Re: Squid 2 - so what? ;-)

Henrik Nordstrom wrote:

> James Zhu wrote:
> >
> > Does squid2 support pipelining on top of persistent connection
> > between a parent squid and a child squid?
>
> No. No pipelining is done between two caches. Only one request at a time
> is sent on a persistent cache-cache or cache-server connection. The
> reason to this is that to many things can go wrong when pipelining
> requests, and HTTP discourages pipelining of requests on a shared HTTP
> connection.
>
>
>
>

Are there any proxy server support pipelining between caches?

Can you elaborate a little bit on the possible problems with pipelining
between proxies? I am not sure if you meant it is hard to implement or it
is fundamentally inefficient...

Here is my understanding, please correct me. Since pipeline responses from
server must be in the same order as the requests, putting different
requests originated from different clients through a same pipe may cause
extra delays to all clients affected by a single timeout (or just slow)
respond. Likely this will degrade the performance of the proxy.

Ok, so is there any other problems associated with pipelining between
proxies? If what I said above is the only problem, how about the idead of
pipelining the requests from the same client through a single TCP
connection between proxies? Http 1.1 clients are doing pipelining anyway,
this wouldn't cause any extra problems. The change would be mostly
modifying or adding a hashing function to pconn? This will make squid a lot
faster, especially when the out going connection is slow. Pipelining over
persistent connection gives us the real win in the performance.

> Squid supports pipelining of client requests, with a limit of four
> outstanding request. Pipelined requests are however not well tested and
> there may be some bugs left, especially on uncacheable replies or
> replies which forces Squid to close the connection (no known
> content-length).

hmm...

> ---
> Hen
Received on Mon Oct 12 1998 - 12:45:08 MDT