The v2 Rotary Interactivity Favor Scheduler

[3766691]

Ok now is the correct one

[3766691]

V3-performance

I have just post a few screenshot to show the interactive performance of RIFS v3

http://code.google.com/p/rifs-scheduler/downloads/detail?name=Screenshot-2012-05-27-2.png

http://code.google.com/p/rifs-scheduler/downloads/detail?name=Screenshot-2012-05-27.png

Have a look. ;-)

[3766691]

Would you please decide whether we need
int preempt; <--- in struct task_struct in include/linux/sched.h
or if the corresponding sequence in check_timeslice_end in rifs.c
if(p->state != TASK_RUNNING)
p->prio --;
else
p->preempt = 0;
is correct. I'm no real programmer but I'm giving it a try with the first choice. (Now compiling :-) ) I hope that it's not the wrong choice. :-?

Best regards,
Diana

The one you have(which cause compilation failure) is v2(Sorry for my fault)
Please download the new v3 at http://rifs-scheduler.googlecode.com

[3766691]

+1 github is a good option imho

btw: on subjective use, RIFS is faster than BFS (I've been comparing both for the last three days). stuff like rendering pdf pages are faster and video plays smoother while compiling. I'll have to measure dropped frames to be sure though.

Chromium slows down sometimes (pseudo-freezes). I don't know if this is beeing caused by the scheduler (is chromium getting RR??) or by flash or by something else.

Thanks for comment :-)
I m preparing the Github site these days.

From now on, I would not rebase RIFS v2 to 3.4.x tree.
RIFS v3 for 3.4.x is now not released because everyday I received a*A LOT OF* bug report about 3.4.x kernel. I would rebase it when it is stabled.

I have mentioned that V3 uses a different algorithm from V2

EVERYONY, TRY IT. :-)

[Diana]

O.k. -- I tried v3 (the correct one ;-) ) with kernel 3.3.7 plus the two BFQ patches and I'm getting severe problems:
o startup takes very long time before X
o KDE loading after login takes ages and mouse cursor is sluggish during any I/O operation during and after loading
o mouse cursor recovers after moving it for a few seconds -- but only for a few seconds

I've noticed in top that all tasks have a PR > 100, I've never seen that with BFS.
The kernel config remained the same as with BFS except for choosing RIFS instead.
There do not appear any error messages in the logs and I experienced no data loss.

My machine is quite "vintage" but BFS is running fine (no comparison to RIFS in the current state!!!):
PIII-1400 Tualatin (single core, single CPU), VIA Apollo PRO133x AGP 4x, IDE disks & DVD, AMD Radeon HD 4350 (RV710) @ AGP 4x with Gallium 0.4, 2GB SDRAM 133. openSUSE 12.1, KDE 4.8.3, MESA 8.1-devel.

I hope your RIFS scheduler is not limited to modern high end multi-core machines to be usable. In the current state I really can't use it.


Best regards,
Diana

[3766691]

O.k. -- I tried v3 (the correct one ;-) ) with kernel 3.3.7 plus the two BFQ patches and I'm getting severe problems:
o startup takes very long time before X
o KDE loading after login takes ages and mouse cursor is sluggish during any I/O operation during and after loading
o mouse cursor recovers after moving it for a few seconds -- but only for a few seconds

I've noticed in top that all tasks have a PR > 100, I've never seen that with BFS.
The kernel config remained the same as with BFS except for choosing RIFS instead.
There do not appear any error messages in the logs and I experienced no data loss.

My machine is quite "vintage" but BFS is running fine (no comparison to RIFS in the current state!!!):
PIII-1400 Tualatin (single core, single CPU), VIA Apollo PRO133x AGP 4x, IDE disks & DVD, AMD Radeon HD 4350 (RV710) @ AGP 4x with Gallium 0.4, 2GB SDRAM 133. openSUSE 12.1, KDE 4.8.3, MESA 8.1-devel.

I hope your RIFS scheduler is not limited to modern high end multi-core machines to be usable. In the current state I really can't use it.


Best regards,
Diana

prio in RIFS is the actual value.So the real prio is actually prio - 120.
I have to spend more experiment on V3, but on my box(Just a Dual Core) it works very smooth even with big stress. On my box, when I am running BFS, it hang my cursor and music.

It seems to have problem with something from your description. I will try to look at it.
Thanks for giving feedback.

[3766691]

The bug fix of RIFS v2 has released.Also I have released BFS-O1,it can improve the turnaround time of BFS by O(1) lowest deadline task searching and it won't affect the behavior of BFS.

RIFS v4 are planning.

[Diana]

No advantage with BFS-O1

Hi Chen,
just want to let you know, that your BFS-O1 patch does not bring any advantage to BFS 420 with kernel 3.3.7. It's making the contrary: I get more video playback hickups (in vlc) when compiling the kernel at the same time. Pure BFS 420 also shows hickups - but not that often.

Diana

[DeiF]

Hi Chen.
Thank you for your effort.

I think you should team up with Con Kolivas and you two together bring us a super-next-gen scheduler.

[3766691]

Hi Chen,
just want to let you know, that your BFS-O1 patch does not bring any advantage to BFS 420 with kernel 3.3.7. It's making the contrary: I get more video playback hickups (in vlc) when compiling the kernel at the same time. Pure BFS 420 also shows hickups - but not that often.

Diana

Sorry I am late to give you reply
BFS-O1 is already failed, so please throw it away.

I am just focusing on RIFS v4 now. RIFS v3 is not responsive on some machine.Actually RIFS V2-bugfixed perform well on any machine and it give well response to the foreground(RIFS V3 does so), but it will cause a task switching lag.