as vs cfq vs deadline

as vs cfq vs deadline

[Paolo Ciarrocchi]

Hi all,
I've run a 'let's check if it does not skip frames test'.
kernel is 2.5.62-mm2.

** How I performed the test **
startx
I've opened to xterminal

terminal1
[test@frodo test]$ glxgears

terminal2
./dbench 16

Following the results:

** booted with scheduler=as ** 
Throughput 19.8092 MB/sec (NB=24.7615 MB/sec  198.092 MBit/sec)  16 procs

[test@frodo test]$ glxgears
609 frames in 5.0 seconds = 121.800 FPS
673 frames in 5.0 seconds = 134.600 FPS
676 frames in 5.0 seconds = 135.200 FPS
591 frames in 5.0 seconds = 118.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
592 frames in 5.0 seconds = 118.400 FPS
-- Here dbench starts --
591 frames in 7.0 seconds = 84.429 FPS
423 frames in 5.0 seconds = 84.600 FPS
422 frames in 5.0 seconds = 84.400 FPS
507 frames in 5.0 seconds = 101.400 FPS
602 frames in 5.0 seconds = 120.400 FPS
676 frames in 5.0 seconds = 135.200 FPS
591 frames in 5.0 seconds = 118.200 FPS
592 frames in 5.0 seconds = 118.400 FPS
422 frames in 9.0 seconds = 46.889 FPS
423 frames in 5.0 seconds = 84.600 FPS
507 frames in 5.0 seconds = 101.400 FPS
507 frames in 5.0 seconds = 101.400 FPS
507 frames in 5.0 seconds = 101.400 FPS
422 frames in 5.0 seconds = 84.400 FPS
338 frames in 5.0 seconds = 67.600 FPS
254 frames in 7.0 seconds = 36.286 FPS
338 frames in 8.0 seconds = 42.250 FPS
507 frames in 10.0 seconds = 50.700 FPS
84 frames in 5.0 seconds = 16.800 FPS
-- dbench stops here --
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
X connection to :0.0 broken (explicit kill or server shutdown).



** booted with scheduler=cfq **
Throughput 24.3661 MB/sec (NB=30.4577 MB/sec  243.661 MBit/sec)  16 procs

[test@frodo test]$ glxgears
689 frames in 5.0 seconds = 137.800 FPS
675 frames in 5.0 seconds = 135.000 FPS
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
591 frames in 5.0 seconds = 118.200 FPS
-- dbench starts here --
507 frames in 5.0 seconds = 101.400 FPS
338 frames in 5.0 seconds = 67.600 FPS
507 frames in 5.0 seconds = 101.400 FPS
616 frames in 5.0 seconds = 123.200 FPS
507 frames in 5.0 seconds = 101.400 FPS
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
591 frames in 5.0 seconds = 118.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
423 frames in 6.0 seconds = 70.500 FPS
338 frames in 8.0 seconds = 42.250 FPS
422 frames in 5.0 seconds = 84.400 FPS
169 frames in 7.0 seconds = 24.143 FPS
507 frames in 16.0 seconds = 31.688 FPS
338 frames in 5.0 seconds = 67.600 FPS
-- dbench stops here --
676 frames in 5.0 seconds = 135.200 FPS
592 frames in 5.0 seconds = 118.400 FPS
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
X connection to :0.0 broken (explicit kill or server shutdown). 


** booted with scheduler=deadline **
Throughput 28.7021 MB/sec (NB=35.8776 MB/sec  287.021 MBit/sec)  16 procs

[test@frodo test]$ glxgears
687 frames in 5.0 seconds = 137.400 FPS
589 frames in 5.0 seconds = 117.800 FPS
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
-- dbench starts here --
423 frames in 5.0 seconds = 84.600 FPS
591 frames in 5.0 seconds = 118.200 FPS
676 frames in 5.0 seconds = 135.200 FPS
338 frames in 5.0 seconds = 67.600 FPS
507 frames in 5.0 seconds = 101.400 FPS
676 frames in 5.0 seconds = 135.200 FPS
592 frames in 5.0 seconds = 118.400 FPS
338 frames in 5.0 seconds = 67.600 FPS
253 frames in 9.0 seconds = 28.111 FPS
169 frames in 7.0 seconds = 24.143 FPS
507 frames in 15.0 seconds = 33.800 FPS
169 frames in 5.0 seconds = 33.800 FPS
-- dbench stops here --
676 frames in 5.0 seconds = 135.200 FPS
676 frames in 5.0 seconds = 135.200 FPS

I can not see any difference in the results above.

I guess the test I've run it is not a good test, isn't it ?

Comments/suggesiontions ?

Ciao,
		Paolo
		
-- 
______________________________________________
http://www.linuxmail.org/
Now with e-mail forwarding for only US$5.95/yr

Powered by Outblaze

Re: as vs cfq vs deadline

[Kevin Puetz]

Paolo Ciarrocchi wrote:

> Hi all,
> I've run a 'let's check if it does not skip frames test'.
> kernel is 2.5.62-mm2.
> 
> ** How I performed the test **
> startx
> I've opened to xterminal
> 
> terminal1
> [test@frodo test]$ glxgears
> 
> terminal2
> ./dbench 16
> 
> Following the results:
> 

<snip>

> I can not see any difference in the results above.
> 
> I guess the test I've run it is not a good test, isn't it ?

a) you're only looking at glxgears average framerate, so you've not measured
whether it drops frames at all, and 
b) glxgears does no disk i/o at all, so it shouldn't care a bit what the I/O
scheduelr is doing (other than how much cpu the kernel is using, which
shouldn't be all that high)

> Comments/suggesiontions ?
> 
> Ciao,
> Paolo
>