On Fri, 21 Jul 2000, Reinoud wrote:

> > This depends on how much work you have to do to serve a single page.
> > If it's just HTML files, then it should be OK (a P100 with 32MB can serve
> > up to 10000 rq/hr). Now if you have to run cgi scripts, then it depends on
> > how much compute power it needs.
> > 
> 
> It's more like 10000 requests simultaniously, even up to 50000.
> The database server is indeed an other machine but connected to the
> webserver. That is, seperated by a firewall and content filter.
> 
	Hmm - you have confirmed that the database server can handle that
	many simultaneous queries?

	The simultaneous number is a hard value to judge - it varies
	based not only on the load but on the size of the data,
	speed of the remote link, random other latency issues and much
	more.

	10,000 simultaneous connections could be:
		   200,000 hits per hour, each averaging three minutes.
		12,000,000 hits per hour, each averaging three seconds.

	Hits per minute or second is probably a better metric.

> > I guess you want to filter on contents only for some protocols, rigth ?
> > proxy softwares may allow you to do this.
> > 
> 
> Okay, any good proxy software you can advise me?
> To point is more like, ppl can do queries on the website which will be
> send to the database server. Now, i wish content filtering to check first
> that ppl wont send very strange queries to try if its vunerable to
> overflows and other stuff, especially in perl so much is still possible.
> This is more the kind of filtering i wish.

	Get a bunch of machines setup with apache with either mod_perl
	or a fastcgi program that filters the queries, then passes them
	onto the database. Then either have the DNS entry point to
	multiple IP addresses (cheap), or put them behind a load balancing
	switch that makes them all appear as one IP address (more flexible).
	You can scale later by just adding more boxes. In the switch case
	you can even have machines go down with minimal disruption.
	
	Another small gain could be to put two network cards in each box -
	one for httpd requests, the other to talk to the database server.

	... Though I think that may be database servers by the time you
	are done.

	Sounds like it will be fun to setup.

                David/absolute
			       -- www.netbsd.org: A pmap for every occasion --