Re: [Alsa-user] new source of MIDI playback slow-down identified - 5a03b051ed87e72b959f32a86054e1142ac4cf55 thp: use compaction in kswapd for GFP_ATOMIC order > 0

[Andrea Arcangeli <aarcange@redhat.com>]

Hi Arthur,

On Thu, Feb 24, 2011 at 12:10:45PM +1030, Arthur Marsh wrote:
> OK, with kswapd-high_wmark + compaction-kswapd-3 + Mel's patch (with the 
> compaction initialisation fix), MIDI playback is fine.
> kswapd0 CPU can very occasionally hit equal highest (17 percent was the 
> highest I noticed, but is generally below the top 4-5 processes and less 
> than 10 percent when working, dropping to around 0.3 percent when swap 
> activity has subsided). This was with loading KDE 3.5.10, konversation, 
> aptitude -u, icedove and iceweasel with several dozen tabs in addition 
> to aplaymidi.

That sounds good... so maybe we don't have to backout compaction out
of kswapd and the new kswapd+compaction logic is working fine without
overloading the system like the older logic did.

Ok so tomorrow I'll get all results on these 3 kernels (
compaction-kswapd-3+compaction_alloc_lowlat-2 vs
compaction-no-kswapd-3+compaction_alloc_lowlat-2 vs
compaction_alloc_lowlat2) on network server load, where throughout is
measured in addition to latency. Then we'll have a better picture to
decide. If throughput is decreased in any measurable way I still
prefer compaction-no-kswapd and to rethink at this later without
hurry. The network load with jumbo frames from e1000 is very good at
stressing compaction+kswapd so there should be no surprises if
throughput and latency are ok with the new compaction kswapd logic.

I suggest you to keep running with this combination
(compaction-kswapd-3 + Mel's patch with the initialization fix that I
sent you + kswapd-high_wmark) if you can, to be sure it works fine for
you.

Thanks a lot for now!
Andrea