[squid-users] squid with large RAM cache on NetBSD

[Ray Phillips]

Dear Squid Users:

The operating system I prefer to use is NetBSD and squid runs on it 
nicely.  However, when I modified squid.conf to make the squid 
process use a large percentage of my machine's RAM (so more objects 
would be held in memory and users of the cache would experience fewer 
fetching-from-disk delays) I ran into trouble.

After the change squid was restarting about once per day and this 
error was appearing in squid's log:

   xcalloc: Unable to allocate 1 blocks of 4104 bytes!

It turned out this was due to NetBSD's default limit of 256 MB on a 
process's size.

I built a new kernel to raise this limit to 390 MB RAM (the machine 
has 448 MB) but then found issuing the command

  squid -k reconfigure

caused the whole machine to freeze.  This doesn't occur when squid is 
using less than 256 MB even with the new kernel.

For reference, this is the URL of my first message to the port-sparc 
NetBSD mailing list:

  http://mail-index.netbsd.org/port-sparc/2002/12/05/0003.html

I've tried again since upgrading from NetBSD 1.5.2 to 1.6.1 but the 
problem is still there [1].

Someone suggested squid might be the culprit, which I doubt, but 
thought I'd check with you.  I presume it's common for people to run 
squid with in-memory caches well above 256 MB in size?

Ray



[1]  http://mail-index.NetBSD.org/port-sparc/2003/09/05/0000.html