Re: Page faults with physical I/O on Linux

From: Tim Thornley <>
Date: Tue, 07 Apr 1998 08:56:08 +1000

I am running a very similar system with similar results. (RedHat 4.2 plus

I see a page fault to request ratio of about 2:1 with NOVM and 0.8:1 with
the VM version. I found it interesting the I get less page faults with the
VM even though memory is more fully used.

ps reports about the same number of major faults.

I have found the GNU malloc makes no difference to paging when install as
suggested in the FAQ on my RedHat system.

I am also interested in any solutions / comments.


At 04:01 PM 4/6/98 -0500, you wrote:
>Okay, I've been lurking on this list for long enough, time to ask a
>question :)
>I've read a good portion of the docs (FAQ, Users Guide, and Release Notes)
>and I think I've decided that the squid box I've set up is paging an
>abnormal amount.
>Machine details:
>RedHat 5.0
>Kernel 2.0.33 with FDs patched to 4096
>128 MB of RAM
>SCSI Disks
>3600 MB of cache space on a single drive (about 1/3 full right now)
>OS and swap on seperate drive on same controller
>Squid versions 1.1.20+retry and 1.NOVM.20+retry tried.
>I've also compiled the GNU Malloc library.
>An average day sees a large number of hits over a short period (lunchtime
>at the school), light activity the rest of the day, and almost no hits
>from 9 pm to 7 am.
>I'm mostly wondering what an abnormal level of paging is (about 1 to 1 or
>greater is whats in the FAQ), and how to adjust the parameters to lower
>the amount of paging. If anyone is interested, I'd love to have a
>reccomendation on how you would configure squid with these amounts of RAM
>and disk space. I've got a webpage setup that has many of the steps I went
>through in setting up the machine, with much thought, and I'd like to
>share it with other smaller school districts who have a need to reduce
>traffic on their Internet connections, but I'd like to get a handle on the
>amount of paging first.
>BTW, that URL is
>if anyone cares to look.
>Also, the cache manager is on in the usual
>place, if that might help the analysis.
>Thank you very much for all help,
>Corey Betka
>Unix SysAdmin
>Hoopeston Area Schools
>Senior in Computer Science, College of Engineering
>University of Illinois at Urbana-Champaign
Tim Thornley (w) 61 2 62420605 (Fax) 61 2 62425728
Received on Mon Apr 06 1998 - 15:57:31 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:39:36 MST