hc-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Tatu Saloranta <cowtownco...@yahoo.com>
Subject Re: HttpCore NIOSSL fatal exception
Date Wed, 05 Dec 2007 18:58:26 GMT

--- Sam Berlin <sberlin@gmail.com> wrote:

> Not exactly, no.  A persistent connection doesn't
> imply a lock/step of
> request/response. As another example, consider a
> protocol like
> BitTorrent or Gnutella.  You send packets to a host
> and read packets

Right, and for things like chat servers this also
applies, as well as other kinds of proxies.
For chat servers number of individual clients (or
downstream proxies) can be pretty high, but traffic
per client is sporadic, and often quite light overall.
And because of blocking issues (buffering, slow
receivers), read and write connections must be handled
separately. Hence naive blocking implementation would
end up with huge number of mostly idle threads.

As to HTTP pipelining, I thought that http browsers
did indeed support it, and I know some http libs do
implement it (like libwww... although it has/had bugs
in it). But server-side systems not as commonly.
It doesn't imply out-of-sequence sending (responses
are still ordered to align with requests I think?),
and is complicated to implement correctly. But it can
improve latency and efficiency a lot, esp. for
long-thin pipes (like traditional end user net

-+ Tatu +-

Be a better friend, newshound, and 
know-it-all with Yahoo! Mobile.  Try it now.  http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ

To unsubscribe, e-mail: httpcomponents-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: httpcomponents-dev-help@jakarta.apache.org

View raw message