Yeah, significant improvements with this revision! :-))) Only running for some hours so far, so I need more time to test, but my first impression is very positive. Seems like I cannot see interactivity differences vs. standard VRQ anymore. Let's read other's experience reports.
Thank you for your continued efforts, BR Manuel Krause
Also after over 24h of uptime and regular use, I can't report any issues, and can't even tell whether this test1 version is equal or maybe even superior vs. standard VRQ. Do your performance test numbers show differences?
@Manuel Sure there are performance improvement or I won't take effect into these new changes. It's about 0.5 to 2 seconds less than -vrq branch show in normal policy 200% to 300% workload kernel compile tests(about 3min for a single test), 50%~150% workload are most identical to -vrq branch.
I am planning to give 1~2 more test release to clean-up the code then merge these new changes to -vrq branch in next release.
I'm currently using 4.5 w/ VRQ, no issues so far, it's working very well, will compile a 4.6 test1 to check it out, but I have a simple question, how do You measure performance for interactiveness to compare changes?
Did You try to measure improvements over mainline or LQX kernels as well?
@anonymous Thanks for testing. Sure, interactiveness and performance are two major factors I want to balance in these changes(and -vrq branch). The last time I check the Cyclictest for the -test branch code a few days ago, there are no regression compare to -vrq branch. For the cyclictest test, you can have a look at my post at http://cchalpha.blogspot.com/2015/12/gc-and-vrq-branch-update-for-v431-and.html
Measure improvements over mainline kernel is not so fair as bfs and mainline cfs has different design targets. The last result I have done such tests are at https://cchalpha.blogspot.com/2015/08/41-gc-vrq-sanity-test-result-and-look.html , you can take it as a reference. I am thinking about having such tests for 4.6 in the rest of time frame in this release, will post the result when they are done.
Yeah, significant improvements with this revision! :-)))
ReplyDeleteOnly running for some hours so far, so I need more time to test, but my first impression is very positive. Seems like I cannot see interactivity differences vs. standard VRQ anymore.
Let's read other's experience reports.
Thank you for your continued efforts,
BR Manuel Krause
Thanks for the quick testing and feedback. :)
DeleteAlso after over 24h of uptime and regular use, I can't report any issues, and can't even tell whether this test1 version is equal or maybe even superior vs. standard VRQ.
DeleteDo your performance test numbers show differences?
BR Manuel Krause
@Manuel
DeleteSure there are performance improvement or I won't take effect into these new changes. It's about 0.5 to 2 seconds less than -vrq branch show in normal policy 200% to 300% workload kernel compile tests(about 3min for a single test), 50%~150% workload are most identical to -vrq branch.
I am planning to give 1~2 more test release to clean-up the code then merge these new changes to -vrq branch in next release.
BR Alfred
Very nice to hear! :-)
DeletePlease, keep us informed about these new releases, as always.
Very promising and successful work that you're doing.
BR Manuel Krause
Hi!
ReplyDeleteI'm currently using 4.5 w/ VRQ, no issues so far, it's working very well, will compile a 4.6 test1 to check it out, but I have a simple question, how do You measure performance for interactiveness to compare changes?
Did You try to measure improvements over mainline or LQX kernels as well?
Thanks
@anonymous
DeleteThanks for testing. Sure, interactiveness and performance are two major factors I want to balance in these changes(and -vrq branch). The last time I check the Cyclictest for the -test branch code a few days ago, there are no regression compare to -vrq branch. For the cyclictest test, you can have a look at my post at http://cchalpha.blogspot.com/2015/12/gc-and-vrq-branch-update-for-v431-and.html
Measure improvements over mainline kernel is not so fair as bfs and mainline cfs has different design targets. The last result I have done such tests are at https://cchalpha.blogspot.com/2015/08/41-gc-vrq-sanity-test-result-and-look.html , you can take it as a reference. I am thinking about having such tests for 4.6 in the rest of time frame in this release, will post the result when they are done.
BR Alfred