[TriLUG] Metrics for concurrent connections in a web basedapplication

[Scott Lundgren]

>We are currently in the process of implementing a web application on MySQL
with apache.
You've asked a number of specific questions without mentioning anything
about the web application, the expected number of concurrent users, or
expected type of use. So it's hard to give a less than general answer.

An open source tool like Jakarta Jmeter, http://jakarta.apache.org/jmeter/,
will allow you to model the page requests and database calls your
application will perform. Granted this leaves out the load such requests
will generate through the processing logic of the web applcation, but since
it sounds like its not written yet, it should give you a place to start.

>Do you have any idea about concurrent connection limits or scalability
requirements?
Depending on how the web application is implemented if the web application &
MySQL are on the same machine, you can get some performance boost by making
the database connections through the unix socket. I would implement this in
such a way that if you decide in the future you need a seperate machine for
MySQL a minor configuration change to the deployment of the application is
all that is needed.

>Like at what number does one wish to move to scsi or raid?
>Does it make sense to implement the DB on a separate machine?
>So that the web transactions are processing on a separate processor?
>Or should I simply put the DB on a separate drive on the same bus?

Model the number of users, their usage pattern (lots of concurrent requests
or singular usage) and what is the acceptable range of response time to find
these answers.

- SL