Skip to contentSkip to conversation

Kernel 2.6 with X (xorg) 4.4 (eats more CPU power)

DISCUSSION Joe Hsu 3 participants 5 messages Dec 1, 2004 – Dec 2, 2004

lore
Share & tools
lore

Joe Hsu

author
#0 lore raw

Dear All:
I've tried libXv to open an Video Overlay port and
XvShmPutImage for 60 frames per second. Each frame is at a size of 800x600 using format YUYV (YUV2). Before each XvShmPutImage, I copy 800x600x2 bytes of non-constant data to XvImage->data. (No user interactive UI)

And I found something interisting happened. In pentium 4
3.0G machine and linux kernel 2.6, X and my program total consumes 5% of cpu resource.

But in pentium 4 2.xG or below, it would consume 10% or
more of CPU resource. (If you try this with XFree86 4.2 and pentium 1.xG machine, it would consume 30% or more of cpu resource at a peak.)

In contrast, I've tried Kernel 2.4 with same X, same
program, and same machine. It consumes almost zero of CPU resource( no matter it runs on a P4 1.xG or P4 3.0G and no matter it runs on 4.4 or 4.2 X-server).

Same phenomenon happened when I ran 4 mpeg4 playback
programs (each 320x240, 30 frames per second, no scaling). It seems that these programs and X consume almost zero of CPU power when the KERNEL HZ is 100. (I've
tried Robert Love's variable HZ patch to kernel 2.4 and change HZ to 1000........Same phenomenon as 2.6)

Could any one explain why??? Thanks.
(I wish to be personally CC'ed the answers/comments posted to the list in response to your posting 'cause I do not subscribe to this mailing list.)

#1 lore raw
Quoted 31 lines · Joe Hsu

And I found something interisting happened. In pentium 4
3.0G machine and linux kernel 2.6, X and my program total consumes 5% of cpu resource.

But in pentium 4 2.xG or below, it would consume 10% or
more of CPU resource. (If you try this with XFree86 4.2 and pentium 1.xG machine, it would consume 30% or more of cpu resource at a peak.)

In contrast, I've tried Kernel 2.4 with same X, same
program, and same machine. It consumes almost zero of CPU resource( no matter it runs on a P4 1.xG or P4 3.0G and no matter it runs on 4.4 or 4.2 X-server).

Same phenomenon happened when I ran 4 mpeg4 playback
programs (each 320x240, 30 frames per second, no scaling). It seems that these programs and X consume almost zero of CPU power when the KERNEL HZ is 100. (I've
tried Robert Love's variable HZ patch to kernel 2.4 and change HZ to 1000........Same phenomenon as 2.6)

Could any one explain why??? Thanks.
(I wish to be personally CC'ed the answers/comments posted to the list in response to your posting 'cause I do not subscribe to this mailing list.)

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#3 lore raw
In reply to Joe Hsu
Quoted 16 lines · Joe Hsu

Dear All:
I've tried libXv to open an Video Overlay port and
XvShmPutImage for 60 frames per second. Each frame is at a size of 800x600 using format YUYV (YUV2). Before each XvShmPutImage, I copy 800x600x2 bytes of non-constant data to XvImage->data. (No user interactive UI)

And I found something interisting happened. In pentium 4
3.0G machine and linux kernel 2.6, X and my program total consumes 5% of cpu resource.

But in pentium 4 2.xG or below, it would consume 10% or
more of CPU resource. (If you try this with XFree86 4.2 and pentium 1.xG machine, it would consume 30% or more of cpu resource at a peak.)

Quoted 4 lines

In contrast, I've tried Kernel 2.4 with same X, same
program, and same machine. It consumes almost zero of CPU resource( no matter it runs on a P4 1.xG or P4 3.0G and no matter it runs on 4.4 or 4.2 X-server).

reported resource usage is an estimate based on sampling. With HZ=1000 you do get a more accurate sampling of reality.....

In addition, do check if your code (or X) doesn't do timeouts of < 10 msec; in HZ=100 kernels those always delay 10 or 20 msec, while with HZ=1000 kernels the delay becomes far more accurate.

#4 lore raw
In reply to Joe Hsu

Its just an idea but... ¿Have you configured your 2.6 kernel as a Preemptive kernel?

Cheers,
MaF

Quoted 7 lines · Joe Hsu

In contrast, I've tried Kernel 2.4 with same X, same
program, and same machine. It consumes almost zero of CPU resource( no matter it runs on a P4 1.xG or P4 3.0G and no matter it runs on 4.4 or 4.2 X-server).

Same phenomenon happened when I ran 4 mpeg4 playback
programs (each 320x240, 30 frames per second, no scaling).
It seems that these programs and X consume almost zero of CPU power when the KERNEL HZ is 100. (I've tried Robert Love's variable HZ patch to kernel 2.4 and change HZ to 1000........Same phenomenon as 2.6)

author
#2 lore raw

Kernel 2.6 with X (eats more CPU power)...with test program

In reply to Arjan van de Ven

I put my test program on internet.
http://bill-goldberg.myweb.hinet.net/monitor3.tgz

Download and extract it. And then read README.TXT. It does make difference between different KERNEL HZ. This program also shows the time it takes to refresh every 600 frames.(Roughly 10 seconds, both in kernel 2.4 and 2.6)

Quoted 28 lines · Joe Hsu
Quoted 27 lines

And I found something interisting happened. In pentium 4
3.0G machine and linux kernel 2.6, X and my program total consumes 5% of cpu resource.

But in pentium 4 2.xG or below, it would consume 10% or
more of CPU resource. (If you try this with XFree86 4.2 and pentium 1.xG machine, it would consume 30% or more of cpu resource at a peak.)

In contrast, I've tried Kernel 2.4 with same X, same
program, and same machine. It consumes almost zero of CPU resource( no matter it runs on a P4 1.xG or P4 3.0G and no matter it runs on 4.4 or 4.2 X-server).

Same phenomenon happened when I ran 4 mpeg4 playback
programs (each 320x240, 30 frames per second, no scaling). It seems that these programs and X consume almost zero of CPU power when the KERNEL HZ is 100. (I've
tried Robert Love's variable HZ patch to kernel 2.4 and change HZ to 1000........Same phenomenon as 2.6)

Could any one explain why??? Thanks.
(I wish to be personally CC'ed the answers/comments posted to the list in response to your posting 'cause I do not subscribe to this mailing list.)