There is something subtly wrong with my Linux install that I just can't locate.

It is Ubuntu Lucid Lynx (10.04) 64-bit. Hardware is a Dell Optiplex 960: Intel Core 2 Quad CPU, 8Gb of RAM, 2x 300Gb HDDs. /home is ext2 on one disk and everything else is on the other (/ is also ext3). I have VirtualBox running a 64-bit Vista image for Outlook calendaring, but the heavyweight apps are IntelliJ, NetBeans, MySQL and Opera. Opera also loads my mail (IMAP) of which there is over 10,000 messages.

The problem is that Opera stalls for a few seconds from time-to-time. Watching the process list shows it's in log_wait_commit which means (as far as I have figured out) the filesystem is holding things up. Sometimes I can make this happen by doing a subversion update, but usually it happens for no reason I can see. It usually happens to Opera, but I've seen NetBeans go under, too. It doesn't make the app crash - it's just completely unresponsive for a few seconds.

Googling has not helped. The closest I got was to remove the sync attribute in the file system. This achieved nothing. On the advice of a Linux guru friend, I lowered /proc/sys/vm/dirty_writeback_centisecs to 300, but that didn't do anything, either. And it was all he could think of.

What is going on and can I fix it? (And how?)

  • Run dmesg from the command line during or just after this happens and look for references to ext2 or ext3, or to sda1, etc. This might help with figuring out what is going on. In particular, if the cause is hardware-related, dmesg will likely show it.
    – Jander
    Jan 11 '11 at 4:31
  • Nothing shows up in the system logs. I thought briefly the flashplugin was causing mischief, but it's not.
    – staticsan
    Jan 12 '11 at 2:13

I've seen quite a few of these hangs in various forums, but never with a fix. Since the problem appears to be fixed on my machine, I am sharing here what I know:

  1. Out of my four Ubuntu systems, all with 10.04 and ext3 file systems, it happened only on an AMD quad system. The three others are Intel quads. This may be a coincidence.
  2. Processes hanging frequently are browsers (regardless of which one I use) and gconf-d
  3. The problem got introduced with an update to 9.04. Unpatched 9.04 does not show the hangs, they immediately appear again after installing the 9.04 updates. This is reproducible. All further versions including 10.10 show the hangs.

What seems to have fixed the problem:

A complete reinstall using ext4 for all filesystems including /home. I tried reinstalls before, of course, but keeping the FS on ext3.

  • That resonates with what I discovered. I found I could reduce the problem to a livable proportion by putting the T attribute on the .opera directory and deleting the .opera/cache directory. This changes the ext3 allocator's behaviour to spread sub-directories around on the disk more. It seems there may be a bug in the ext3 allocator under multi-core that browsers trigger.
    – staticsan
    Mar 7 '11 at 22:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.