SCIENTIFIC-LINUX-USERS Archives

September 2004

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Troy Dawson <[log in to unmask]>
Reply To:
Troy Dawson <[log in to unmask]>
Date:
Tue, 7 Sep 2004 11:32:34 -0500
Content-Type:
text/plain
Parts/Attachments:
text/plain (49 lines)
Hi,
There has been several people that have been having bad performance on their
desktops with Fermi Linux LTS, and/or Scientific Linux 30x.  One of the
reasons for this is that RedHat has tuned their kernel to get maximum server
speed, and this has occasionally made some peoples desktop's slower ...
depending on what you do on it.
The first change (setting the file /proc/sys/vm/inactive_clean_percent to 30)
has already been made a default in the 2.4.21-15 and higher kernels.  So your
first action should be to upgrade your kernel if you can.

The second performance boost we've found deals with the I/O read write latency
variables.  Basically setting them back to RedHat 9 settings.

If you do a
   elvtune /dev/hda
(substitue /dev/hda with whatever your drive is)
you will see something like

/dev/hda elevator ID            0
         read_latency:           2048
         write_latency:          8192
         max_bomb_segments:      6

For a RedHat 9 (and earlier) releases, the read_latency was set at 64.  This
meant that you needed to read something from the disk, the maximum you had to
wait was 64 milliseconds.

HOW TO FIX
Do the following for each of your hard drives.

   elvtune -r 64 -w 8192 /dev/hda

and to make sure that it keeps after a reboot, you need to put those lines in
your /etc/rc.d/rc.local file.

We're thinking of making a SL_ rpm (like a zz_ rpm for Fermi guys) that will
change this for those that need the change.  But if those of you who are
experiencing bad desktop performance, could test this, and see if it helps
improve things, we would appreciate it.

Thanks
Troy

--
__________________________________________________
Troy Dawson  [log in to unmask]  (630)840-6468
Fermilab  ComputingDivision/CSS  CSI Group
__________________________________________________

ATOM RSS1 RSS2