Problems with memory allocation / renicing unlinkd

Hello,

we're using squid in a medium-traffic environment. We've got the problem that squid eats all memory after running for a while:

squid 15015 0.5 79.1 544972 409568 ? S May19 44:54 (squid)
squid 15016 0.0 0.0 1112 84 ? S May19 0:34 (unlinkd)

As mentioned in the config file, squid may eat up to twice of cache_mem. 532MB is nearly *seventeen* times of my set cache_mem (which is set to 32MB).

We're using Version 2.3stable2 on a dual-PIII 550 machine with 512MB RAM and U2W SCSI under Linux-2.2.14, glibc 2.1.2, compiled with egcs-2.91.66. The ChangeLog states nothing about memory allocation changes for 2.3stable3.

The perhaps most memory-relevant lines from squid.conf:

cache_mem 32 MB
maximum_object_size 40960 KB
cache_dir ufs /usr/local/squid/cache 512 128 128

I searched the list archive but found only references that memory allocation over time is normal. But *this* much?

Does anybody know if this is a possible memory leak or how I could restrict this behaviour in some way?

Periodically, unlinkd starts to work and eats lots of memory, too. There is *lots* of disk activity then (squid has got a 4GB U2W disk for himself) which slows down the entire system way too much. Is there a way to renice unlinkd in any way so it's a bit friendlier to the rest of the system?

Must we use a dedicated squid machine to get rid of these issues or did I miss something?

Thanks for any help.

MfG, P. Schindler
--------------------------------------------------------------------
Technische Administration / Technologiezentrum
                                                   Fon 0781-504-4040
PGP-Key auf Anfrage Fax 0781-504-4099
Received on Wed May 24 2000 - 03:51:08 MDT