Kernel Sluggish

[TrentonAdams]

Hi Guys,

I have had a sluggish kernel whenever I upgrade my kernel. My last stable one was 2.6.17-gentoo-r8. I tried 2.6.25-r7 and 2.6.27-r7. When I tried 2.6.25, I figured it'd be fixed soon enough, so I just waited to try a newer version. Now I'm starting to think this is an issue that will not be resolved. Also note that it seems that 2.6.27 is a little better than I remember 2.6.25 was. 2.6.25 was unbearable.

Here's my CPU info. If you need anything else, let me know. Do you want the kernel configuration? I will post the scheduling configurations, and the cpu configurations as well...

[Dairinin]

Try:

[TrentonAdams]

I'll try those. But, I did try a partially preemptible kernel already (desktop), and it was WAY worse. That is why I switched it to PREEMPT=n.

[TrentonAdams]

Dairinin,

Could you explain to me what each of those options really does?

I'm running them now, and I'm going to see if it makes a difference. Hopefully it will.

Thanks.

[Dairinin]

Better see kernel docs (at least what hides under <Help> button in menuconfig). Well, I'm not a kernel hacker, but AFAIK.

1. Group scheduler enables CPU resource allocation based not on PID (classics), but on other criteria (on user id by default). Was the source of "sluggish" behavior when first introduced in 2.6.24. Good for servers and multiuser enviroment, pretty useless for desktop/laptop.

2. NO_HZ Disables timer interrupts when they are not needed (tickless system). You are mobile, this can make your sys live longer

3. Preemption allows your system to switch context when higher priority event occurs (interrupt, for example) without waiting for current task to complete or time 'slice' expiration. Helps reduce latency

4. Anticipatory elevator is good, but CFQ is a new shiny io scheduler, which makes use of a process niceness for i/o and provide "fairness" for concurrent processes doing simultaneous read/writes. Also makes your life a little more comfortable.

[Hu]

What do you mean by sluggish? What operations feel slower than you expect? Is it slow for all operations, or are there particular events that are noticeably slower than others? What type of workload(s) are you running?

[albatroz]

Try with the new kernel version 2.6.28, I was also experiencing some of this sluggish. With this new version it feels much better.

[TrentonAdams]

[pappy_mcfae]

If you want help with a sluggish kernel, post the entire thing. Also, post the results of lspci -n and cat /proc/cpuinfo as well as your /etc/fstab file. While I can pretty much guess that you are suffering under the lash of the usual suspects, I'd like to see your entire .config (unedited, from the first # to the last m,n, or y).

Blessed be!
Pappy
_________________
This space left intentionally blank, except for these ASCII symbols.

[energyman76b]

don't do this:
CONFIG_HZ_1000=y (rest of CONFIG_HZ_x are turned off)

it will only drain your battery and the kernel wastes time. Try 300.
_________________
Study finds stunning lack of racial, gender, and economic diversity among middle-class white males

I identify as a dirty penismensch.

[TrentonAdams]

[TrentonAdams]

Now that I'm looking at the difference between the two kernel configurations, I'm noticing this...


CONFIG_IDEPCI_SHARE_IRQ is y on the old kernel, and not set on the new kernel.

CONFIG_IDEDMA_PCI_AUTO=y on the OLD kernel
CONFIG_BLK_DEV_IDEDMA_SFF=y on the NEW kernel
CONFIG_BLK_DEV_IDEDMA_PCI=y on the NEW

CONFIG_IDEDMA_AUTO=y on the OLD kernel and NEW kernel

Also, the CONFIG_SCSI_SATA_INTEL_COMBINED=y in the OLD kernel and NOT in the new kernel, and is not even available in the new kernel.

Anyhow, I'm thinking of trying out AHCI ATA in the SATA configuration, as ATA was completely off by default for some reason. I had even copied my 2.6.17 configuration to ".config" before doing a genkernel oldconfig, etc.

[TrentonAdams]

I don't know which SATA thing to enable, it doesn't appear to be affecting the performance at all.

[TrentonAdams]

I am also getting these...
Jan 19 16:04:48 reuben kernel: hdc: packet command error: status=0x51 { DriveReady SeekComplete Error }
Jan 19 16:04:48 reuben kernel: hdc: packet command error: error=0x54 { AbortedCommand LastFailedSense=0x05 }

And also recently, I have noticed that my system is turning the main file system into read-only mode, so I have to reboot. I'm starting to think that is a new problem with the kernel that did not happen in the previous kernel. I'm running the previous kernel now, and I'm not seeing that stuff.

[pappy_mcfae]

Sorry I took so long to get back to you. It was laundry day, which is usually also melt brain with TV day.

Anyway, I took a look at your kernel and decided it would be better to start with one of my seeds. As far as which SATA driver to use, you would want the

[TrentonAdams]

Thanks pappy. I'll try it out. I don't need to worry about drive naming, as I use LVM, and I have labeled /boot.

Is mrproper really needed? Wouldn't "clean" be enough, and for me to replace .config? I just hate waiting soo long for the compile. It's whole 6 minutes for the kernel, and 10-20 for the modules. hehe

[pappy_mcfae]

Clean can and does work, but I'm a believer in all new parts, no old comebacks. I don't even want a hint of stale object code anywhere near my kernel or its modules. I use make mrproper all the time. I've been stung by make clean once. It was bothersome, so I don't go there anymore.

How can you go wrong with pristine source code?

Blessed be!
Pappy
_________________
This space left intentionally blank, except for these ASCII symbols.

[TrentonAdams]

Besides that, I don't know what I was thinking. clean will make the code recompile anyhow, so it's still going to take a long time with that. What I usually do is ommit clean or mrproper. But, I'll try that when I get back to work. What exactly did you change in the new config? I couldn't tell because vimdiff is showing me a whole lot of changes. Was it only the stuff you mentioned? PIIX, etc.

[pappy_mcfae]

I gave you one of my seed kernels with your devices. I set my kernels to be fast, uncluttered, and functional. They are made to run a computer, and run it fast. If the user wishes to experiment with the settings, delve deeper into what's there, or analyze my settings to find the method to my madness, all the better. I hope I'm just the stepping stone for you to begin playing with your own kernels. After all, if the kernel isn't right, nothing else will be.

Usually, when I send a custom .config like that, it's because there is much that isn't right, and it's just easier for me to start with a known good base, and build upon it. Either that, or I'm just feeling lazy and don't want to go through every setting to make sure all is well.

Blessed be!
Pappy
_________________
This space left intentionally blank, except for these ASCII symbols.

[devilheart]

this is a very old (and unresolved) kernel bug
http://bugzilla.kernel.org/show_bug.cgi?id=12309

[TrentonAdams]

This is horrible. How can the kernel guys just let this go on and on? That describes my problem exactly. And, it even says that the problem started with 2.6.18, which is later than the kernel I have. I wish they would fix it already. I guess I'll have to stick with 2.6.17 with all of my Gentoo machines. I'm seeing this on my x86_64 install as well.

They should revert to a known working good, and port changes one at a time backwards.

[mikegpitt]

[devilheart]

i don't know if it has been solved. we will see what happens with 2.6.29

[TrentonAdams]

I found this too...
http://linux-ata.org/faq.html

I'm going to try it out.

[TrentonAdams]

Perhaps it would be worthwhile for the Gentoo folks to add 2.6.17 back into the portage? I noticed there is a 2.6.16 in there though.