| || |
Simulating a production like test
I need to run a performance test that mimics the production usage.Iam trying to analyse the production logs to come up with a realistic transaction distribution mix for my test.
I began with collecting the following data for all the transactionsfor a period of two months.(From Sept 1st 2008 to Oct 31st 2008,its okay to use these two months as its not a shopping cart application which are usually busy during Christmas & Thanks Giving time)
-Total number of user visits for each transaction per day
-Average Time Viewed in Seconds for each Transaction
-For all Transactions Iam considering Max(Visits) for the entire two months as "Peak Usage" on a day.
Therefore, Max(Visits) will give me peak usage per day for a transaction.
Max(Visits)/24 peak usage per hour for a transaction
(Iam dividing by 24 because the application will be used across the world at different time zones)
Can anyone guide me on what should be my next steps.
Re: Simulating a production like test
Hey Smith -
A few questions, and some thoughts.
What's your perf environment like? Similar to production, nowhere close? Identical?
Next, I'd wonder what the goal of your testing is - are you looking to test performance just under normal levels of load, or to take it to the next level to do some scalability testing to ensure you can handle your busy period?
When I'm prepping throughout the year to do testing for our busy time (HR application, busy time is NOW! - we see approximately double the access during Open Enrollment season than for the rest of the year), I also analyze the rates at which we're growing so that I can estimate the kind of activity I think we'll see.
For example, I might also look up Sep/Oct/Nov/Dec of 2007 for your application (if it's still available). Then calculate the % of growth month over month - if your application hasn't seen a ton of growth, you might be able to make some assumptions from that as to the kind of growth you'll see this year from October to Nov/Dec busy time and know what you'll need to support.
A lot of it really depends on what the goal and focus of your testing is.
One caveat I'd mention to the dividing by 24 - I'd take a sample of several days across different weeks as well. It may be that even though you're a global application, a lot of your access is still focused within a particular 12 hour period, and you don't want to mislead yourself on hourly peak numbers too much.