BMQ v5.5-r2 is released with the following changes
1. Spring code clean-up to better aligned with mainline scheduler code.
These are minor clean-up code changes, other lager code changes and feature add-on found during code clean-up is planned in next release.
Enjoy BMQ for your linux kernel.
Full kernel tree repository can be found at https://gitlab.com/alfredchen/linux-bmq
And all-in-one patch can be found at gitlab.
Bug report at https://gitlab.com/alfredchen/bmq/issues
running on 4 machines without problems.
ReplyDeleteAlfred Chen'S Blog: Bmq V5.5-R2 Release >>>>> Download Now
Delete>>>>> Download Full
Alfred Chen'S Blog: Bmq V5.5-R2 Release >>>>> Download LINK
>>>>> Download Now
Alfred Chen'S Blog: Bmq V5.5-R2 Release >>>>> Download Full
>>>>> Download LINK 1W
I tested BMQ on a 3970x.
ReplyDeleteI benchmarked with geekbench 5.
If i ran the benchmark with default nice level BMQ performs better but with more variation between the scores. single core score is almost always better or equal.
But if I run the vanilla scheduler with "schedtool -R -p99 -n -19" the single core is equal to BMQ and the multicore score is much higher than with BMQ.
vanilla: https://browser.geekbench.com/v5/cpu/1401196
BMQ: https://browser.geekbench.com/v5/cpu/1342578
any idea why this could be or a solution to get the same behavior on BMQ?
I tried SCHED_RR and SCHED_FF but with no effect. nice levels also doesn't seem to do much for BMQ.
My .config for BMQ:
Deletehttps://pastebin.com/beRvm5vS
I also tried idle nohz and 250 hz with no luck so far.
Also here a more recent BMQ run:
https://browser.geekbench.com/v5/cpu/1411175
The timeslice setting also doesn't seem to make a big difference.
vanilla scores consistently over 34k BMQ bettes 26k and 28k.
It's a known behaviour for real-time scheduler policy(FIFO&RR), as BMQ currently designed to be optimized for NORMAL policy only. For real-time tasks, they are all put into a single double-linked data-structure. Improvement can be make but it was at low priority.
DeleteNice level doesn't works for SCHED_RR and SCHED_FF, unless they changes back to NORMAL tasks.
schedtool -N -10 or -19 also doesn't improve the scores.
DeleteMaybe BMQ is just a little bit too fair with vanilla scheduler and SCHED_RR the console is visibly lagging with BMQ never.
Nice level doesn't improve benchmark score, it just adjust priority of the given task among the others. Obviously there is no others when your benchmark is running.
DeleteFor the score, I can't tel if it is comparable or not, as it has been reported BMQ doesn't work well with ryzen last year. The ticket is closed, but I don't have ryzen HW to test/improve it myself.
Maybe start a small croudfunding campaign for the hardware.
ReplyDeleteI would spend spend something.
BMQ is really nice the compile performance is still better even with the lower geekbench score.
Maybe I should ran some more benchmarks or some real software to test if this is the only regeression.
Here is a benchmark from user last year. https://openbenchmarking.org/result/1910077-AS-KERNELPAT43
DeleteThe benchmark is not bad for intel hardware.
Alfred Chen'S Blog: Bmq V5.5-R2 Release >>>>> Download Now
ReplyDelete>>>>> Download Full
Alfred Chen'S Blog: Bmq V5.5-R2 Release >>>>> Download LINK
>>>>> Download Now
Alfred Chen'S Blog: Bmq V5.5-R2 Release >>>>> Download Full
>>>>> Download LINK Rr