Significance of 90 Percentile response time in performance testing

Loadrunner

We all would have seen the 90 percentile response time in our analysis summary report and this percentile value can be customized according to our NFR's. In this post we will take close look at what this 90 percentile response time is and its significance in performance testing.

What is 90 Percentile response time?

It is the value for which 90% of data points are smaller (or) The permissible response time within which 90% percentage of the users fall.

Here data points refer to the transaction response time.

Lets take an example to understand this.

Consider 10 instance of Login transaction with values 2,1,5,8,3,7,10,9,4,6 in seconds

1. Sort the Response Time values in descending order 1,2,3,4,5,6,7,8,9,10 seconds respectively.
2. Remove the top 10% of the Response Time values. Here the top 10% is 10 seconds. So remove it.
3. Now the higher value left if the 90 percentile Response time. So 9 seconds is our 90 percentile Response time for Login transaction.

Significance of 90 Percentile response time:

Consider you have the NFR's for 90 percentile response time as 9 seconds and from our above example 90 percentile response time is also 9 seconds, which means our NFR is met.

• It means that 90% of the transactions or users are within the permissible limit of the 90 percentile NFR value.
• It is more accurate and dependable than the Average response time or some times called as 50 percentile value because average response time means that only 50% of the transactions or users are within the permissible limit of the Average response time NFR. 
• The average response time will always be less than the 90 percentile response time. Using the below formula find out the 90 percentile to Average response time ratio.

(90 Percentile RT - Average RT) /  Average RT

• If the ratio is small it means that 90 Percentile RT & Average RT are close enough and the application is having good and stabilized performance.
• If the ratio is high it means that applications performance is poor.