On Fri, Jun 13, 2008 at 2:31 PM, Jens Wirth <jewirth@xxxxxxxxx> wrote: > Oleg Galimov <Galimov <at> gmx.de> writes: > >> I have a dual-core cpu. When I run cyclictest on both cores and then start a > time hog (like cp /dev/zero >> /dev/null), then the average latency on the core where it runs goes down > significantly (on my machine it is >> like 30us before and 10us after, when using rt-patch). It returns to the > original values, when I kill the cp process. > > I made a similar observation while measuring the jitter in > the time of execution of usleep(). The more load the machine > has the less becomes the average jitter. Of course, there is > a point of inflection when the system is fully stretched. > > I observed this behavior on a ppc405 based embedded system > (no SMP) with and without CONFIG_PREEMPT_RT. Interestingly > enough, I was unable to reproduce your observation with the > cyclictest tool on this system. I have to correct myself: When running multiple threads I get the same behavior as you described. Remarkably, this is a single core system with a fixed cpu frequency. Jens -- To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
