Patch to enable NO_NEW_FAIR_SLEEPERS scheduler feature

[GoogleCodeExporter]

As requested on the list.

Some people have seen responsiveness improvements on desktop systems by
enabling the "NO_NEW_FAIR_SLEEPERS" option in the kernel scheduler
options.

I thought it'd be fun to try it out with Android on my FR but soon got
sick of reenabling it by hand through adb every time I turned it on so I
used this quick fix to my init.rc to get it going.  It does seem a little
snappier but it's hard to tell if it's a real effect or just confirmation
bias. ;-)

The attached patch is against the init.rc from Panicking's images but
should easily be adapted to any other image, it's just 3 extra lines. :-)

Original issue reported on code.google.com by ozch...@gmail.com on 1 Oct 2009 at 12:41

Attachments:

• scheduler.patch

[GoogleCodeExporter]

This is an enhancement, not a defect, but the web interface doesn't let me set 
that.
:-(

Original comment by ozch...@gmail.com on 1 Oct 2009 at 12:42

[GoogleCodeExporter]

Fixed that for you =) For some reason you need to be added to the project and 
get
privileges to change that.

Original comment by larlin...@gmail.com on 1 Oct 2009 at 1:49

• Changed state: Started
• Added labels: Type-Enhancement
• Removed labels: Type-Defect

[GoogleCodeExporter]

Before we add this in, I'd really like to have some indication of wheter it does
anything. Ideally we should have a standard performance test with some baseline 
data.
Then we would know if changes like this do any good.

Original comment by scarhill on 2 Oct 2009 at 5:06

[GoogleCodeExporter]

The later Cyanogen roms used BFS:
http://ck.kolivas.org/patches/bfs/
I'm running it on my G1 and it has been a massive improvement in 
responsiveness. I
think it could make a very big difference on the Freerunner also.

Original comment by kieran.f...@gmail.com on 5 Oct 2009 at 7:46

[GoogleCodeExporter]

Hmm tested BFS on my eeePC and got some small time improvments in some types of
loads, so i agree with kieran.fleming that it migth be intresting to test on the
freenode.

Original comment by bla...@gmail.com on 23 Oct 2009 at 12:22

[GoogleCodeExporter]

I have CaffeineMark3 installed.  It's a sort of benchmarking tool.  Would that 
be a 
useful measure?  I don't know that it's designed to test scheduling though...

Original comment by montg...@gmail.com on 24 Oct 2009 at 4:13

[GoogleCodeExporter]

I think you want to measure latency more than sheer performance, so perhaps
latencytop might help more there ?  It depends on having a kernel build with
CONFIG_LATENCYTOP though, and probably would need to be ported to Android in 
its own
right.

Original comment by ozch...@gmail.com on 24 Oct 2009 at 11:50