Hi all,

As a follow up to my previous post. I did some testing and have come
to the conclusion that firefox is indeed at the heart of my problem
re: high I/O wait times. See below.

On Wed, Mar 31, 2010 at 6:39 PM, john <lists.j...@gmail.com> wrote:

> Oh, heck, I'll throw out another thing :-)
>
> I was surprised to see that there was so much disk write activity. I
> am trying to figure out what is getting written where. I found a tool
> called IOTOP that should correlate disk i/o to particular apps.
> Unfortunately it uses some kernel hooks that aren't supported by
> Ubuntu kernels so i am in the process of compiling an ubuntu kernel
> with the proper stuff included.

So I compiled a custom kernel by copying my .config and following directions
here https://help.ubuntu.com/community/Kernel/Compile

I turned enabled I/O accounting so that I could use IOTOP
http://guichaz.free.fr/iotop/

CONFIG_TASK_IO_ACCOUNTING=y

compiled the kernel and rebooted.

IOTOP has a number of interesting features, including the ability to
show all processes, all threads, only active process/threads,
cumulative or real-time disk I/O etc. Running IOTOP while opening
web pages in firefox, browsing, watching youtube etc, showed my that
firefox does a LOT of disk writes and very few reads. Here's some
sample output in with the "cumulative" switch turned on

# iotop -a -P

total DISK READ: 0.00 B/s | Total DISK WRITE: 0.00 B/s
  PID  PRIO  USER     DISK READ  DISK WRITE  SWAPIN     IO>    COMMAND
 4423 be/3 root         12.00 K   1984.00 K  0.00 %  0.13 % [kjournald]
 2581 be/3 root          0.00 B    636.00 K  0.00 %  0.03 % [kjournald]
 5099 be/3 ntp          12.00 K      4.00 K  0.00 %  0.00 % ntpd -p
/var/run/ntpd.pid -u 115:127 -g
10442 be/4 john          8.00 K      0.00 B  0.00 %  0.00 % gnome-terminal
 7207 be/4 john         16.00 K      0.00 B  0.00 %  0.00 %
gnome-panel --sm-client-id default1
10688 be/4 john          4.00 K     17.20 M  0.00 %  0.00 % firefox-bin
 7172 be/4 john          0.00 B    336.00 K  0.00 %  0.00 % gconfd-2 11
 5005 be/4 syslog        0.00 B     52.00 K  0.00 %  0.00 % syslogd -u syslog
 5623 be/4 root          0.00 B    180.00 K  0.00 %  0.00 % winbindd

In 10 minutes of futzing about firefox wrote 17.2 M to disk and read
off 4K. All the writing apparently happens in the users ./mozilla
directory. Indeed of the 47Gigs used on my disk 26Gigs (55%) are given
over to my 570 users firefox profiles (eg ~45 M per user).

The space isn't a problem, but I am really beginning to think all of
that disk activity is really hurting our performance. I
often feel like some of the best aspects of LTSP are nullified by
Firefox's affect on the multi-user environment. At the same time I
like firefox as a web-browser. I would love to find a way to make
firefox feel like less of a liability, perhaps pushing /home to a
faster disk will do that. I think Firefox's problems under LTSP really
color our users perception of the usefulness of LTSP/ubuntu.  Perhaps
I just don't know how to configure Firefox correctly...

I have followed
https://help.ubuntu.com/community/UbuntuLTSP/Firefox3Optimize and I
hope that it will make a difference. I am also going to mount /home on
a separate disk either under raid 0 or 10 and/or perhaps buy a SAS or
solid state disk. I am still mulling that one.

I'm still interested in folks ideas about the "fastest" approach to
take re: disk writes, and especially ideas about taming firefox under
LTSP.

Thanks to all for your ideas!

John

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
_____________________________________________________________________
Ltsp-discuss mailing list.   To un-subscribe, or change prefs, goto:
      https://lists.sourceforge.net/lists/listinfo/ltsp-discuss
For additional LTSP help,   try #ltsp channel on irc.freenode.net

Reply via email to