Squid fail using 1.0.20

Hi. Our Squid-1.0.20 running on Linux 2.0.23 died for the first time this
afternoon. In fact it died twice :( I found this in cache.log

FATAL: icpSendERROR: FD 23 unexpected port 3128.
Squid Cache (Version 1.0.20): Terminated abnormally.

Ideas?

I've also been getting buckets of these errors as well since I'm in error
reporting mode :)
[14/Nov/1996:19:37:30 +1300] comm.c:1235: commHandleRead: FD 21:
read failure: (104) Connection reset by peer

I'm interested to know if that's anything nasty.

And lastly... This one is interesting because we have a 1.2Gb cache spool
which is only 25% full at the mo.

[14/Nov/1996:15:56:22 +1300] store.c:1986: storeGetMemSpace:
Allocating beyond the MAX Store with total size of 4199814
[14/Nov/1996:15:56:22 +1300] store.c:1988: Start Deleting
Behind for every pending objects
[14/Nov/1996:15:56:22 +1300] store.c:1989: You should really
adjust your cache_mem, high/low water mark,
[14/Nov/1996:15:56:22 +1300] store.c:1990: max object size to
suit your need.
  
Cheers,
 Graeme

--
Graeme Merrall 
System Admin
Manawatu Internet Services Ltd, Palm Nth, New Zealand