> Ideally, my cache should say 'oh well, I'll discard that response and try
> via icp to my other siblings or my parents', or it should send an icp
> query just to the sibling it thinks (via cache-digest) has the object,
> thus treating cache-digests as merely a 'which sibling is *most* likely to
> have the object'
As far as I know squid (in non-beta form) currently doesn't deal with the
retries involved if neighbors refuse a request. Before cache-digests it was
less apparent, but we did see it with ICPs (they sibling says yes we have it
buit throws it away before the http request gets there). We had to stop
peering with at least one cache which had "miss_access deny" rules because the
denied messages got back to users.
If I'm out-of-date on this please let me know. I'd love to switch to
cache-digests but can't justify it if we will either fetch stuff through our
neighbors or cause our users to see denied messages.
-- Jon
Received on Fri Nov 20 1998 - 07:20:18 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:43:11 MST