httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From George_Carre...@iacnet.com
Subject Re: IIS doesn't scale
Date Thu, 07 May 1998 12:47:05 GMT
It is possible to 'configure' IIS to use a symbolic host name in a redirect
even if the web browse does not send a host header.
But the GUI for 'ease of use in configuring' the thing is so deep
and confusing that it creates a lot of fear. One warning here,
in order to get *any* web server to do exactly what I've wanted to
do I've always had to write a bit of C code. That goes for Netscape,
and Apache, and even IIS. And it just so happens that apache is
the easiest one to write C code for.

I could give you plenty of other horror stories about IIS, just name the
version 2.0, 3.0, or 4.0

It is truly intimidating in a way that Apache will never be.

By the way, I really wish that in the Apache NT port that somebody
had figured out a way to use CreateProcess along with the WINSOCK 2.0
way of passing handles to open sockets from one process to another
in such a way that the forgiving, non-intimidating single-thread-per-process
model of unix apache could have been preserved.

I've got plenty of C programmers who will never be able to write
multithreaded code. It is just too intimidating.

The Windows NT IIS 4.0 server appears to have ways of supporting 
non-multithreaded applications programming model, which could make it
a lot easier to use for applications programmers.

Did people working on the Windows NT port of Apache look at WINSOCK 2.0
calls that I'm talking about? All I've heard are gribes about how the fork()
model is different than unix. Well, unix is somewhat unique in the fork() 
model. And sorry,
the fork() model never worked as well as it did in the days of PDP-11 with
64k program space and 64k data space. 


To: new-httpd@apache.org @ igate
cc:  
From: dgaudet@arctic.org @ igate @ IGATE
Date: 05/06/98 06:49:18 PM PDT
Subject: IIS doesn't scale

IIS insists on using the IP address of the webserver when generating
redirects for requests w/o the Host: header. i.e.

GET /blah HTTP/1.0

results in this header:

Location: http://10.0.0.1/blah/

Just imagine how badly this works behind a tcp rewriter like Cisco's
LocalDirector... so much for scalability.  According to the fellow who
told me this, microsoft claims this is a feature, not a bug.

It's nice to know that we get this right with UseCanonicalNames on or off
-- we'll use the ServerName if there's no Host: header. 

Dean
 


Mime
View raw message