On 20/09/14 03:27, Christopher Samuel wrote:

> To be honest I'm suspicious of kernel autotuning here..

OK, did a quick bit of poking around.

On our M4 iDataplex I see 0 everywhere.

*but* on our M2 cluster I see:

[root@merri-m ~]# xdsh all -v cat /proc/sys/vm/zone_reclaim_mode | xcoll
====================================
idataplex,login,terri
====================================
1

====================================
merri081,merri083,turpin,merri082
====================================
0

So all the M2's have 1, along with an (old, 2011, Westmere) SGI UV10.

The nodes that are all 0 are x3690 X5 (Westmere with Maxx5's).

Now they're all booting the same statelite image and so I know it's
not something we've configured:

provmethod=rhel64_gpfs35016_updates

So I'd say that's highly likely to be kernel autotuning.


Aha - there was a patch merged for 3.16 to disable it by default!

commit 4f9b16a64753d0bb607454347036dc997fd03b82
Author: Mel Gorman <[email protected]>
Date:   Wed Jun 4 16:07:14 2014 -0700

    mm: disable zone_reclaim_mode by default
    
    When it was introduced, zone_reclaim_mode made sense as NUMA distances
    punished and workloads were generally partitioned to fit into a NUMA
    node.  NUMA machines are now common but few of the workloads are
    NUMA-aware and it's routine to see major performance degradation due to
    zone_reclaim_mode being enabled but relatively few can identify the
    problem.
    
    Those that require zone_reclaim_mode are likely to be able to detect
    when it needs to be enabled and tune appropriately so lets have a
    sensible default for the bulk of users.
    

That modified the text describing the sysctl in vm.txt, and that
old text confirms that the kernel does autotune this..

zone_reclaim_mode is set during bootup to 1 if it is determined that pages
from remote zones will cause a measurable performance reduction. The
page allocator will then reclaim easily reusable pages (those page
cache pages that are currently not used) before allocating off node pages.

So there you go!

Best of luck,
Chris (off to catch a plane)
-- 
 Christopher Samuel        Senior Systems Administrator
 VLSCI - Victorian Life Sciences Computation Initiative
 Email: [email protected] Phone: +61 (0)3 903 55545
 http://www.vlsci.org.au/      http://twitter.com/vlsci


------------------------------------------------------------------------------
Slashdot TV.  Video for Nerds.  Stuff that Matters.
http://pubads.g.doubleclick.net/gampad/clk?id=160591471&iu=/4140/ostg.clktrk
_______________________________________________
xCAT-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/xcat-user

Reply via email to