httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
Subject A warning about threads in commercial unix systems
Date Thu, 16 Apr 1998 20:35:51 GMT
Let me sound a warning about the threading implementations in commercial unix 
This is based on recent expertise from people inside some of the biggest 
vendors who are responsible for basic server/service infrastructure.

It's a bitch. Especially on multi-cpu machines some of the basic mutex 
simply don't work or need massive amounts of expensive debugging to get to
work reliably, or usefully on a per-application basis. Vendors still supply 
their own threading
packages for various reasons on solaris and hp-ux. Take the JDK sources for 
on Solaris machines it used to supply its own threading package because of 
its authors (employees of sun of course) had with the solaris native threads.

Of course, things change, and I'm the first to condem people
who spread FUD (fear, uncertainty, and doubt) about a technology.

Word on the street is that for whatever inconveniences and foibles people
have with Windows NT that at least the threading works consistently.

At IAC for example we run massive numbers of web transactions through Netscape 
each day, and even with the most recent fixes and support from Netscape we 
dozens of crashed HTTPD processes each day, something we did not see in
the single-threaded-multi-process older version of Netscape servers unless
there was some seriously buggy user-written NSAPI code involved.

Since netscape turns off core dumping, and doesn't record process exit status
in the error logs most people would never notice this kind of thing unless they 
turn on the
unix process level accounting and write a custom 
binary-accounting-file-scanning process.

There is nothing keeping Apache people from doing a much better job of course,
as you would expect. But when you do it, you will deserve a badge of 
bigger than you might think you deserve now.

View raw message