[boost] asio formal review

From: Matthew Nourse (matthew_at_[hidden])
Date: 2005-12-12 00:40:15

I vote to accept asio into boost.

Potential bias disclosure: I'm friends with Chris Kohlhoff.

I have been writing asynchronous servers and clients with asio and
boost for about 20 months.  The following asio & boost -based server
applications are currently running in commercial production systems:

* a gateway from a proprietary networking API to HTTP.  Up since
November 2004. Windows+VC7.

* a custom authenticating HTTP proxy.  Up since February 2005.  Linux+gcc.

* an HTTP gateway to a proprietary IM system. Up since May 2005 except
short upgrades to my code.  Linux+gcc.

* an XMPP (Jabber) gateway for the same proprietary IM system.  Up
since August 2005 except short upgrades to my code.  Linux+gcc.

* a lightweight publish-subscribe mechanism.  Up since August 2005.  Linux+gcc.

I've found asio's implementation to be extremely efficient.  For
example the publish-subscribe mechanism can handle 500 requests/second
without any detectable CPU usage on a 3Ghz P4 even though my code is
untuned & does more memory allocations than is strictly necessary.

I found the asio design philosophy and documentation relatively easy
to understand however I before I used asio I had already written aynsc
socket applications using Windows' IO completion ports.  I am not a
template-based design expert nor am I a longstanding boost user.  My
first use of boost was when I started my first asio application in
August 2004.

Matt