Question on Overall Approach
Hi, I have lots of reference material and have attended a Loadrunner course (which was excellent by the way) but I would like to get peoples opinions on how to best approach the *planning* of the testing.
Using Webtrends to analyze the web server logs I have been able to identify 2 things:
1. The most common operations (.ie login, do x, do y etc.)
2) The kind of paths most common (which helps identify the user models)
All of this is showing up with great clarity (thank god for predictable customers).
I intend to create scripts to simulate the common operations etc.
My question is what approach do people take at this point?
I intend to simulate the current load on the site , then ramp up. I also plan to do failover to test the load balancing software: finally I plan to do some endurance testing running scripts over a period of days to identify mem leaks etc.
I am wondering what other people do etc.
Re: Question on Overall Approach
<BLOCKQUOTE><font size="1" face="Verdana, Arial, Helvetica">quote:</font><HR>Originally posted by Namco:
I intend to create scripts to simulate the common operations etc. My question is what approach do people take at this point?<HR></BLOCKQUOTE>
I tend to use queuing networks as my model for performance simulation by doing workload characerization and performance modeling. The details of this would probably be more than you want for your post. I mainly do this because it leads right into capacity planning exercises.
Simulating the current load is good as is ramp up - but only if you are sure of what you are looking for in the ramp up. In other words, have valid SLA's and boundary conditions that you are looking to stay within. Basically, your approach sounds good. Of course, the real key is the analysis you do after the fact. (In other words, when you see your transactions per second are going down, what are the things you check? If you see your response time climbing but your transaction rate is steady, what do you look for? If the CPU getting is maxed out in some cases with the same script but not in other cases, what do you check for? The analysis is always the most important.)
Remember to consider bursty traffic (not the same as peak load) as well as heavy-tail distributions - two things often overlooked in the world of performance. Also remember that there is a distinction, in the world of performance, between concurrent users and simultaneous users - and neither should be the basis for your metrics.