Hi Fu, Luotao Fu wrote: > Hi, > > I took some time today and went through Wolfgangs scenarios partly. Now > some results from my side. I ran my tests on a 2.6.24-rt1 > > Wolfgang Grandegger wrote: > > I also did some more measurements and made, by chance, interesting >> observations. I will summarize in more detail later on. Here are some >> preliminary results. My high latencies of up to 570us (without latency >> tracer) seem to be caused mainly by the following setting: >> >> CONFIG_RCU_TRACE=m >> > > I also got high latencies without CONFIG_RCU_TRACE set at all. setting > CONFIG_RCU_TRACE to y or m causes also high latency though in my test > runs. Hence I doubt the rcu Tracer is really the problem. > > As I mentioned in my last mail. Only thing I could reproduce reliably is > that the measurement results depend heavily on kind of the non-rt > Workload. For with hackbench or cache calibrator I couldn't produce > abnormal high latencies. Nor could I produce the hight latencies on a > system booted via flash. Hence my suspects stays on the fec irq thread > and filesystem access routines. > >> With CONFIG_NO_HZ=y or CONFIG_PPC_BESTCOMM_GEN_BD=y the latency >> increases by approx. 100..150us, each. >> > > Since I was producing high latencies independently from the rcu > settings. I didn't spend a lot of time playing around with the GEN_BD > and dynamic clock. All I can say is that disabling them also cause high > latency. ;-) The average results don't differ significantly in my test > runs.
In the meantime I have measured the impact more carefully and posted the results to the list: - CONFIG_RCU_TRACE or CONFIG_PREEMPT_RCU_BOOST not set causes high latencies up to 500..600us. - CONFIG_NO_HZ=y increases the latency by 90 us, at least. - CONFIG_PPC_BESTCOMM_GEN_BD=y seems not to harm. Do you still get high latencies with: CONFIG_PREEMPT_RCU_BOOST=y CONFIG_RCU_TRACE=y CONFIG_NO_HZ is not set With this setting I have not yet realized latencies > 150us. Could you please give it a try? If I change one of the parameters above, latency increases in short time. Thanks. Wolfgang. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
