Hi,
I've read the user guide/config guide/faq and searched the mailing lists.
However, apologies if this has been answered before and I couldn't find it.
Can someone please explain to me the mechanism squid uses to select
a sibling to fulfil a request if more than one sibling has the object,
and I am using cache digests ?
The faq says that it will use NetDB to select a cache.
What info in here does it use:
1) the closest sibling cache to this cache (is this used/computed if not using ICP)
2) the closest sibling->origin server (my interpretation of the ICMP RTT info)
3) some total time combination of the two ?
If it is 2, then this is different from ICP. If I actually wish ICP like
sibling selection (i.e. quickest/closest/lightest loaded sibling since I
know they are all on the same network segment and origin ICMP RTT info is
therefore not useful and would simple add to useless pings everywhere), is
there a way I can do this ?
assuming I used cache digests, AND specified no-netdb-exchange, how would
it make its decision ?
Also, assuming no sibling has the object, it would seem beneficial in some
circumstances to still ICP or HTCP check all siblings, since:
1) the cache digest could be nearly an hour old
2) the lossiness of the digest could have created a false miss.
for example, if internet bandwidth is particularly limited and/or expensive.
It is unclear how/if ICP/HTCP works with cache digests on.
It seems to hint that digests are an alternative to ICP/HTCP without
specifically saying that they are mutually exclusive ?
Stuart C. Smith
Technical Architect
Venation
www.venation.com
Received on Thu Jan 22 2004 - 10:15:17 MST
This archive was generated by hypermail pre-2.1.9 : Sun Feb 01 2004 - 12:00:08 MST