VM tests on 2.4.14 2.4.15-pre[12] and Athlon opt

rwhron@earthlink.net
Fri, 9 Nov 2001 20:46:22 -0500


Summary: Athlon optimized kernel uses higher percentage of
CPU (36% vs ~ 25%) and takes a little longer on memory
fill test. Athlon kernel a little faster on mmap test.

Test: Fill and write 80% of virtual memory with "mtest01 -w -p80"
mmap, touch, msync, and munmap 500000 pages "mmap001 -m 500000"
Listen to mp3 sampled at 128k. (usual test).
Test with/without Athlon processor configuration.

There were no mp3blaster skips on the mmap001 tests.

config differences for athlon:

diff p5 k7
< CONFIG_M586=y
> CONFIG_MK7=y

< CONFIG_X86_L1_CACHE_SHIFT=5
< CONFIG_X86_USE_STRING_486=y
< CONFIG_X86_ALIGNMENT_16=y

> CONFIG_X86_L1_CACHE_SHIFT=6
> CONFIG_X86_TSC=y
> CONFIG_X86_GOOD_APIC=y
> CONFIG_X86_USE_3DNOW=y
> CONFIG_X86_PGE=y
> CONFIG_X86_USE_PPRO_CHECKSUM=y

2.4.14

mp3 played 248 of 301 seconds (82%)

Averages for 10 mtest01 runs
bytes allocated: 1236900249
User time (seconds): 2.014
System time (seconds): 7.045
Elapsed (wall clock) time: 30.853
Percent of CPU this job got: 28.90
Major (requiring I/O) page faults: 106.5
Minor (reclaiming a frame) faults: 302773.1

Average for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 19.518
System time (seconds): 18.276
Elapsed (wall clock seconds) time: 186.31
Percent of CPU this job got: 19.80
Major (requiring I/O) page faults: 500163.4
Minor (reclaiming a frame) faults: 42.6

2.4.15-pre1

mp3 played 257 of 303 seconds (84%)

Averages for 10 mtest01 runs
bytes allocated: 1233649664
User time (seconds): 2.060
System time (seconds): 5.280
Elapsed (wall clock) time: 30.314
Percent of CPU this job got: 23.70
Major (requiring I/O) page faults: 106.3
Minor (reclaiming a frame) faults: 301976.7

Average for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 19.360
System time (seconds): 17.950
Elapsed (wall clock seconds) time: 188.15
Percent of CPU this job got: 19.40
Major (requiring I/O) page faults: 500165.2
Minor (reclaiming a frame) faults: 41.0

A note from Alan Cox on lkml said 2.4.14 should handle buggy VIA
chipsets. My KT133A now runs without oops compiled as Athlon.
(kernels <= 2.4.9 didn't on my IWill KK266 mobo).

The next two tests are athlon optimized. To me, the biggest
revelation is the higher "Percent of CPU this job got" on the
mtest01 runs.

2.4.15-pre1 - athlon optimized

mp3 played 275 of 317 seconds (86%)

Averages for 10 mtest01 runs
bytes allocated: 1238577971
User time (seconds): 5.475
System time (seconds): 6.032
Elapsed (wall clock) time: 31.683
Percent of CPU this job got: 36.00
Major (requiring I/O) page faults: 106.4
Minor (reclaiming a frame) faults: 303181.0

Average for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 19.218
System time (seconds): 18.214
Elapsed (wall clock seconds) time: 182.07
Percent of CPU this job got: 20.00
Major (requiring I/O) page faults: 500163.4
Minor (reclaiming a frame) faults: 42.6

2.4.15-pre2 - athlon optimized

mp3 played 268 of 314 seconds (85%)

Averages for 10 mtest01 runs
bytes allocated: 1235851673
User time (seconds): 5.447
System time (seconds): 6.099
Elapsed (wall clock) time: 31.400
Percent of CPU this job got: 36.40
Major (requiring I/O) page faults: 106.8
Minor (reclaiming a frame) faults: 302515.9

Average for 5 mmap001 runs
bytes allocated: 2048000000
User time (seconds): 19.566
System time (seconds): 18.472
Elapsed (wall clock seconds) time: 183.45
Percent of CPU this job got: 20.20
Major (requiring I/O) page faults: 500164.4
Minor (reclaiming a frame) faults: 41.8

-- 
Randy Hron

- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/