Arch Linux

MilenKid

Member

Registered: 2013-04-21

Posts: 86

High CPU Wait on I/O

Hi,

When copying from one partition to another or doing intensive HDD operations, my system starts going really slow. It's actually lagging big time.

Let's say from ext3 to ntfs. mount.ntfs CPU usage is 40%-50%.

Is this normal? Should I change anything?

Last edited by MilenKid (2013-05-12 18:10:56)

gedgon

Member

From: Poland

Registered: 2011-01-27

Posts: 95

Re: High CPU Wait on I/O

There is no simple answer.

It's actually lagging big time.

https://bugzilla.kernel.org/show_bug.cgi?id=12309
https://bbs.archlinux.org/viewtopic.php … 69#p985069

Let's say from ext3 to ntfs. mount.ntfs CPU usage is 40%-50%.

FUSE driver, http://www.tuxera.com/products/tuxera-n … rformance/

Also depends of used tool, size and number of files. For example, FM's relaying on KDE KIO tends to be very slow and CPU intensive. https://www.youtube.com/watch?v=M_MDlPa9jeE

MilenKid

Member

Registered: 2013-04-21

Posts: 86

Re: High CPU Wait on I/O

Well, I have set dirty & background_ratio to 1 and it makes things a little better, but not how it should.

This happens no matter of FS, I use a simple dd to an ext4 f.system to test.

dirty_ratio 1
dirty_background_ratio 1

I am also using cf-kernel. Setting bfq did not made any notable difference.

CPU Wait actually goes up. Someone should fix this, it's embarrassing for Linux...

triplesquarednine

Member

Registered: 2011-04-12

Posts: 630

Re: High CPU Wait on I/O

Hi,

When copying from one partition to another or doing intensive HDD operations, my system starts going really slow. It's actually lagging big time.

Let's say from ext3 to ntfs. mount.ntfs CPU usage is 40%-50%.

Is this normal? Should I change anything?

I use the 'fix bouncing cow syndrome' patch with my kernel to get around the performance degradation (that you are describing above, and that is also discussed in the bugzilla report). the patch in question; http://pastebin.com/KHb331FQ

For me, where the my system would get painfully slow was with my SDcard/multi reader and USB hdd(s). the partition transfers and intensive operations certainly improved after using this patch, but where the benefit was extremely dramatic was with the external devices (USB/SDcard), which without using this patch, the performance was brutal...  VMware also seemed to speed up on certain operations.

i don't know what all kernel code -ck kernels/patchets touch, or if this patch is relevant to that kernel, but on mainline (3.8.x possibly other versions) it should be beneficial. I say 'should' because i am only really using realtime kernels, so that is my personal experience / point of reference. - but the patch works very well for me.

cheerz