The /proc stress test is now perfect, latencies max 2.5ms.
The "latency jitter" in all stress tests is much lower than on a unpatched
kernel. The figures are looking really flat.
Unfortunately the patch gives me still problem on disk I/O,
as usual I performed the tests with mem=256m , mem=128m and mem=64m
With mem=256m I get up to 28ms on the disk write test, with mem=128m the
latency goes down to 20ms and with mem=64m I get about 11ms.
The read test with mem=256m sometime gives me 5-6ms latencies
and with mem=64m the latency goes down to about 3.5ms.
Seems that we still need to tweak thing to get below 5ms reliably on the disk
read test.
The problems seem to be still correlated to the RAM size,
Ideas about which kernel function could cause this ?
here the tests: (note the decrease of latencies with reducing ramsize)
mem=256m
http://www.gardena.net/benno/linux/audio/2.2.10-mingo2-256m-1/3x256.html
mem=128m
http://www.gardena.net/benno/linux/audio/2.2.10-mingo2-128m-1/3x256.html
mem=64m
http://www.gardena.net/benno/linux/audio/2.2.10-mingo2-64m-1/3x256.html
Note that the write performance of writing a large file decreases by about 30%
using Mingo's patch, and looking at the harddisk led, the led flashes
intermittently instead of being always on.
The disk read perfomance is ok, it's the same like on an unpatched kernel.
Another fact to note is that re-running the same test a second time, gives me
better values, than the first test after a reboot , is this maybe related to
memory page allocation / buffering ?
It would be interesting to hear about the values that other people gets using
mingo's patches on PII (UP) +EIDE + large RAM ?
comments ?
regards,
Benno.
-- Benno Senoner E-Mail: sbenno@gardena.net Linux scheduling latency benchmarks http://www.gardena.net/benno/linux/audio
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/