> Duane Wessels:
> I'd like to get an idea of to what extent people think the ICP HIT_OBJ
> feature is a good idea to continue supporting.
The main reason to support ICP_HIT_OBJ is speed, and the limiting factor today is setting up the TCP connection instead of getting the (small) object via ICP. With persistent connections this slow-down is removed, and there is no reason to keep ICP_HIT_OBJ. My gut feeling is to slay ICP_HIT_OBJ the minute we have persistent connections.
Regarding the evils of ICP_HIT_OBJ:
- fragmentation does occur (I'll try to dig up my measurement of this if needed) for ICP and is evil
- congestion has not been a problem caused by ICP, mainly because HTTP/1.0 over TCP takes care of congesting links/networks
- reading from disk takes much less time than sending over network in my experience (but then my caches are some thousand/hundred kilometers apart;)
- traffic minimization is not a big issue for those with several parents/sibling, as ICP_HIT_OBJ generates extra traffic (only the ones use are logged by Squid, there is more traffic on the link)
- <paranoid> it is possible to spoof ICP_HIT_OBJ replies with false documents, this is easier to do with multicast ICP</paranoid>
Ingrid
--
Ingrid.Melve@uninett.no UNINETT, N-7034 Trondheim, Norway
Phone +47 73 55 79 07 Fax +47 73 55 79 01
http://domen.uninett.no/~im/eng.html
"Sometimes it is better to light a flamethrower than curse the darkness"
Received on Thu Nov 20 1997 - 00:09:42 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:41 MST