One way to approach this is to baseline your application on hardware that you already have. You can use a load generating program like Jmeter (http://jakarta.apache.org/jmeter/) to create load that would be similar to what you experience during your peak times. It is pretty easy to record a transaction while navigating the application in different ways, and then replay those transactions to create load. You can tweak the number of simulated issues, frequency of hits, wait time, etc...while running the transactions.

This method can take some of the guess work out of it and can help to guarantee that the platform will scale. It is also a great way to identify any bottlenecks (memory, io, etc).

Just be careful to ramp the load up to realistic levels...it can be easy to misinterpret the settings for Jmeter and create ridiculous levels of load, leading you to think that the system is under performing.

Jason Hiller
Oracle Corporation

 

On 2007-10-05 23:56, Bryan O'Neal wrote:
> Ok, so I am thinking about moving some stuff in house.  In particular a
> mostly dynamic website that users some backend java and a mail server.
> The mail server is pretty low key, handling maybe 30K messages a day.
> The web server is expected to get about 300K hits a day.  How big of a
> server do I really need?  
> I was thinking Cent OS5 running Apache/TomCat/MySQL with procmail.  And
> I was thinking I could easily getaway with something like a dual core
> 3GHz(x1333FBS) with 4GB Ram and a 4 SAS/SATA disks in a RAID5 area.  
> Suggestions?  Is this going to be big enough or am I low balling it?
> 
> 
> 
> Bryan O'Neal
> Cornerstone Homes