On Fri, 30 Nov 2018 15:42:35 +0000 Yordan Karadzhov <ykaradzhov@xxxxxxxxxx> wrote: > These two patches aim to restore the log(n) time complexity of > the latency plotting, by adding Data Collections for Sched events. > > Steven, please try playing with this modification before applying it. > If you can test this, when doing some real work this would be great. This works awesomely! Check out this screenshot: http://rostedt.org/private/ks-latency.png The old way I could not see that huge wake up latency. Also, I noticed that fsync got preempted (see the red square to the left). I have to ask, how hard is it to click on the wakeup (the start of the green box) and get the marker to go there? I can find it difficult to measure because I can't get the marker to set to the start without doing a search for the wakeup. -- Steve > > Yordan Karadzhov (2): > kernel-shark-qt: Add a method for adding a new collection to a list > kernel-shark-qt: Make Sched event plugin use its own data collections > > kernel-shark-qt/src/libkshark-collection.c | 45 ++++++++++++++++++++- > kernel-shark-qt/src/libkshark.c | 1 + > kernel-shark-qt/src/libkshark.h | 9 +++++ > kernel-shark-qt/src/plugins/SchedEvents.cpp | 33 ++++++--------- > kernel-shark-qt/src/plugins/sched_events.c | 21 ++++++++++ > kernel-shark-qt/src/plugins/sched_events.h | 6 +++ > 6 files changed, 93 insertions(+), 22 deletions(-) >
|