AMD64 system slow/unresponsive during disk access (Part 2)

[mamunata]

[F-0_ICE]

don't know if the issue is related but i get high cpu usage when writing data to hard drive.

i have a sata 1 motherboard/drive and for some reason i only have udma5 available.

i have two questions:
when writing data to my drive what is normal cpu usage?
and why don't i have anything higher than udma5 available?

i have been trying to find an answer to these on my own for a while now so i would appreciate any insight.
_________________
~AMD64
AMD: Athlon64 X2 3800+
2G PC3200
ATI: RADEON HD 4350
Linksys: WMP54G

True Knowledge is Best Acquired Through Experience.

[sidamos]

OK, the problem is solved for me!

I have AMD64X2, IDE disks, running 32 bit Gentoo. My system was pretty unresponsive (scrolling in Firefox for example paused a lot) when running mplex or dvdauthor with source and dest files on the same disk, despite the fact that I used nice and ionice.

I tried:
- 2.6.29 with CFQ
- 2.6.29 with AS
- 2.6.30 with CFQ
- 2.6.30 with AS
- 2.6.30 with DEADLINE

All with timer 100 Hz. The best combination (with only a little "feeled" improvement) was 2.6.29 with CFQ.

Now I tried:
- 2.6.31-zen5, BFS, BFQ, 1000 Hz

And everything is good now! While running mplex or dvdauthor or copying large files over Gigabit, I can make normal use of other programs with almost no hiccups!

Of course I don't know, which of the changes fixed it:
- 2.6.31
- ZEN patches
- 1000 Hz
- BFS
- BFQ

[mamunata]

[d-fens]

[kernelOfTruth]

[snipe]

[albright]

using
System uname: Linux-2.6.31-zen7-x86_64-Intel-R-_Core-TM-_i7_CPU_920_@_2.67GHz-with-gentoo-1.12.13

with BFS, BFQ andSQLB

I read somewhere that in kde 4.3 strigi works OK so I started it up.

System bogged right down when indexing began - mouse very slow to
respond, painful window movement etc Real ugly

system began to work properly in a minute (indexing still going on in
the background of course)

Has anyone seen this problem with strigi indexing?
_________________
.... there is nothing - absolutely nothing - half so much worth
doing as simply messing about with Linux ...
(apologies to Kenneth Graeme)

[Hallerich]

Hey all,

I suffered a lot from this problem since I switched to amd64. Now I just updated to the latest gentoo-sources (2.6.32 released yesterday), because I was curious to see if it had anything to do with the replacement of the pdflush threads (which I noticed to always stall my system when I was copying large amounts of data) with a new per backing device based writeback (see this summary on kernelnewbies.org).

Now I am happy to say that this update seems to solve all the problematic behaviour, with the system now running really smooth even with heavy IO in the background

So I expect this might help others who are affected by this as well!
Good luck guys

[kernelOfTruth]

thanks Hallerich !

there's also some nice info on the new per backing device writeback (the improvements or replacement of pdflush you wrote about) which really looks like it could fix it (currently I'm still running 2.6.31 due to fglrx so I don't know whether it'll fix it for me)
_________________
https://github.com/kernelOfTruth/ZFS-for-SystemRescueCD/tree/ZFS-for-SysRescCD-4.9.0
https://github.com/kernelOfTruth/pulseaudio-equalizer-ladspa

Hardcore Gentoo Linux user since 2004

[Sujao]

[kernelOfTruth]

[Sujao]

[zehnan1]

I can cofirm much better performance with BFS, BFQ, SQLB combo.

For me a test is considered the following: try to do things while unraring 8GB large file, running a game in windowed mode and playing xvid movie in smplayer. Kwin desktoip effects were on. Game ran at 130 fps (normally 155fps) without hicups, movie ran fluid without framedrops and konqueror starts in 6 seconds, firefox in 15. Desktop effects were fluid without hicups except for the resizing which doesn't work anyway.

If I try comparing this to the previous situation with CFS-CFQ; the game would would have systematic frame drops, movies exhibit similar hicup problems, the konqueror wouldn't start for half a minute and firefox wouldn't start until the rar finished and if it did I probably couldn't type in the web address because of the horrible lag. Desktop effects would become usuable.

System is quadcore with 2gb ram, kernel 2.6.31-zen7 ZEN SMP PREEMPT.

[PaulBredbury]

The pf-kernel includes BFQ and BFS, as an alternative to the zen patchset. In case this is useful for people to try.

Edit: Updated URL.

[teapot]

I have tried the new 2.6.32 kernel that comes with a new low-latency mode for CFQ enabled by default and I can confirm a huge performance increase (latency-wise , that is ).

My previous experience with 64-bit Linux on this computer was just terrible. The system "only" has 1Gb of RAM , which with firefox and a few tabs is used up pretty quickly. Let me just say that even light disk-swapping on top of a system with the all to well known I/O-induced latency problem of the Linux kernel made things completely awful.

But I am still confused and not at all convinced that this problem is solved once and for all. It cannot have been entirely CFQ's fault. The first thing I tried when I first noticed latency problems was to use another I/O-scheduler , but with little improvement.
Why does this new CFQ-mode suddenly make things a lot better ?

[Yzzyx]

[quote="Sujao"]

[DigitalCorpus]

[haarp]

[pste]

First I have to say, I may be repeating something already mentioned since this is a quite extensive thread...

I seem to have the same problem (Intel Core2Duo = amd64 as arch, currently gentoo-sources-2.6.33) and this IO-sluggishness is really annoying me. I think I saw something about a kernel bug posted, any progress from there? What I'd like to contribute is some hopefully useful info (if its not already posted, that is).

The problem becomes really evident when having a vpn-tunnel going! My log is filled with occasional

[lagalopex]

2.6.33-ck1 bahaves pretty nice for me. But most hickups happend after some uptime, so lets wait and pray

[Mekoryuk]

I have seen next to zero improvement, and I've been using BFS/BFQ/SQLB for several kernel releases. I'm currently on zen-2.6.31-r5. Heavy I/O still grinds my system to a halt. Just today I tried to copy a multi-gig file to a memory stick and my computer was unusable until it finished. Preallocation of multi-gig torrent files like Linux ISOs also results in the system becoming effectively unusable until the process finishes, and any emerges installing large amounts of data (like game files) does it too.

About the only improvement I've seen is gameplay in Quake Wars doesn't hiccup any more.

I remember reading somewhere there was an issue with my HDD's (Seagate Barracuda) NCQ implementation, so I'm beginning to think that I'll never see decent I/O performance until I buy a new HDD. :C

[Yamakuzure]

All problems with responsiveness are gone since I use gentoo-sources-2.6.32* ... I don't know why, though.
_________________
Edited 220,176 times by Yamakuzure

[matze_na]

I wouldn't go as far as to say the problem has totally vanished, but it certainly got a lot better over the last months.
Especially the move to 2.6.33-zen1 with BFS and CFQ (since BFQ isn't available yet) was quite an improvement over 2.6.32-zen with BFS and BFQ.
At the moment, I am totally happy with my system.. Ok, there sure is a little slowdown when moving large files I guess, but it doesn't really bother me now. It's way better than a few months ago.

[PaulBredbury]

BFQ is back in pf-kernel, for anyone interested to try it.