On 1/27/2014 2:03 PM, riel@xxxxxxxxxx wrote:
The current automatic NUMA balancing code base has issues with
workloads that do not fit on one NUMA load. Page migration is
slowed down, but memory distribution between the nodes where
the workload runs is essentially random, often resulting in a
suboptimal amount of memory bandwidth being available to the
workload.
In order to maximize performance of workloads that do not fit in one NUMA
node, we want to satisfy the following criteria:
1) keep private memory local to each thread
2) avoid excessive NUMA migration of pages
3) distribute shared memory across the active nodes, to
maximize memory bandwidth available to the workload
This patch series identifies the NUMA nodes on which the workload
is actively running, and balances (somewhat lazily) the memory
between those nodes, satisfying the criteria above.
As usual, the series has had some performance testing, but it
could always benefit from more testing, on other systems.
Changes since v4:
- remove some code that did not help performance
- implement all the cleanups suggested by Mel Gorman
- lots more testing, by Chegu Vinod and myself
- rebase against -tip instead of -next, to make merging easier
Acked-by: Chegu Vinod <chegu_vinod@xxxxxx>
---
The following 1, 2, 4 & 8 socket-wide results on an 8-socket box are an
average of 4 runs.
I) Eight 1-socket wide instances (10 warehouse threads/instance)
a) numactl pinning results
throughput = 350720 bops
throughput = 355250 bops
throughput = 350338 bops
throughput = 345963 bops
throughput = 344723 bops
throughput = 347838 bops
throughput = 347623 bops
throughput = 347963 bops
b) Automatic NUMA balancing results
(Avg# page migrations : 10317611)
throughput = 319037 bops
throughput = 319612 bops
throughput = 314089 bops
throughput = 317499 bops
throughput = 320516 bops
throughput = 314905 bops
throughput = 315821 bops
throughput = 320575 bops
c) No Automatic NUMA balancing and NO-pinning results
throughput = 175433 bops
throughput = 179470 bops
throughput = 176262 bops
throughput = 162551 bops
throughput = 167874 bops
throughput = 173196 bops
throughput = 172001 bops
throughput = 174332 bops
-------
II) Four 2-socket wide instances (20 warehouse threads/instance)
a) numactl pinning results
throughput = 611391 bops
throughput = 618464 bops
throughput = 612350 bops
throughput = 616826 bops
b) Automatic NUMA balancing results
(Avg# page migrations : 8643581)
throughput = 523053 bops
throughput = 519375 bops
throughput = 502800 bops
throughput = 528880 bops
c) No Automatic NUMA balancing and NO-pinning results
throughput = 334807 bops
throughput = 330348 bops
throughput = 306250 bops
throughput = 309624 bops
-------
III) Two 4-socket wide instances (40 warehouse threads/instance)
a) numactl pinning results
throughput = 946760 bops
throughput = 949712 bops
b) Automatic NUMA balancing results
(Avg# page migrations : 5710932)
throughput = 861105 bops
throughput = 879878 bops
c) No Automatic NUMA balancing and NO-pinning results
throughput = 500527 bops
throughput = 450884 bops
-------
IV) One 8-socket wide instance (80 warehouse threads/instance)
a) numactl pinning results
throughput = 1199211 bops
b) Automatic NUMA balancing results
(Avg# page migrations : 3426618)
throughput = 1119524 bops
c) No Automatic NUMA balancing and NO-pinning results
throughput = 789243 bops
Thanks
Vinod
Changes since v3:
- various code cleanups suggested by Mel Gorman (some in their own patches)
- after some testing, switch back to the NUMA specific CPU use stats,
since that results in a 1% performance increase for two 8-warehouse
specjbb instances on a 4-node system, and reduced page migration across
the board
Changes since v2:
- dropped tracepoint (for now?)
- implement obvious improvements suggested by Peter
- use the scheduler maintained CPU use statistics, drop
the NUMA specific ones for now. We can add those later
if they turn out to be beneficial
Changes since v1:
- fix divide by zero found by Chegu Vinod
- improve comment, as suggested by Peter Zijlstra
- do stats calculations in task_numa_placement in local variables
Some performance numbers, with two 40-warehouse specjbb instances
on an 8 node system with 10 CPU cores per node, using a pre-cleanup
version of these patches, courtesy of Chegu Vinod:
numactl manual pinning
spec1.txt: throughput = 755900.20 SPECjbb2005 bops
spec2.txt: throughput = 754914.40 SPECjbb2005 bops
NO-pinning results (Automatic NUMA balancing, with patches)
spec1.txt: throughput = 706439.84 SPECjbb2005 bops
spec2.txt: throughput = 729347.75 SPECjbb2005 bops
NO-pinning results (Automatic NUMA balancing, without patches)
spec1.txt: throughput = 667988.47 SPECjbb2005 bops
spec2.txt: throughput = 638220.45 SPECjbb2005 bops
No Automatic NUMA and NO-pinning results
spec1.txt: throughput = 544120.97 SPECjbb2005 bops
spec2.txt: throughput = 453553.41 SPECjbb2005 bops
.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>