On Fri, Jul 14, 2000, Roger Larsson wrote: > Iain Sandoe wrote: >> I've posted the results of Benno's latency tests applied to 2.2.17pre10-ben2 >> on LinuxPPC. [...] >> No big surprises - although the PPC /proc fs seems to be a little slower >> than the x86 one... >> >> Working on 2.4.0-test4 & thence to Andrew's patches on LinuxPPC... [...] > The green ticks on X11 test are interesting - something with a higher > prio than > the SCHED_FIFO max takes the processor away from it and runs for more > than 1 ms. > > The only thing that can do that is interrupt routines / back handlers... > (try to find this one) Yeah, I've got a concern there... I have Jun Sun's IRQ latency test code ported. I'll re-do the patch and try it again. I have a feeling that the first time through (with the IRQ code) I may have made a mistake in the value of the freq of the VEA Timebase counter. If this is true then the IRQ times for the PPC are truly awful on occasions.. I.E. I had the freq as 300 MHz - but I think it is actually approx 18MHz. If that is the case then there is/are something/some things that are holding IRQs off for up to.... > 9 or so ***ms*** Yikes. I'll check this out tomorrow... but locating them might be hard because most of the longer times are recorded in 'softira.c' & 'irq.c' - which doesn't narrow things down much. Expect a re-do of the IRQ patch tomorrow - and a request to try it out :-) > Disk read looks nice - others looks ... ummm /proc was the one that looked worst to me... Iain.
