Hello Steven, On Fri, 19 Aug 2022 at 23:29, Steven Rostedt <rostedt@xxxxxxxxxxx> wrote: > Hmm you need to turn off line wraps when pasting in output like this. > Because the line wraps make it very difficult to follow. * True, I should've done that; I'll remember to do it. > To get the vsocket CID id, you can run the agent: > > # trace-cmd agent > listening on @3:823 > > The @3 is the CID (823 is the port). > > Hit Ctrl-C on to kill the agent. > > Then on the host, run: > > # trace-cmd agent -P 3 > > Where 3 is the CID of the guest you want to connect the host (the same > number from the 'trace-cmd agent'. That is, if it was @4:823, then use > '-P 4'). The host agent above should also have a: > > listening on @2:823 > > or something. You'll need the host's CID too. > > Then on the guest, run: > > [ You do not need -p nop if you have -e used ] > > # trace-cmd record --poll -M 1E -e all --proxy 2 --name host -e kvm -e sched -e irq /home/test/rt-tests/oslat --cpu-list 1-4 --rtprio 1 --duration 12h -T 30 oslat V 2.30 > > And then when it ends, you'll have two files. One that is trace.dat and one > that is trace-host.dat. If everything worked well, the two would also be in > sync with each other. Doing: > > # trace-cmd report -i trace-host.dat -i trace.dat > > Will show the host events happening with the guest events. If the host > caused the latency, you will definitely see it. > > If you have the latest kernelshark, you can run: > > kernelshark trace-host.dat -a trace.dat > > And then select "Plots" -> "KVM Combo Plots" and in the popup, select the > box for "all". > > Then you'll see plots for each virtual CPU and the host thread that represents it. * I see, thank you so much for these interesting details. I'll try these steps and get back to you soon. Thank you. --- - P J P
|