Re: [PATCH] sched/fair: use dst group while checking imbalance for NUMA balancer

2020-09-21 Thread Jirka Hladky
Resending - previously, I forgot to send the message in the plain text mode. I'm sorry. Hi Mel, Thanks a lot for looking into this! Our results are mostly in line with what you see. We observe big gains (20-50%) when the system is loaded to 1/3 of the maximum capacity and mixed results at the

Re: [PATCH] sched/fair: use dst group while checking imbalance for NUMA balancer

2020-09-21 Thread Mel Gorman
On Thu, Sep 10, 2020 at 11:50:04PM +0200, Jirka Hladky wrote: > 1) Threads stability has improved a lot. We see much fewer threads > migrations. Check for example this heatmap based on the mpstat results > collected while running sp.C test from the NAS benchmark. sp.C was run > with 16 threads on

Re: [PATCH] sched/fair: use dst group while checking imbalance for NUMA balancer

2020-09-10 Thread Jirka Hladky
Hi Hilf and Mel, thanks a lot for bringing this to my attention. We have tested the proposed patch and we are getting excellent results so far! 1) Threads stability has improved a lot. We see much fewer threads migrations. Check for example this heatmap based on the mpstat results collected

Re: [PATCH] sched/fair: use dst group while checking imbalance for NUMA balancer

2020-09-07 Thread Srikar Dronamraju
> > > > On Mon, Sep 07, 2020 at 07:27:08PM +1200, Barry Song wrote: > > > Something is wrong. In find_busiest_group(), we are checking if src has > > > higher load, however, in task_numa_find_cpu(), we are checking if dst > > > will have higher load after balancing. It seems it is not sensible to

Re: [PATCH] sched/fair: use dst group while checking imbalance for NUMA balancer

2020-09-07 Thread Mel Gorman
t.com; pet...@infradead.org; juri.le...@redhat.com; > > vincent.guit...@linaro.org; dietmar.eggem...@arm.com; > > bseg...@google.com; linux-kernel@vger.kernel.org; Linuxarm > > ; Mel Gorman ; > > Peter Zijlstra ; Valentin Schneider > > ; Phil Auld ; Hillf Danton > >

RE: [PATCH] sched/fair: use dst group while checking imbalance for NUMA balancer

2020-09-07 Thread Song Bao Hua (Barry Song)
t; bseg...@google.com; linux-kernel@vger.kernel.org; Linuxarm > ; Mel Gorman ; > Peter Zijlstra ; Valentin Schneider > ; Phil Auld ; Hillf Danton > ; Ingo Molnar > Subject: Re: [PATCH] sched/fair: use dst group while checking imbalance for > NUMA balancer > > On Mon, Sep 0

Re: [PATCH] sched/fair: use dst group while checking imbalance for NUMA balancer

2020-09-07 Thread Mel Gorman
On Mon, Sep 07, 2020 at 07:27:08PM +1200, Barry Song wrote: > Something is wrong. In find_busiest_group(), we are checking if src has > higher load, however, in task_numa_find_cpu(), we are checking if dst > will have higher load after balancing. It seems it is not sensible to > check src. > It

[PATCH] sched/fair: use dst group while checking imbalance for NUMA balancer

2020-09-07 Thread Barry Song
Something is wrong. In find_busiest_group(), we are checking if src has higher load, however, in task_numa_find_cpu(), we are checking if dst will have higher load after balancing. It seems it is not sensible to check src. It maybe cause wrong imbalance value, for example, if dst_running =