Concurrent users or page hits per second

People that work in performance test usually want to know how many users can run concurrently in a system in order to produce a capacity planning document to help engineers to setup new environments correctly.

The big questions to answer in order to write the document are:

- How many users can support the system by each CPU?

- How many transactions per minute can support the system by each CPU?

- How many app servers, JVMs, I need to support a certain number of users?

- How is the response time given a certain number of users?

etc, etc.

But an item to clarify talking about performance is "how to define the number of concurrent users in the system?"

A good practice to define the number of concurrent users of a web interface is to calculate how many page hits has been requested along the time of the test execution and divide it by the total number of the seconds.

That Page hits per second is the a good representation of the concurrency.

So even if your test is running with 1000 users logged in the system that could be not the right number to represent the concurrency. Yes sure, 1000 users are running their scenarios but probably they are spread along the work time, but not the entire day and continuously. Consider that each user when interacts with a web interface some of the time thinks, start mumbling what is the next step to do. Real users are not like machines so they are not active 100% of the time. It is for that the real number to take in consideration is the total number of requests to the server and not the total users that are logged in the system.

"Credits to a workmate that inspired me for this post. Thanks Pier!"