Re: page faults with physical i/o ???

From: Jonathan Larmour <>
Date: Mon, 13 Jan 1997 14:29:33 +0000

At 22:12 11/01/97 +0100, Michael Boettger wrote:
>I have a little problem with Squid v1.1.3 and Linux v2.0.27.
>Even on starting up squid IŽll get massive i/o faults and also after about
>100000 requests ther are about
>70000 i/o faults.

Here's a reply I gave to someone asking the same question:

At 21:12 28/11/96 +0100, Michael Boettger wrote:
>> From: James R Grinter <>
>> On Thu 28 Nov, 1996, "Michael Boettger" <> wrote:
>> > Is it normal to get that much physikal i/o page faults ????
>> > Page faults with physical i/o: 28122
>Thanx for the quick answer but that canŽt be true. Here is my "cat

I think you'll find that 28122*PAGE_SIZE i.e. 28122*4096 = 115Mb, then add a
few Mb of shared libraries (from ps -axu), will not be too dissimilar from
the total size of Squid (from ps -axu or mallinfo() (NOT necessarily what
squid declares as its usage)). There will be a little difference but not much.

The reason is that while Linux may allocate you the space in principle when
you malloc, it does not actually give you real pages in memory (whether
immediately available or not). It instead relies on page faulting do this.
This is because malloc() calls brk() which calls into the kernel as
sys_brk() which calls do_mmap() i.e. it is the same process as mmap.

I would imagine the same holds for most half-decent modern OS's.
Origin UK,323 Cambridge Science Park,Cambridge,England. Tel: +44(1223)423355
------[ Do not think that every sad-eyed woman has loved and lost... ]------
April 12th! Ra!Ra!----[ she may have got him. -Anon ]-----April 12th! Ra!Ra!
Help fight spam! These opinions are all my own fault
Received on Mon Jan 13 1997 - 06:37:43 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:34:03 MST