/for the physical cpu/hardware vs more general use (VM).
> 
> Thanks for insights and experiences.
> 
> Scott
> 
> _______________________________________________
> Linux-PowerEdge mailing list
> Linux-PowerEdge at dell.com
> http://lists.us.dell.com/mailman/listinfo/linux-poweredge
> Please read the FAQ at http://lists.us.dell.com/faq
> 



More information about the Linux-PowerEdge mailing list