I've had my fixes running on the phone for 12950+ samples at 2Hhz and I
believe the fixes I've pushed into the eventstat repo address the key
issues:
1. no more huge CPU consumption after a long run time
- fixed by re-working the timer cache into a hash table
- fixed by reaping old expired timers that have not been used for > 600
seconds
2. fixed the weird extra CPU "total events" escalation over time:
- improved hashing on the event information with a better and wider hash
- optimized expensive code paths to lower the time to correlate events . This
avoids the time between each short sample time expanding and causing us
to measure more "total events" per sample time because the time interval
is no longer expanded over the time delay interval.
3. misc optimizations. I've run the code through perf and gcov and worked on
the
more CPU and cache expensive code paths.
I'm going to push a new release out in the next 48 hours.
** Attachment added: "eventstat-stats.ods"
https://bugs.launchpad.net/ubuntu/+source/eventstat/+bug/1467932/+attachment/4420460/+files/eventstat-stats.ods
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/1467932
Title:
eventstat output suddenly accounts kernel threads as userspace
processes after a prolonged amount of time
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/eventstat/+bug/1467932/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
