Subject: Re: NT v UNIX web servers - question



>My question to you is, how might NT be a better OS than UNIX for hosting a
>web site?  Or would a UNIX server really do the better job?  I'd be most
>interesting in hearing back from grrls who have used both for web servers.
>I'm looking for practical, somewhat quantifiable pros and cons, to help me
>make a decision (read, MS bashing wouldn't be helpful, unless it has a
>point).


for straight file serving, under low/moderate load, the differences
between unix and NT would probably be minimal.   the two biggest
advantages to unix, IMO, are stability under high load, and ease of
integration between the webserver and other software.   

Windows and the Mac OS are designed around the concept of a computer
which interacts with a user sitting at the keyboard.   the generic term
is 'single-user context'.   SUC is great for highly interactive
applications, especially those like graphics development, which involve
making lots of small, arbitrary changes to a base document.   the
defining feature of single-user context is the stepwise ability to undo
the last action, and try something else.

unix, OTOH, is designed for multiple simultaneous users OR no users at
all.   its specialty is batch processing applications, where a
well-defined set of actions are applied to a large set of documents.  
multi-user contexts tend to follow a pass/fail model, rather than an
undo/retry model.


single-user machines have a built-in tendency to want user interaction.  
that can certainly be minimized, but it exists nonetheless.   their
ultimate fail case is generally to present a screen which tells the user
what went wrong, then wait for a reboot.   multi-user/blind machines
generally try to fail a single process in ways that don't require
shutting down the entire system.   nobody wants their private machine
spontaneously rebooting in the middle of their work, and nobody on a
shared machine wants to stop working and wait for a reboot because
somebody else had a software failure.

the basic assumptions of a multi-user context imply a resonably clean,
blind protocol for passing information between processes.   single-user
contexts can get away with messier protocols and pulling more shortcuts,
because they assume the user knows what's happening, and is willing to
live with the consequences.


a webserver is basically a blind system, and if it serves dynamic
content, prefers to do so in a pass/fail manner, with protocols that
allow it to communicate with other software cleanly and simply.   as
part of its character, the application wants a multi-user context.  
it's certainly possible to port most of the important features of that
application to both single and multi-user contexts.   the multi-user
context will offer the most support for unexpected demands and
flexibility for growth, though.