Hello,
These results represent running the following benchmarks on a true NUMA
machine (16-way with 32GB of RAM). To make the table somewhat less wide,
the following convention has been adopted:
***THIS IS DIFFERENT THAN BEFORE***
1) == 2.6.12-rc1
2) == 2.6.12-rc1 + current CKRM core patches + CONFIG_CKRM=n
3) == 2.6.12-rc1 + current CKRM core patches + CONFIG_CKRM=y
4) == 2.6.12-rc1 + current CKRM core patches + CONFIG_CKRM=y + 100
classes
5) == 2.6.12-rc1 + current CKRM core & memrc patches + CONFIG_CKRM=n
As should be clear from the choice of results, I am not able to run with
the memory controller enabled on NUMA. Chandra is aware of the issue.
As before, instead of posting raw numbers, I have provide only the
summary percentage change per each run.
kernbench -- 64 threads, 20 iterations
percentage of elapsed time relative to mainline
----------------------------------------------------------------------
1) 100.00
2) 99.86
3) 100.08
4) 100.06
5) 100.01
dbench -- 20 clients, 20 iterations
percentage of throughput relative to mainline
----------------------------------------------------------------------
1) 100.00
2) 99.14
3) 99.82
4) 99.14
5) 100.71
tbench -- 10 clients, 20 iterations
percentage of throughput relative to mainline
----------------------------------------------------------------------
1) 100.00
2) 100.07
3) 99.94
4) 99.54
5) 101.00
SPECjbb [1] -- starting 1 warehouse, by 1s, to 20 warehouses,
20 iterations
percentage of score relative to mainline [3]
----------------------------------------------------------------------
1) 100.00
2) 99.73
3) 109.10
4) 103.63
5) 97.68
SDET [2] -- 20 iterations
percentage of throughput relative to mainline
# scripts: 1 4 16 64 128
----------------------------------------------------------------------
1) 100.00 100.00 100.00 100.00 100.00
2) 99.60 99.91 99.49 99.70 99.99
3) 100.90 100.56 99.55 100.10 100.01
4) 100.68 99.76 99.40 99.97 100.23
5) 100.31 100.52 100.15 100.13 100.16
I have profiler output for all runs in all benchmarks, if anyone cares
to see what may have caused certain runs to behave as they did, I can do
a diffprofile and mail the results.
Thanks,
Nish
[1] Disclaimer:
SPEC(tm) and the benchmark name SPECjbb(tm) are registered trademarks
of the Standard Performance Evaluation Corporation. The benchmarking was
conducted for research purposes only and were non-compliant with the
following deviations from the rules:
1. It was run on hardware that does not meet the SPEC
availability-to-the public criteria. The machine was an
engineering sample.
[2] DISCLAIMER: SPEC(tm) and the benchmark name SDET(tm) are registered
trademarks of the Standard Performance Evaluation Corporation. This
benchmarking was performed for research purposes only, and the run
results are non-compliant and not-comparable with any published results.
[3] There seems to be a problem with SPECjbb(tm) on NUMA; each of these
runs was labelled invalid, so the score outputs were all estimates
(which SPECjbb produced). Thus, I am pretty sure we should disregard the
output of SPECjbb for the NUMA run and concentrate on the PPC64 one
(which I shall be posting shortly).
-------------------------------------------------------
This SF.Net email is sponsored by: New Crystal Reports XI.
Version 11 adds new functionality designed to reduce time involved in
creating, integrating, and deploying reporting solutions. Free runtime info,
new features, or free trial, at: http://www.businessobjects.com/devxi/728
_______________________________________________
ckrm-tech mailing list
https://lists.sourceforge.net/lists/listinfo/ckrm-tech
