[Bug 367377] Re: High load average, disk read, no apparent reason - 2.6.28-11

[Bryce Harrington]

** Description changed:

  Binary package hint: linux-generic
  
  I am not at all sure this is a kernel bug, but I don't know what else to
  file it against.  Perhaps someone who understands the VM subsystem could
  look at my logs and provide some indication of what may be the culprit
  here.
  
  Symptoms:
  
  At apparently random times when I am working, the system will be come
  extremely slow.  This frequently occurs on starting a new application,
  which may be Firefox, Open Office or Okular for example.  The load
  average spikes to a large value (greater than 10) and the hard drive
  light lights up, showing continuous disk I/O.  The user interface
  becomes unusable and the mouse pointer update is very slow (several
  seconds lag).  More often than not the system becomes so unresponsive
  that I cannot change to a text VT and kill X, or even use ctrl-alt-
  bkspace.  I have to resort to Magic-SysRq-RSEIUB and reboot.
  
  This is happening several times a day on a system with 2GB of RAM under
  a light workload of KDE4.2, Thunderbird, Firefox and sometimes
  OpenOffice or Netbeans.  At the time of onset the memory usage as shown
  by 'htop' is around 400MB used of 2GB physical (this excludes 'cached').
  It is impossible to predict just what action may start a frenzy of disk
  activity and a loss of responsiveness.  Sometimes even opening a page in
  an existing Firefox instance is sufficient to effectively DOS the
  system.  **IMPACT IS SEVERE**
  
  I am currently running up-to-date Jaunty with kernel 2.6.28-11, but I
  have had similar behaviour with Intrepid in the past.  I believe that
  upgrading this machine from 1GB to 2GB has made matters worse.  I have
  no swap device configured, but I have tried configuring swap in the past
  using a swapfile, and this has not prevented the problem from occurring.
  
  My confusion is what could be causing the disk I/O that is bogging down
  the system, and driving the load average up as processes wait on I/O?
  It can not be swap in the sense of using a swap partition or file, since
  I have none configured.  I don't have any reason to believe it is
  application I/O either - the problem occurs on starting many different
  apps, and appears to be triggered by requesting more memory.  I can only
  think it is paging executables / shared libs in and out, trying to make
  room.  I can't understand why this is happening, with so my physical
  memory free.
  
  Logs:
  
  This problem has proven difficult to diagnose because usually when it
  occurs I lose control of the machine and existing programs like top and
  so on stop updating.  I grabbed a little script from another thread on
  Launchpad which runs 'top' and 'vmstat' and dumps the output to a file.
  Today I had this script running while I had a 'high load average' event.
  In this case, I did not lose control of the machine as sometimes
  happens, but the load average spiked to 6 or so for no apparent reason.
  
  Running KDE 4.2, with Thunderbird open and maybe two Dolphin windows.
  Started OpenOffice Word Processor - it took maybe 5 mins to start it and
  shut it down.  Meanwhile load average is 6 or so, and mouse pointer
  unresponsive, disk is thrashing.  Once OO had shut down, the thrashing
  stopped.  I was able to repeat this behaviour by starting OO Calc, which
  also took minutes to start up and shut down.  During this time, I was
  logging 'top' and 'vmstat' output, which I will attach to this bug.  The
  script logs 'vmstat 1' and 'top -b -d 1' to 'vmstat.log' and 'top.log'
  respectively.
  
  I am not sure what other evidence I can capture.  I noticed that 'iotop'
  showed OO was reading from disk at around 5MB/s when the disk was
  thrashing.  What it could be reading @ 5MB/s for several minutes I
  cannot imagine, unless pages are being thrashed in and out of memory for
  some reason.
  
  Can someone understand my log files and point to the offending process?
  If you can suggest other information to capture, please let me know.  I
  am an experienced Linux user and I am happy to spend some time on this
  as it is rendering my main development machine unusable!
  
  I don't think this problem is tied to a memory leak in any particular
  user-space end-user application (although it could be a leak of some
  sort in the X-Server or KDE4 components).
  
  Hardware:
  HP Pavilion dv1680ea
  Core Duo 1.87GHz 
  2GB RAM
  Intel integrated graphics
  
  Software:
  Kubuntu 09.04 2.6.28-11 kernel
  KDE 4.2
  Firefox 3.0.9, OOo 3.0 and other apps
  NO SWAP CONFIGURED
+ 
+ [lspci]
+ 00:00.0 Host bridge: Intel Corporation Mobile 945GM/PM/GMS, 943/940GML and 945GT Express Memory Controller Hub (rev 03)
+     	Subsystem: Hewlett-Packard Company Device 30a0
+ 00:02.0 VGA compatible controller: Intel Corporation Mobile 945GM/GMS, 943/940GML Express Integrated Graphics Controller (rev 03)
+     	Subsystem: Hewlett-Packard Company Device 30a0

-- 
High load average, disk read, no apparent reason - 2.6.28-11
https://bugs.launchpad.net/bugs/367377
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.

-- 
ubuntu-bugs mailing list
ubuntu-bugs at lists.ubuntu.com
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs