Hi
one big doubt...
I am looking Linux I/O scheduler.....

In linux 2.6.12.3 we are having 4 I/O schedulers.....

i am running multimedia application so I thought i will go for DEADLINE I/O SCHEDULER..

I tryed tuning of deadline i/o scheduler.....
after tuning of deadline i/o scheduler,i just switched from deadline I/O scheduler to Anticipatory scheduler........

ANTICIPATORY is giving good throughput compared to DEADLINE scheduler....!!!!!!!!

I am really VERY VERY VERY confused........

my doubts are

1)for Soft realtime application (multimedia) which is better i.e DEADLINE or ANTICIPATORY????????

2)what sort of application we should select DEADLINE and ANTICIPATORY


Please help me, i am in big confussion



Kotresh

deadline is terrible
anticatory is sort of ok

**you should use CFQ (DEFAULT_CFQ)** as your default i/o schedular

for mutimedia this is the way we do the kernel
apply this patch
http://people.redhat.com/mingo/realtime-preempt/
then in the config choose
Default I/O scheduler (CFQ)
and "Complete Preemption" as the preemption model
use 1000 nanoseconds on the high res timer and i use a tickless system.
(tickless most likely doesn't effect realtime preemption but can keep the system from running hot all the time) killed my last cpu pretty quick ;-)
make sure you choose Extended RTC operation
and of course Use RTC as default sequencer timer

thanks for u r kind reply studioj

but I am using 2.6.12.3 in that anticipatory is default one(NOT CFQ).

CFQ became default scheduler after 2.6.18 I think. and u have mentioned some patch also but that is for 2.6.20.

please mention me any patch is there for 2.6.12 ???

and i want to know which is beter in 2.6.12 kernel.




thanks


Regards
Kotresh

CFQ exists in 2.6.12 so just choose that i'm not sure about the options for default but i bet that is there as well. if you can't set it as default just choose only CFQ.

you are right the older realtime patches seem to have been taken down -- the older patches often had problems so for such an old kernel i guess i would skip that part.

it also helps quite a bit to alter the kernel Makefile after it is built and set optimization flags for your architecture.

Example:

i got a 13% overall kernel speed increase by using
-march=athlon-xp -O3 -ffast-math
instead of the default optimization flags

thank you studioj

ya, i think kernel opimization is one more way to increase speed...

But still i want to do both ie
1)Section of right scheduler and tuning parameters of that scheduler to give best performance for vedio applications
2)Kernel optimization....

but one thing, i am struck up with 2.6.12....

plz tell me in wat manner CFQ is better because deadline perfomance
is better than CFQ (in 2.6.12) in terms of throughput..

plz tell me some ways to decide which scheduler to select..

this is how i understand it anyway.

CFQ is designed specifically to maintain minimal worst case latency on each and every IO read and write. so it is made specifically to prevent dropped frames etc in multimedia apps.

with anticipatory for instance better throughput for reads can cause writes to wait for vm to fill with dirty writes before the actual write will kick in. etc.

this causes CFQ to appear to use less bandwith in simple tests but in real world multimedia works better and safer. the only thing you care if you run CFQ is the worst case
latency of all reads and writes, never the throughput, 128k/sec is more than enough as
far as you never wait 2 seconds before you can get the next 128k no matter whats going on.

now in a way anticipatory is a form of deadline and in theory deadline should work ok for multimedia but this would require very carefull use of the correct i/o structures by the programmers. BUT if the programmers use RT disk i/o and you use a deadline type schedular
you are going to get large bursts of disk i/o that will get in the way of the overall picture.

hope that helps explain it somewhat