Any conclusion on the Xeon context-switching issue?

Richard Huxton <dev@archonet.com> writes:

Subject says it all really. I've got a new client who seems to be
suffering from it, and I'm not sure if any conclusion was reached.

What's he using? 8.1 seems to have alleviated the problem somewhat,
and I've done more work in CVS tip. It'll never go away entirely,
because these chips are just not very good at sharing memory, but
we've certainly reduced it quite a bit from where it was in 7.x.

regards, tom lane

Tom Lane wrote:

Richard Huxton <dev@archonet.com> writes:

Subject says it all really. I've got a new client who seems to be
suffering from it, and I'm not sure if any conclusion was reached.

What's he using? 8.1 seems to have alleviated the problem somewhat,
and I've done more work in CVS tip. It'll never go away entirely,
because these chips are just not very good at sharing memory, but
we've certainly reduced it quite a bit from where it was in 7.x.

7.4.12 (.12 as of last week). I've seen context-switching peak at 80000
on a quad-Xeon that really shouldn't be straining.

An upgrade is possible, but obviously needs testing against. Would your
CVS changes be against 8.2-to-be or 8.1.x?

--
Richard Huxton
Archonet Ltd

Richard Huxton <dev@archonet.com> writes:

Tom Lane wrote:

What's he using?

7.4.12 (.12 as of last week). I've seen context-switching peak at 80000
on a quad-Xeon that really shouldn't be straining.

Try 8.1.3.

regards, tom lane

Tom Lane wrote:

Richard Huxton <dev@archonet.com> writes:

Tom Lane wrote:

What's he using?

7.4.12 (.12 as of last week). I've seen context-switching peak at 80000
on a quad-Xeon that really shouldn't be straining.

Try 8.1.3.

Thanks. I'll see if we can run a parallel installation overnight or
something.

--
Richard Huxton
Archonet Ltd

On Mon, 2006-02-27 at 16:54 +0000, Richard Huxton wrote:

Tom Lane wrote:

Richard Huxton <dev@archonet.com> writes:

Subject says it all really. I've got a new client who seems to be
suffering from it, and I'm not sure if any conclusion was reached.

What's he using? 8.1 seems to have alleviated the problem somewhat,
and I've done more work in CVS tip. It'll never go away entirely,
because these chips are just not very good at sharing memory, but
we've certainly reduced it quite a bit from where it was in 7.x.

7.4.12 (.12 as of last week). I've seen context-switching peak at 80000
on a quad-Xeon that really shouldn't be straining.

Is this causing an actual problem, or do you just like to see lower
numbers under the cs column? Certainly many people have asked about
this issue but few have pointed to application-level performance
problems that may have resulted.

-jwb