Mailinglist Archive: opensuse-buildservice (119 mails)

< Previous Next >
Re: [opensuse-buildservice] Linux scheduler bugs
On Mon, May 2, 2016 at 7:05 AM, Jan Engelhardt <jengelh@xxxxxxx> wrote:

On Friday 2016-04-29 01:23, Greg Freemyer wrote:

I believe in the Intel world at least, all 6 core (12 thread) single
socket (CPU) are all SMP, not NUMA.

You make it sound like NUMA machines are not SMP, which is almost
universally not the case. Anyhow, SMP and NUMA are separate concepts.

Hmm..

I am surprised. I guess I assumed that "symmetric" and "non-uniform"
were contradictions.

But it is just a matter of terminology and I'm not the one defining
them, so I won't argue. (I will go do some reading.)

If you have multiple sockets (CPUs) you are definitely into NUMA (per
my understanding).

I would question that for FSB generations (era Pentium 2), but it is
hard to find information on those long gone processors.

I'm happy to restrict my comment to Pentium 4 class CPUs or newer.

I'm not sure when NUMA kicks in for an individual Intel CPUs (socket)
(but I know that it does eventually).

It is there whenever the machine description says there is. (See
ACPI tables.) You can play around with that in VirtualBox IIRC.

But you didn't answer the fundamental question.

The scheduler bugs seem to affect systems using the NUMA feature of
the scheduler more than those that don't.

lscpu | grep NUMA

provides information about NUMA. Are the servers in the public
instance of OBS using NUMA based hardware. i.e. Do they have more
than one NUMA node?

Greg
--
To unsubscribe, e-mail: opensuse-buildservice+unsubscribe@xxxxxxxxxxxx
To contact the owner, e-mail: opensuse-buildservice+owner@xxxxxxxxxxxx

< Previous Next >
Follow Ups
References