> > On Mon, Aug 05, 2002 William Lee Irwin III wrote:
> >>> Look for pagetable space consumption. BTW, what kernel version?
> > On Tue, Aug 06, 2002 at 09:20:44AM -0700, Cliff White wrote:
> >> What is the best way to measure pagetable space? /proc/meminfo?
> >> The kernel version for these runs is stock 2.4.18. We have run
> >> some other kernels, there is a rough table of results with links to
> >> details at http://www.osdl.org/projects/dbt1prfrns/results/
> >> cliffw
> > Sorry about that, I should have mentioned 2 things:
> > (1) this is not the only issue, this is one of the things I'm particularly
> > concerned about and know is an issue arising with databases
> > (2) it'll take some backporting from 2.5.x to get the pagetable accounting
> > into 2.4.x kernels.
> If you're running stock 2.4.18, you don't have highpte in any form.
> So cat /proc/meminfo, and if you're not low on low mem, this is not
> the problem. If you are low on low mem, it may or may not be the problem ;-)
> The output of /proc/slabinfo during a run would be useful too.
For the STP we've been getting /proc/meminfo at the start and end of the run.
The 2-way has only 1GB physical memory (sigh), so 2-way runs might be less
I don't think we care that much about 2.4.18, unless
you have other motives for doing the backport, don't do it for us.
We are working mostly on the 8-cpu machine for >4G memory runs, (it has 16G)
but we need more driver hardware
which will take another day or so. Would appreciate a suggested 2.4.* kernel
version/patch for the 8-way.
Per previous request, i am building 2.5.30 with the akpm 'everything' patch,
we'll try that on the
1-tier rig first and let you know more.