Re: [MLUG - DISCUSSION] Computer Crime

[Rick Buford <EMAIL:PROTECTED>]

Christian M. Cepel wrote:
> I'm confused...  Jerry (I think) keeps talking about how many hits per 
> so long a time period to fill up a T1. I.e., bandwidth.
>
> I got the impression that it has absolutely nothing to do with 
> bandwidth, but instead with the server daemon (httpd) and the computer 
> itself (the server) being able to service all those different sockets 
> simultaneously.  I.e., you crash the computer by telling it to run 3000 
> processes when it's designed, and reasonably expected to handle a peak 
> of 100 simultaneous processes (and the designers consider that a highest 
> load scenario), and when perhaps it can actually functionally load only 
> 1500 simultaneous processes w/o failing.
>
> Taking up bandwidth may slow a connection, but only multiple processes 
> would slow and then crash a server.  Their T1 (if they have one) could 
> be running at like 30% capacity when the attack was at it's worst.
>
> Or am I misunderstanding here.
>
> Frankly, I thought good web server daemons were designed to have limits 
> on the number of simultaneous sockets and processes running.

Well, you are right and you are wrong. To properly serve a busy website, 
you need both adequate server and network resources. The biggest server 
in the world won't help your traffic flow smoothly if you have 
inadequate bandwidth nor will the converse be a happy situation.

For example, my WAN links usually run 4-5x more outbound traffic than 
inbound. This makes perfect sense once you realize that customer 
requests to the web servers are far smaller than the response to that 
request. On the rare occasion that either the inbound or outbound side 
of the WAN links reach capacity, the customer experience will 
degrade...quickly. It's pretty obvious why cpu or memory can degrade 
performance, but if you have inadequate bandwdith for the traffic 
demands, you can just as easily have either customers queuing on the 
router or server processes locked trying to send data outbound.

Rick

_______________________________________________
discussion mailing list
EMAIL:PROTECTED
http://mlug.missouri.edu/mailman/listinfo/discussion