Siduction Forum

Siduction Forum => Scripting & Kernelhacking => Topic started by: ralul on 2012/06/14, 01:30:45

Title: linux-3.4 better with bfs - a little
Post by: ralul on 2012/06/14, 01:30:45

I am just testing the new recently bug fixed BrainFuckScheduler:
http://ck-hack.blogspot.de/2012/06/bfs-0423-34-ck2.html

This runs a little better than mainline unpatched Linux-3.4 kernels. In the past patching linux-3.2 with BFS was clearly a lot better for the desktop. Towo did this for siduction.

But this has narrowed now, because mainline CFScheduler has been improved a lot recently. But still I feel "a little" better using BFS patches for my Desktop system! Though I would not use it if my machine ran many server tasks, because I know throughput could be "a little" less...

Title: Chen developes RIFS
Post by: ralul on 2012/06/18, 20:43:38

If this benchmark is truly measuring differences, this shines! Perhaps we will end up having a better scheduler in mainline Linux! The author, Chen, claims to keep his coding in the frames the modular design of mainline CFS provides (edit: looking at his source this doesnt seem to be true).

My try to compile fails missing an include stats.h :(

Title: Chen developes RIFS
Post by: ralul on 2012/06/19, 19:05:55

I testet Linux-3.4.3-with-RIFSv1 and had 3 audio interrupts watching an avi file. This was medium load: a parallel kernle compile and another interruption when ktorrent in action. Perhaps above ultra latency statistic of RIFS is significant with high load having 8 cpus in action.

But I only have core2 and medium load.
I am back using BFS-423 as my scheduler...

Title: Chen developes RIFS
Post by: ralul on 2012/06/21, 13:36:45

I am just running a new version of the RIFS Scheduler:
http://rifs-scheduler.googlecode.com/files/RIFS.ES-v1-low-spec-kernel3.4.x
This is a version running smoothly for my system!
As Chen said, it is not punishing sleepy threads as much as before.

I think with his statistics Chen wants to show some developer specifics. LogN efficient algo to create or insert new processes or something like that.

Title: RE: Chen developes RIFS
Post by: dibl on 2012/06/21, 15:51:14

@ralul thanks for doing this.  I for one am very interested to learn whether this will give a more responsive desktop experience.

Title: RE: Chen developes RIFS
Post by: jaegermeister on 2012/06/23, 14:04:52

This is a project which was already suggested at sidux times, really really interesting. A desktop-oriented system like siduction ought to have a desktop-oriented kernel, way more than vanilla debian, which has instead to remain a one-good4all system.

Is the patched kernel you're running already long in production? Any glitches? If it runs smooth it could be a great candidate for inclusion in next kernels...

Title: RE: Chen developes RIFS
Post by: ralul on 2012/06/23, 14:54:38

@jaegermeister
I recently had a wlan disconnect:
Because I watched the Danzig soccer play for long using dvb-T, an internet wlan0 timeout?

This is Chen discussing and searching for testers:
http://phoronix.com/forums/showthread.php?71658-RIFS-ES-Linux-Kernel-Scheduler-Released/page4

Title: RE: Chen developes RIFS
Post by: jaegermeister on 2012/06/23, 15:44:02

It was due to Greece's temporary 1-1 :D