On Sun, Sep 14, 2014 at 08:07:17PM +0200, Stefan Fritsch wrote:
On Sunday 14 September 2014 12:13:44, Brent Cook wrote:
Results for an old Athlon (hmm, I don't remember it running at 10.8
Ghz before)
cpu0: AMD Athlon(tm) 64 X2 Dual Core Processor 6000+, 10823.06 MHz
forktest 0m19.23s real 0m0.72s user 0m22.46s system
forktest+ 0m16.33s real 0m0.50s user 0m18.22s system
kernel -j2 3m0.68s real 5m10.95s user 0m41.58s system
kernel -j2+ 3m0.45s real 5m10.14s user 0m41.53s system
On Sunday 14 September 2014 19:16:41, Job Snijders wrote:
The speedup kernel seems to perform slightly better:
[root@speedup ~]# time ./forktest
real 0m33.134s
user 0m1.230s
sys 0m53.710s
[[email protected] ~]# time ./forktest
real 0m37.600s
user 0m1.200s
sys 0m59.530s
Thanks to both of you for testing.
These results are more in line with my Core i7 and I expected the
effect to be larger the large the number of CPUs is. Just my 6-year-
old Core2 Duo does not fit in there and gives a much larger speedup.
Maybe it is particularily bad at something the current pmap code does.
Hello Stefan,
This machine is an old AMD Turion64 @2GHz, and all tests were done with X off.
cpu0: AMD Turion(tm) 64 Mobile Technology ML-37, 1990.08 MHz
I tested by forcing CPU speed both low and high.
-current
========
apm -L
Performance adjustment mode: manual (800 MHz)
41.62s real 2.26s user 24.32s system
40.35s real 2.21s user 24.65s system
38.40s real 2.18s user 23.27s system
40.15s real 2.30s user 23.64s system
41.15s real 2.34s user 22.77s system
38.77s real 2.03s user 23.49s system
----------- ---------- -------------
~ 40.24s real 2.22s user 23.69s system
apm -H
Performance adjustment mode: manual (2000 MHz)
19.51s real 0.80s user 11.91s system
19.95s real 1.00s user 12.44s system
19.28s real 0.76s user 12.29s system
19.53s real 0.74s user 11.81s system
18.49s real 0.90s user 11.51s system
19.94s real 0.94s user 12.44s system
----------- ---------- -------------
~ 19.45s real 0.85s user 12.06s system
kernel -j2 9m22.59s real 8m17.30s user 47.90s system
speeddup
========
apm -L
Performance adjustment mode: manual (800 MHz)
39.67s real 1.88s user 22.77s system
38.91s real 2.32s user 24.78s system
39.75s real 2.35s user 23.11s system
38.19s real 2.02s user 23.18s system
40.46s real 2.18s user 24.10s system
40.56s real 2.24s user 22.53s system
----------- ---------- -------------
~ 39.59s real 2.16s user 23.41s system
apm -H
Performance adjustment mode: manual (2000 MHz)
21.66s real 1.03s user 12.72s system
19.81s real 1.08s user 11.44s system
19.67s real 0.75s user 11.99s system
19.28s real 0.76s user 11.56s system
19.38s real 0.92s user 12.10s system
19.71s real 1.00s user 12.17s system
----------- ---------- -------------
~ 19.91s real 0.92s user 11.99s system
kernel -j2 9m38.43s real 8m32.10s user 49.10s system
As you can see, there is not much difference in speed, but it looks stable.
But as it's SP only, it's probably expected.
Unfortunately, my MP AMD PhenomII x8 is currently used as a SmartOS
virtualizer. I may try to load a MP VM to test the patch on it, but it will
have to wait a bit more.
Thierry
