BMQ v5.5-r3 is released with the following changes
1. Implement sched_exec(). Take this change to balance cpu loading as task has very load memory&llc footprint at this point of time.
2. Optimize sg_balance_check(). Cut off one of the two sg balance checking branch based on analysis.
3. Introduce per cpu sched_cpu_llc_mask. And also deploy bottom up topology cpumask setup, used more code but hopefully it is the final version suitable for all major CPUS.
These are major code changes for the found-out during spring code clean-up. I believe that all for 2020's Spring code clean-up and it will be the final release for 5.5 kernel release.
After more than one year of deployment(officially release today one year ago), the main features of BMQ are stable and in a good shape(IMO). So from next kernel release, I'd like to start picking up some previous low priority work items.
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
Hi,
ReplyDeleteIm running this patch now and tested vs vanilla with a threadripper(3970x) as you recommended:
https://openbenchmarking.org/result/2003119-HV-1910077AS28
Something seems up with the context switching benchmark from stress-ng.
Running on arch-x64 (i5-2520M) and arch-p4 (Celeron 353 UP). Fyi, there's a compile warning (and a few 'notes'):
ReplyDeletehttps://gist.github.com/jeremywh7/d78c68085bfb5aacc905ad2bd1f8dbc0
Confirmed. The waning happened in 32bit system. Fix will be deliver in next release.
DeleteBTW, this warning can be safely ignore.
I forgot to mention that this is running on Ryzen@home and i7@work since ages and works well.
ReplyDeleteThanks, Alfred!
BR, Eduardo