scheduler latency
Hi,
I've done some scheduler latency tests and come up with some pretty plots, so I thought I'd post them. It is nicksched vs mainline.
Note: this is not realtime scheduling latency, but latency as measured by my random program. It also does not indicate interactivity, because the latencies measured were all under 15ms which is probably below perception. Even if not, it doesn't measure latency as you might see it.
Finally, some of the tests involved me moving the mouse around, so factor in by subliminal (or not) bias toward my scheduler.
Quoted 10 lines · Nick Piggin
I've done some scheduler latency tests and come up with some pretty plots, so I thought I'd post them. It is nicksched vs mainline.
Note: this is not realtime scheduling latency, but latency as measured by my random program. It also does not indicate interactivity, because the latencies measured were all under 15ms which is probably below perception. Even if not, it doesn't measure latency as you might see it.
Quoted 4 lines
Finally, some of the tests involved me moving the mouse around, so factor in by subliminal (or not) bias toward my scheduler.
Nick, when I did the SOFTRR hack, I also measured sched latencies using another measurement/load apps:
http://www.xmailserver.org/linux-patches/softrr.html
At the bottom you can find the measurement app (note: x86 only since it uses "rdtsc") and stressful (latency-wise) loads.
- Davide