Memory usage - accounted 200+Mb physical 128 ???

Hi all,

I've got something weird here that I can't understand. I'm running a
cache with 128Mb ram, 4Gb cache.

The accounted memory usage goes like thus :-
        
StoreEntry 428834 x 52 bytes = 21776 KB
URL strings = 21173 KB
IPCacheEntry 920 x 36 bytes = 32 KB
FQDNCacheEntry 0 x 56 bytes = 0 KB
Hash link 0 x 12 bytes = 0 KB
Pool MemObject structures 13 x 80 bytes = 1 KB ( 0 free)
Pool for Request structur 11 x 4408 bytes = 47 KB ( 0 free)
Pool for in-memory object 0 x 4096 bytes = 0 KB ( 0 free)
Pool for disk I/O 20886 x 8192 bytes = 167088 KB ( 0 free)
NetDB Address Entries 0 x 76 bytes = 0 KB
NetDB Host Entries 0 x 8 bytes = 0 KB
NetDB Peer Entries 0 x 24 bytes = 0 KB
ClientDB Entries 1615 x 292 bytes = 460 KB
Miscellaneous = 644 KB
Total Accounted = 211272 KB

Making the total squid is reporting as using well beyond the physical,
but there's no swapping to speak of taking place (about 40k). About
the same time as the accounted usage broke 150Mb the performance
started to dip something terrible (not too good on a proxy also doing
acceleration).

As a second issue, not critical but something ppl might be interested
in I've found a possible problem with accelerators in general.

It looks like putting an accelerator/redirector in place on the proxy
opens up a possible dos attack.

GET / HTTP/1.0 on the IP of the proxy (rather than the IP of an
accelerated site) causes the proxy to shoot off into a death spiral of
requests. When without the redirector the request is trapped and
junked by squid. I've stuck a few extra lines into my code to trap
requests for the IP of the proxy, but I think it's something to watch.

        Mark

-- 
Mark Lowes <hamster@lspace.org>			http://hamster.wibble.org/
Wibble.org					http://www.wibble.org/
Lspace.org					http://www.lspace.org/