Poor performance (unresponsive, pausing, stuttering etc.)

[Fitzcarraldo]

I've been using SL since 3.26 on the same hardware (re-installing from LiveDVD for every version up to 3.4f, and then rolling updates since then so that I am currently running all the latest versions of every package in Portage and the SL overlay).

However I have noticed with the last few kernel versions (2.6.24 onwards, if I remember correctly) a degradation in performance, all other things being more or less equal (i.e. AIGLX, Compiz-Fusion, etc. all running).

Watching a video in a window, perhaps whilst browsing the Web (or not), results in 'stuttering' i.e. short freezing of the image, and keyboard input, and the sound going "duh-duh-duh-duh-duh...". Sometimes there is just a brief single pause/freeze at seemingly random intervals whilst at other times the sound may stutter while the screen (and keyboard) is frozen. Similarly, if dragging the vertical slider in Firefox the mouse and screen freeze briefly (may be only a second or several seconds) and then continue.

If significant disk access is involved, say running a merge or imagemagick on a batch of files, or whatever, then the screen and keyboard and sound freeze (and windows may grey-out) for a while. All I can do is sit there and wait for the system to 'wake up' again.

Running htop shows me that my CPU (dual core) is nowhere near 100% loaded and neither is my 2 Gb of RAM fully used. Sometimes 257 Mb of my 517 Mb swap partition is used (I suspect that's due to the Firefox 2.0.0.17 memory leak problem), but I just can't see my laptop being overloaded.

It seems to me that the problem stems from some sort of disk read and/or write bottleneck, but I'm not sure.

This poor performance has got to point where it is actually interfering with my productivity, apart from being very annoying. I have PCs with much lower specs, running PCLinuxOS, Windows XP and Vista with better performance than my laptop running SL.

Now, I have been studying the Gentoo forums in detail. Three threads I have noticed are:

Why is my machine becoming unresponsive?

AMD64 system slow/unresponsive during disk access... (covers other CPUs too)

[solved] Slow buffered disk reads

I've read all 28 pages of the second thread and it makes very interesting reading (SL performance is also discussed -- favourably -- in that thread, by the way). Some people have suggested kernel changes, other changes in userspace. What worries me is that some very knowledgeable Gentoo users can't seem to find the root cause of the problem.

Anyway, I have yet to start trying the various suggestions in the above-mentioned threads, but wondered if any of you SL users have encountered similar problems and have performed any tuning to your SL installations (kernel or userspace) that have resulted in improved performance. I'd love to hear what you have tried, successful or not.


EDIT: I never really used the 2.6.23 kernel (I built it but immediately built the later kernel), so the degradation in performance could have been introduced in the 2.6.23 kernel rather than 2.6.24.

[dunsurfin]

I'm afraid that I've found the opposite. 3.5 x86-64 for me gives better performance than 3.4f did. My system is not fully updated as I have stuck with kernel 2.6.25-r1 (I have the latest kernel installed but am just waiting for the opportunity to do the necessary work on my nvidia drivers) but all other packages are fully up to date. I hesitate to say this (think of granny and sucking eggs!) but I wouldn't pass up the opportunity of a fresh install of a distro and wonder if your problem is because you just kept updating 3.4f rather than doing a fresh install.


Dell Dimension E521 (onboard sound turned off at BIOS)
AMD Athlon 64X2 4200+ processor
2 GB RAM 533Mhz DDR2
Nvidia Geforce 7300 LE 256MB
PCI Trust SC-5200 5.1 sound card
PCI V-Stream DVB-t DVBTV card
TSSTcorp DVD+-RW TS-H553A
Logitech S510 wireless USB Keyboard and Mouse
Broadcom BCM 4401-BO 100BASE-TX
200GB SATA hard disc

[Fitzcarraldo]

Thanks for replying.

The thing is, though, that, in effect, I do have 3.5 installed. Well, 3.5+ in fact. If all my packages and all my config files are up to date (3.5 or newer), then why would SL performance improve if I re-install? As far as I know, SL is not like Windows, which does benefit from periodic re-installation. In Linux, for example, there is no Registry to cause slow-down by becoming fragmented due to left-over entries and holes. I'll hazard a guess that installing 3.5 on my PC would result in the same performance as I'm getting now.

In all 28 pages of the second thread I mentioned earlier, no one suggested re-installing. Well, actually they did, but in the context of installing 32-bit instead of 64-bit Gentoo, as it appears that this particular problem is more likely to occur in 64-bit Gentoo (I think it was explained as being due to the cache, but can't remember now). What the posters in that thread have been doing is investigating in great detail things like the kernel config, fstab etc., as it appears, albeit not conclusively, that this is a kernel problem. Some of the posters experimented with installing earlier versions of the kernel, and several reported much better performance with earlier kernel versions. This has been my experience: performance with 2.6.22 was much better than with 2.6.26 in my case. Actually, performance under 2.6.26 feels better to me than under 2.6.25-r1.

The only other thing I didn't mention in my first post that might have an impact on what I'm experiencing is the AMD ATI fglrx driver, as the effect of video drivers on performance is notorious. For example, in the above-mentioned 28-page thread, some posters reported the Nvidia driver as a cause of badly degraded performance, and regressed to an earlier version. In my case ati-drivers-8.532 is performing slightly better than 8.522, but I have not ruled it out as a possible cause, or contributor to, the stuttering/pausing problem. But I'm still inclined to think it's a disk I/O or cache bottleneck, or perhaps a prioritisation issue of some kind.

[Fitzcarraldo]

I've added commit=60,noatime,nodiratime to my fstab (see the third Gentoo forum thread I listed earlier) to see if it makes any difference. Will report back after a fews days of evaluation.

[dunsurfin]

The thing is, though, that, in effect, I do have 3.5 installed. Well, 3.5+ in fact. If all my packages and all my config files are up to date (3.5 or newer), then why would SL performance improve if I re-install? As far as I know, SL is not like Windows, which does benefit from periodic re-installation. In Linux, for example, there is no Registry to cause slow-down by becoming fragmented due to left-over entries and holes. I'll hazard a guess that installing 3.5 on my PC would result in the same performance as I'm getting now.

I may be entirely wrong but this is my thinking; If the list of packages installed in 3.5 were exactly the same as in 3.4f then that would be the case. I haven't examined the list but what you have is 3.4f with all it's packages updated and any different packages in 3.5 would not be there.

[wolfden]

The thing is, though, that, in effect, I do have 3.5 installed. Well, 3.5+ in fact. If all my packages and all my config files are up to date (3.5 or newer), then why would SL performance improve if I re-install? As far as I know, SL is not like Windows, which does benefit from periodic re-installation. In Linux, for example, there is no Registry to cause slow-down by becoming fragmented due to left-over entries and holes. I'll hazard a guess that installing 3.5 on my PC would result in the same performance as I'm getting now.

I may be entirely wrong but this is my thinking; If the list of packages installed in 3.5 were exactly the same as in 3.4f then that would be the case. I haven't examined the list but what you have is 3.4f with all it's packages updated and any different packages in 3.5 would not be there.

Portage is portage - your world file isn't install specific. So kwrite in 3.4 is kwrite in 3.5, the only thing that changes is version of kwrite. 3.4 and 3.5 get/got their packages from the same place -> portage.

He is maintaining his own configs and world file. I am not experiencing this at all as I am about same boat as him for install, but mine is x86

[Fitzcarraldo]

In all 28 pages of the second thread I mentioned earlier, no one suggested re-installing. Well, actually they did, but in the context of installing 32-bit instead of 64-bit Gentoo, as it appears that this particular problem is more likely to occur in 64-bit Gentoo

I am not experiencing this at all as I am about same boat as him for install, but mine is x86

Some quotes from that second Gentoo thread I listed earlier:

hm, I've seen for a lot of people going to x86 fixed the problem, could it be that the reduced memory required when compiling in x86 userland is actually making a difference?

And another interesting test would be to use say 1gb of that memory for compcache, and tmpfs for /var/tmp/portage and see if that'd be even faster :p

Could be the reduced disk/io and no swapping would be a very nice solution.

The memory could be a good explanation, maybe it's the main reason why i'm not having the problem as I have 8GB.

One thing that I notice is that the x86 is able to cache more, it uses 90% of available memory. On x86_64 only 10% is used for cache.

@likewhoa: fortunately or unfortunately, I haven't seen any freezes or lock ups since I moved to x86. I like x86. Everything works as it should, there is no emul crap and no duplicate java packages to support a firefox-bin. The benchmarks are owned by amd64 (see my other thread) but this system is for my personal use and not for running benchmarks. It has to feel faster and smooth.

I won't go back to amd64 until this issue is completely taken care of.

I won't go back to amd64 until this issue is completely taken care of.

Seems you're convinced. I think it's time I take the plunge and put x86 on a spare partition and try it out too.

Actually devsk did return to x86_64 later because of an unrelated problem he had with x86.

For the first time in 19 months of using SL I'm beginning to wonder if I made the right choice in going for x86_64 instead of x86.

Anyway, I'm going to rebuild the kernel without preemption, as several posters in that Gentoo thread said it made a big difference. I'll post when I've done it.

[dunsurfin]

Portage is portage - your world file isn't install specific. So kwrite in 3.4 is kwrite in 3.5, the only thing that changes is version of kwrite. 3.4 and 3.5 get/got their packages from the same place -> portage.

I bow to your superior knowledge, but a quick look at the package list shows packages in 3.5 that aren't in 3.4f (e.g. sys-fs/logfsprogs-1) and in 3.4f that aren't in 3.5 (e.g. sys-fs/devfsd-1.3.25-r9). Presumably changes are made to the package list for good reason and a world update won't change the list.

[wolfden]

Portage is portage - your world file isn't install specific. So kwrite in 3.4 is kwrite in 3.5, the only thing that changes is version of kwrite. 3.4 and 3.5 get/got their packages from the same place -> portage.

I bow to your superior knowledge, but a quick look at the package list shows packages in 3.5 that aren't in 3.4f (e.g. sys-fs/logfsprogs-1) and in 3.4f that aren't in 3.5 (e.g. sys-fs/devfsd-1.3.25-r9). Presumably changes are made to the package list for good reason and a world update won't change the list.

I guess I'm not understanding what your getting at. So I must be misunderstanding.

Here is what I assume:
I assume he has a 3.4 install, but as he stated he has a rolling release with this own custom config files. I know he has also done the correct migrations as portage has evolved otherwise he would be dead in the water long ago. So if he is running custom configs than I am sure his package list varies greatly from the default 3.4 install as he would of removed and installed what he needed. Regardless tho, the packages are portage based, portage maintained, portage migrated. This would include removing depreciated files and converting to newer methods.

I run x86_64 also, but I don't do a rolling release with it as most of our testers are x86 so I'm always testing the x86_64. I will not touch my x86 as that is running solid since beta of 3.4. I use the x86 as my main daily usage, x86_64 doesn't have enough benefits to warrant running it as main OS. If you have some odd reason to run x86_64 than I say go for it, otherwise stick with x86 for now.

It is an interesting topic tho and wonder if other distros are hit with this. I am curious to see what the finding will be.

[dunsurfin]

I guess I'm not understanding what your getting at. So I must be misunderstanding.

Perhaps my knowledge is insufficient to contribute fully to this thread. I had just wondered whether a fully updated 3.4f has some problems that a fully updated 3.5 didn't have (My reasoning being that I am using the x86-64 3.5 and not seeing the problems this thread is about). Reading your comments suggests that this is very unlikely, if not impossible, and I now know a bit more about the way things work. Looking back through my old posts I found that I had tried an x86 version (3.4e) but found it sluggish compared to my 3.3 x86.-64 and installed the x86-64 3.4e. I'm also curious as to the results of the thread.