Mort wrote:
> 1)You speak of "idle connection". How do you define this "idle connection".
idle == squid is not currently waiting for a reply from the server on this
connection. The time between where Squid has received a server reply and it
schedules another request on this server connection.
> 3) If squid supports pipelining, we expect that these multiple requests can
> be issued by the client to go over one connection. Our experience is that
> in this scenario, Squid creates multiple connections. Are we
> mis-interpreting the documentation?
RFC2616 discourages proxies from using pipelining due to the complexity
involved in request retransmission requirements on pipelined connections when
the connection is closed prematurely.
To countereffect the performance drop seen by this by high speed clients Squid
processes up to two requests concurrently per client connection using
pipelined requests. See the pipeline_prefetch directive in squid.conf.
The fact that a client sends a pipeline of requests has no bearing on how
these requests will arrive at the server other than that the requests will
arrive approximately in cronological order. There is no relevance between
client side pipelining and proxy<->server connection management. Again,
pipelining is a hop-by-hop property of HTTP.
hop-by-hop == a business between the client and the proxy, OR the proxy and
the server.
end-to-end == a business between the final client and the origin server
(subject to caching)
The most notable hop-by-hop features of HTTP/1.1 is
a) Connection management (i.e. persistent connections, pipelining etc)
b) Transfer encodning
Regards
Henrik
Received on Wed Aug 21 2002 - 11:52:19 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:09:46 MST