Re: Cisco Cache Director (Was RE: Does Squid beat the rest? )

On Sat 20 Sep, 1997, Gregory Maxwell <nullc@nightshade.z.ml.org> wrote:
>`Cisco's Cache Engine is thread based' (True, given for ref)
>`Cisco's Cache Engine uses a special OS with low context switch
overhead'(true)
>`..Cisco's Cache Engine is the only cache system which is able to overcome
>expensive context switches when dealing with multiple requests' That is a
>flat lie, squid is a single thread, so no switch.

Your system is context switching for every DNS lookup, if you think about it.
What would be a worthwhile development is to use Darren Reed's arlib to perform
asynchronous DNS lookups. I think this might actually be in the 1.2 track.

But I think Cisco are lining themselves up against Netscape's proxy, with that
statement. Netscape's proxy still runs a process per connection, I believe.

There are also a lot of sites out there still running CERN httpd for their
proxy (having done a quick recce of one day's user-agent logs on a busy
server), and a few using old versions of Squid[1] and very old Harvest code.
Clearly these people don't spend time following this list, or much time in
keeping up to date with new software. These are the people that are targets for
this sort of product - plug in and forget.

Henrik's idea of redirecting packets to just a simple statement on
configuration - that's fine, but the primary reason people are looking at
transparent proxying is because of support costs. It's very difficult to
explain to many users, each with different versions and different types of
browsers, how to configure them to use a proxy: and a lot of people out there
do (sadly) find it impossible to follow simple instructions.

James.
([1] New versions of Squid don't show up, because they don't append themselves
to the User-Agent header, instead using Via.)
Received on Sun Sep 21 1997 - 04:28:20 MDT