RE: [R] R on AMD64 (Opteron)

From: Liaw, Andy <andy_liaw_at_merck.com>
Date: Mon 26 Jul 2004 - 23:55:01 EST


> From: Jerome Asselin
>
> From: "Liaw, Andy" <andy_liaw@merck.com>
>
> > Just one more thing for folks intending to try Opterons with >1 cpu:
> You'll
> > probably want to use a NUMA kernel, rather than a SMP one.
> >
> > Cheers,
> > Andy
>
> Thanks for the advice, but is there another reason why you
> write this beside
> a large (I mean >8) number of CPUs?
>
> At http://lse.sourceforge.net/numa/faq/ I read,
>
> What is the difference between NUMA and SMP?
> The NUMA architecture was designed to surpass the scalability
> limits of the
> SMP architecture. With SMP, which stands for Symmetric
> Multi-Processing, all
> memory access are posted to the same shared memory bus. This
> works fine for
> a relatively small number of CPUs, but the problem with the shared bus
> appears when you have dozens, even hundreds, of CPUs
> competing for access to
> the shared memory bus. NUMA alleviates these bottlenecks by
> limiting the
> number of CPUs on any one memory bus, and connecting the
> various nodes by
> means of a high speed interconnect.

I was told by folks at Penguin Computing (where we bought our Opterons) and AMD that because of the way memory controller is integrated in the Opteron, when running two CPU intensive processes on a dual CPU box, one cpu will be at near 100% while the other will be only around 90%, if you use a SMP kernel. Indeed that's what we saw when we ran two R processes simultaneously. Switching to the NUMA kernel for the most part `solves' that problem.

Andy



R-help@stat.math.ethz.ch mailing list
https://www.stat.math.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Tue Jul 27 08:32:12 2004

This archive was generated by hypermail 2.1.8 : Wed 03 Nov 2004 - 22:55:16 EST