Code review comment for ~johill-lanl/epics-base/+git/epics-base:timer-queue-fix
Revision history for this message
| Jeff Hill (johill-lanl) wrote | # |
© 2004
Canonical Ltd.
•
Terms of use
•
Data privacy
•
Contact Launchpad Support
•
Blog
•
Careers
•
System status
•
73416ba
(Get the code!)
Martin,
I will hopefully attend to to the build errors tomorrow. Suspect that MSVC will never work with valgrind? If so we can simply ifdef the valgrind macros out? Need to take closer look.
Also need to look at your performance metrics.
Jeff
_______
From: <email address hidden> <email address hidden> on behalf of Martin Konrad <email address hidden>
Sent: Friday, April 24, 2020 1:27 PM
To: Hill, Jeff
Subject: [EXTERNAL] Re: [Merge] ~johill-
I haven't looked at the code, yet, but it builds fine on Ubuntu 19.10 with GCC 9.2: no warnings, all tests pass :-) A few compiler/OS combinations fail to build on Travis (https:/
Before digging into the code, I benchmarked the old and the proposed timer queue implementation. You can find the results here:
https:/
The old implementation takes more than 100 us to create and start a new timer if there are already 30,000 timers in the queue (this number is based on a real-world scenario with Asyn/Stream at FRIB) and the new timer needs to be inserted at the beginning of the list (30,000 existing timers with 60 s, the new timer has 30 s duration). In this case the old code has to walk a linked list from the back all the way to the front (the code was optimized for inserting timers with the same duration). The new implementation performs two orders of magnitude better in this scenario. Not bad! None of the cases I benchmarked is performing significantly worse than with the old implementation. For comparison I also benchmarked boost::asio's timers - its performance is in the same order of magnitude.
See https:/
--
https:/
You are the owner of ~johill-