Re: [TCLUG:16088] High-load websites

[Luke Francl <fran0382@tc.umn.edu>]

On Wed, 12 Apr 2000, Yaron wrote:

>   Hi,
> 
> I'm sure I've seen articles about this before, but I can't FIND any.
> 
> Anyone know a place I can find an overview of how real high-load websites
> (like Yahoo) handle the heavy load? I'm looking for # of servers,
> backends, round-robin things, http-accelerators, etc. 
> 
> Anyone have any pointers to this?

I'd recommend what I feel is the number one site for information like
this: photo.net. In particular, you might want to check out: the ArsDigita
Server Architecture:
http://photo.net/wtr/arsdigita-server-architecture.html

This primarily regards database-driven websites, but which high load
website isn't?

Another tip I've seen: Using a dedicated, fast, threaded webserver for
static content (Zeus, Roxen, thttpd, or khttp (the kernel http server --
very fast!)) like images. Don't ever use CGI scripts.

Also, for very high demand websites, Apache may not cut it due to the fact
that it is not threaded. Apache pre-forks child web servers, which is
good, but under heavy load, Apache has no choice but to do expensive
fork() system calls. Other choices for web servers can include: AOLserver
(nee NaviServer), which includes the ability to embed TCL scripts into
webpages a la PHP (but predating it by 3+ years); Roxen, which I believe
has embedded Pike scripting; and Zeus, which I know nothing about...

But anyway, read that web page. Lots of good info there.

Luke