Re: Large Scale Transparency - Disadvantages?

Miguel A.L. Paraz wrote:

> Based on the messages on this topic for the past few months, simple router IP
> redirection to the squid box is problematic?

* Problems with reliability. What happens if the cache box
dies/malfunctions?

* ICMP TCP/IP related problems. For example Path MTU discovery not
working from the cache to the clients. Possible workarounds include
disabling MTU discovery on the cache and/or redirecting all ICMP must
fragment packets to the cache and enable forwarding in case the message
was not for the cache connection..

Other problems precent in all TCP redirection techniques are:

* All traffic for the TCP session must pass tru a the redirection point
in the network. The net effect is that TCP redirection basically cannot
be done in a backbone network unless your backbone does not have (or
will have) any redundant/alternate paths. It is therfore recommended to
make the redirection closer to the clients.

* In case there is a problem with the redirection/cache, the the client
has no way of selecting to access the origin server directly. (mostly
political issue between you and your clients)

* HTTP is not designed to be transparently redirected to a cache. HTTP
assumes that a client will know if a proxy/cache is being used or not
and adjust accordingly, and some browser (IE < 5.01) "correctly" assumes
that if there is no proxy settings then no proxy/cache is being used.

> How about WCCP, does it get around the problems that are due to the fact that
> the browser doesn't know the presence of a cache?

It solves the first. Not sure about how ICMP is handled when WCCP is
used.

WCCP does not solve the browser/HTTP protocol problems, only some
technical issues in how the TCP/IP traffic is redirected.

> How do the commercial caches deal with this, or do they have the same problem?

They share mostly the same set of problems I think.

--
Henrik Nordstrom
Squid hacker