it sounds like squid's WCCP implementation and/or TCP stack isn't ensuring
that the total packet size (advertised TCP MSS + ip-header size + tcp
header size) is less than the MTU.
remember that WCCP is based on layer-4. if fragmentation can occur prior
to the WCCP intercept, things can "go bad".
squid isn't unique in not ensuring that the TCP MSS is set to a reasonable
value; many other commercial cache vendors get this wrong too, no matter
how many times i've told them. :-(
cheers,
lincoln.
At 12:04 PM 17/03/2001 +0800, Miguel A.L. Paraz wrote:
>Hi,
>
>I have WCCPv1 working between a 7206 with 12.0.14(S), and squid 2.3stable4,
>*if* the clients are on the leased lines connected directly to the 7206.
>A redirect-list controls the clients to redirect by IP.
>
>For clients on other routers that point default to the 7206, it works for
>a while, and then later on they fail. The end-users' web sessions just stall.
>The hits on the access-list counters stop increasing, and the squid access
>log stops registering requests coming from them.
>
>Any ideas as to what to look at?
>
>If this really won't work by design, then I'd have to enable WCCP on the
>access routers/RAS's. If the remote RAS is at a location without any cache
>server, is it a good idea to make it point to a distant squid server? Or
>should the two be "close" or on the same LAN?
>
>Thanks!
>
>
>--
>
>http://www.internet.org.ph The Philippine Internet Resource
>Mobile Voice/Messaging: +63-917-810-9728
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:32 EDT