Poor performance (unresponsive, pausing, stuttering etc.)

[chsims1]

Simply an observation that dunsurfin will be correct in saying that there are likely to be differences in a fully updated 3.4 and 3.5, because both installations did not (presumably) produce identical initial results. So, is it not possible that things installed under an initial 3.4 could be updated to produce the detrimental effects?

[wolfden]

It shouldn't make no difference as long as one is properly maintaining it. If he wasn't, his 3.4 wouldn't be working worth a hoot with current portage. Portage will eventually force you make the necessary changes. So if a package is depreciated since 3.4 release, you will have to manually make the changes. 3.5 has the changes and fixes needed to work with current system at the time of it's release.

Openrc is a perfect example - 3.4 has to migrate manually to the current changes. When 3.5 was released, it had that done already. Portage changes a lot, especially with the ~x86 branch. That is just one example of the many changes 3.4 has gone through. A fresh install of 3.4 and than doing a world update on it would be for the insane. You would be able to build several gentoo installs faster.

One good thing about Fitzcarraldo, he doesn't give up on something and digs deep and usually always produces a solution.

[KelloggsFrosties]

I am experiencing the same poor performance on one of my pc's. It is a clean SL 3.5 installation, never used emerge, except for updating mono and monodevelop on it. Although it is an AMD 64 i am using x86-SL on it.

The setup is as follows:
AMD 64 3200+
ATI Radeon HD 3650
1 GB Ram
60 GB Drive

I am quite sure, that it is not caused by the low amount of Ram, because i never run out of memory and swap is nearly never used. I am quite sure, that it has to do with the fglrx-driver. I tried to update driver to latest via emerge, but did not solve the problem by doing so.

Fitzcarraldo: Are you using Compiz? Is disabling Compiz solving the performance issue for you?

[KelloggsFrosties]

Uuuuh. And another idea. Are you using Wlan? And if yes, which driver?

Mine is Broadcom 4318 WLAN using b43.

[Fitzcarraldo]

Fitzcarraldo: Are you using Compiz?

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).

Is disabling Compiz solving the performance issue for you?

As one would expect, the performance is significantly better when switching Compiz-Fusion off. However, I can still easily get SL to pause or stutter just by dragging one window rapidly or by scrolling up and down rapidly in a Firefox window, so, no, it does not solve the performance issue (especially when it should not occur at all: see further on).

As expected, htop shows that, under KWin or Metacity, there is less load on the CPU and less RAM is used than under Compiz-Fusion. Interestingly, I can see clearly from htop that Metacity puts much less load on the CPU and uses less memory than KWin. And neither Metacity nor KWin get anywhere near overloading the laptop. Mind you, neither does Compiz-Fusion. The one exception -- also the case when Compiz-Fusion is enabled -- is at the end of a pause/stutter, when the CPU load for both cores shoots up to virtually 100% and then back down again. I don't know whether this is actually happening at the end of the pause/stutter, or whether htop is frozen during the pause/stutter and can only show the higher load after it can run again. I don't notice significant flashing of the HDD LED when this occurs.

Are you using Wlan? And if yes, which driver?

No, I'm not using Wlan. I'm using the native iwl3945 driver built into the kernel.

I conducted three more experiments, all using the same 33.5 Mb DivX music video that causes a lot of pauses and/or stuttering under SL on my Acer laptop:

1. Dual boot into 32-bit Windows XP Professional on the same laptop (Intel Core 2 Duo T7200 @ 2 GHz + 2 GHz, 4 Mb L2 cache & 667 MHz FSB, with 2 Gb RAM and 160 SATA HDD, 256 Mb ATI Mobility Radeon X1600).
Excellent performance. The video plays without any pauses or stutter even when I am copying very large files to and from an external USB thumb drive and dragging windows about at the same time or scrolling very rapidly up and down in a Firefox window. I could not get the video to pause or stutter.

2. 32-bit Vista Home Premium (SP1) on a new Acer Aspire SA90-LB7H tower PC (Intel Pentium Dual Core E2160 @ 1.8 GHz + 1.8 GHz, 1 Mb L2 cache & 800 MHz FSB, with 1 Gb RAM and 250 Gb SATA HDD, 128 Mb ATI Radeon HD2400 Pro).
Excellent performance. The video plays without any pauses or stutter even when I am copying very large files to and from an external USB thumb drive and moving windows about at the same time or scrolling very rapidly up and down in a Firefox window. I could not get the video to pause or stutter.

3. 32-bit PCLinuxOS (KDE 3.5.9, 2.6.18 kernel) on a Gateway Solo 9300 laptop bought in 2000 (Pentium III Coppermine @ 795.92 MHz with 281.9 Mb RAM and 10 Gb HDD, ATI 3D Rage P/M Mobility AGP 2X).
Good performance. The video plays without any pauses or stutter if nothing else is going on (unlike on my high-powered Acer laptop), although there is periodic pausing whilst copying very large files to and from an external USB thumb drive and/or I am moving windows about at the same time or scrolling very rapidly up and down in a Firefox window. Overall, though, the performance is good.

Having performed these experiments I now realise just how bad SL is performing on my laptop! When you look at the specs of the three machines, the performance of Windows XP on the same hardware, the performance of Vista in half the RAM, and the performance on an eight-year-old single-core laptop with a fraction of the RAM and less than half the CPU frequency, there is something very wrong with SL's performance. I can now see why so many people in that 28-page Gentoo forum thread are so keen to find the cause (the "silver bullet" as one of them put it) of the poor performance. This is awful. Pity the old SL kernel sources 2.6.22 and earlier are not available in the overlay any more, otherwise I'd try installing one of them to see if performance got back to the way it was when I had 3.3 and 3.4f installed.

[Fitzcarraldo]

Turning off pre-emption (see an earlier post) didn't work. In fact, although I didn't benchmark it, I think it made things slightly worse. I've still got the commit=60,noatime,nodiratime in fstab, as that seems harmless enough and may possibly have made a very slight difference despite the following:

a) Performance before making any changes:

Code: Select all

# hdparm -tT /dev/sda

/dev/sda:
 Timing cached reads:   6192 MB in  2.00 seconds = 3102.36 MB/sec
 Timing buffered disk reads:  142 MB in  3.04 seconds =  46.78 MB/sec

b) Performance with the commit=60,noatime,nodiratime in fstab:

Code: Select all

# hdparm -tT /dev/sda

/dev/sda:
 Timing cached reads:   5148 MB in  2.00 seconds = 2578.84 MB/sec
 Timing buffered disk reads:  122 MB in  3.02 seconds =  40.36 MB/sec

Read rates are worse for case (b)!


Well, below are a few of the suggestions on the Gentoo forums:

1. Disable pre-emption. (Didn't work for me. )
2. commit=60,noatime,nodiratime in fstab (Didn't work for me. )
3. Look at CFLAGS. (I have not tried it yet.)
4. Change from ext3 to ext2, ext4 or reiserfs. (I have not tried it yet.)
5. Install Gentoo x86 instead of x86_64. (I have not tried it yet.)
6. blockdev --setra 16384 /dev/sda (Didn't work for me. )
7. Change the I/O scheduler from CFQ to deadline-scheduler in grub.conf or kernel config. (I have not tried it yet.)
8. Adjust SWAPPINESS from 60 (default) to 20. (Didn't work for me. )
9. Device Drivers --->
ATA/ATAPI/MFM/RLL support --->
<*> XXXXXXXX chipset support (if it exists)
(ATA/ATAPI/MFM/RLL support is disabled in the SL 2.6.26 kernel, being superseded by SCSI support, and, in any case, I couldn't see an entry for my particular Intel chipset. )
10. Turn off nmi_watchdog in grub.conf (I've tried it and it does not seem to have any effect. )
11. Set libata (CONFIG_ATA) but not IDE (CONFIG_IDE) (This is the 2.6.26 SL config default anyway. )
12. Turn off CONFIG_CC_OPTIMIZE_FOR_SIZE (This is the 2.6.26 SL config default anyway. )
13. Use compcache (I have not tried it yet.)

So, where am I now? Well, I performed Step 10 above and rebooted (with Compiz-Fusion enabled) and it didn't make any difference. Neither did it when I then exited from Compiz-Fusion. But I left Compiz-Fusion off and shut down, and I've just rebooted and at least now the video does not pause/stutter while I don't touch anything. So is that just down to not having Compiz-Fusion starting after boot, or have Steps 2 and 10 also played a part? I shall have to remove them one by one to see.

Well, at least I can get the SL x86_64 performance on my laptop to be a bit better by disabling Compiz-Fusion, although it's still not nearly as good as the performance under Windows XP on my laptop or under Vista on my tower PC (see my earlier post). And viewing videos with Compiz-Fusion enabled is a total non-starter.

I'm going to continue to investigate. My gut feeling is still that this is a kernel issue. I've noticed that if I just leave the video running with Compiz-Fusion then it pauses and/or stutters periodically, and so does the disk LED on my laptop, so I still believe this is somehow tied to SATA I/O. I wonder what kernel 2.6.27 will bring?

EDIT: Fixed a few typos (to help searching in future.)

[dunsurfin]

Just as some sort of "control" to your experiments (even though I have nvidia and not ATI graphics) I tried to get my machine to stop responding by playing a DVD video (using kaffeine), copying and then deleting a 950 MB folder of files from one partition to another, running Thunderbird (it almost always is), and doing some very rapid scrolling up and down in Firefox (3.0.2). I also installed and ran htop rather than top. I had no degradation in performance at all and the DVD (Belleville Rendezvous) played perfectly. When I was doing the rapid scrolling, both CPUs went up to about 75% but memory usage was about 30% and no swap at all. Compiz was running all the time.

[Fitzcarraldo]

Now that's the sort of performance I would expect. What sort of hardware have you got, dunsurfin? Is that a laptop or desktop? SATA or PATA HDD? Diameter, capacity and RPM? HDD Controller? Mobo, CPU, RAM, etc.? Can you also paste a copy of your kernel config file /usr/src/linux/.config for me to study? It doesn't matter if it's the config file for 2.6.25-r1, I will take that into account. (Actually I still have my 2.6.25-r1 .config file, so it will be interesting to compare those too.) Thanks.

BTW, my internal DVD drive is connected via USB according to Windows XP, and it can't keep up well even under Windows XP, so that's one of the reasons my experiments have all been with video files on the HDD. I'd expect a HDD to perform better than an optical drive.

[dunsurfin]

It's a desktop, Fitzcarraldo, with these specs: I've also attached a copy of the .config file (in my case it was in usr/src/linux-2.6.25-sabayon-r1/ as I have two kernels installed), a copy of dmesg output and also of cpuinfo. Hope this helps as that was all a hard task to do before I'm properly woken up
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]

Many thanks, dunsurfin. It'll take me a while to wade through the .config file to see if there is anything worth my trying.

By the way, /usr/src/linux is a symlink, so you should still have it on your PC. When you do an "eselect kernel list" it will show what the symlink is currently pointing to (2.6.25-r1 in your case, 2.6.26 in my case). When you do an "eselect kernel set <x>" it changes the symlink to point to the kernel number <x> you select from the list:

Code: Select all

# ls -la linux
lrwxrwxrwx 1 root root 20 2008-10-06 16:46 linux -> linux-2.6.26-sabayon
localhost src # eselect kernel list
Available kernel symlink targets:
  [1]   linux-2.6.25-sabayon-r1
  [2]   linux-2.6.26-sabayon *
localhost src # eselect kernel set 1
localhost src # ls -la linux
lrwxrwxrwx 1 root root 23 2008-10-07 12:17 linux -> linux-2.6.25-sabayon-r1
localhost src # eselect kernel set 2
localhost src # eselect kernel list
Available kernel symlink targets:
  [1]   linux-2.6.25-sabayon-r1
  [2]   linux-2.6.26-sabayon *
localhost src # ls -la linux
lrwxrwxrwx 1 root root 20 2008-10-07 12:18 linux -> linux-2.6.26-sabayon
localhost src #