-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Hiya Joe,
On 19/10/13 23:10, Joseph Landman wrote:
> Did you try taskset to pin the CPU and the memory on the same numa
> node?
I wasn't, but using numactl to do that instead (which has more
powerful options from what I can see) shows that what I'm seeing is
the best case. :-(
if I pin the memory on the Max5 expansion unit I get slightly worse
performance and if I pin it to the RAM hanging off the other socket
it's worse again.
That correlates with what the kernel reports for distances between
NUMA nodes too (cheaper to hit RAM on the Max5 expansion than on the
other socket).
BTW numactl can show you the default binding, which is to the current
node:
[samuel@turpin Stream]$ numactl --show
policy: default
preferred node: current
physcpubind: 1
cpubind: 0
nodebind: 0
membind: 0 1 2
> What does numastat report?
It confirms it's writing only to the home NUMA node.
[root@turpin ~]# numastat stream_c.exe
Per-node process memory usage (in MBs) for PID 19275 (stream_c.exe)
Node 0 Node 1 Node 2
--------------- --------------- ---------------
Huge 0.00 0.00 0.00
Heap 0.00 0.00 0.00
Stack 0.02 0.00 0.00
Private 2062.07 0.00 0.00
- ---------------- --------------- --------------- ---------------
Total 2062.09 0.00 0.00
Total
---------------
Huge 0.00
Heap 0.00
Stack 0.02
Private 2062.07
- ---------------- ---------------
Total 2062.09
cheers!
Chris
- --
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
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.14 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
iEYEARECAAYFAlJlyNEACgkQO2KABBYQAh9lvwCfaNuwixvlFzP8z4Ejuxlzxjb2
ExUAnjehW/Cc+ejDGmAxAosqxxvhtlns
=GhfV
-----END PGP SIGNATURE-----
_______________________________________________
Beowulf mailing list, [email protected] sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
