Application Response Time - Thresholds

[Harvey]

This is related to availability management, but more of a technical question.

We have tools in our organization which can measure the response time of our applications continuously daily, from an end-users perspective.

For example we could have data which says, at 05:00 it took 3 seconds to load www.itilcommunity.com, at 05:05 it took 3.2 seconds to load the same, all throughout the days. So maybe at 13:00 when everyone is back from lunch the site is busy and it takes 4.6 seconds to load the same.

Then, we can set thresholds for this.

For example anything below 4 seconds will be "green", anything between 4 and 6 seconds will be yellow, and if the site takes more than 6 seconds to load the status will be "red" and an alert will go to the application owner.

Now, my question is, is there some scientific way that we can consistently define these thresholds?

Currently we take a chunk of data, take out the outliers, average it, and we'll say okay, this average will be "green". Then we look at the slowest it has ever responded during normal operation and make anything above that, "red". Yellow is defined somewhere above the average..

So as you can see, the method kinda makes sense... but is not something that can be consistantely repeated...

Any ideas on this?

Thanks for any help!

- Harvey

[javierarcal]

Hi Harvey,

I suggest to define your threshold somewhat of statistica procedure, your current method seems to be good, but from my point of view it may have a slight problem.

This problem could be with "singular measures" too low or too small, if for some reasons you have obtained a very low or very high response this funny measures are not represantive of the normal behaviour and should not be considered to define your threshold.

From my point of view what should be done to define your threshold is considering your data distribution as a Gauss distribution what is quite true in an standar enviroment, calculate your data average and your data standard deviation (sigma) and take as threshold:
average+sigma(90%)
average+2*sigma (95%)
average+3*sigma (99%)

Average and sigma can be obtained using any statistical package or for example throug an excell data sheet.

Hope it helps
Javier
ITIL Consultant
javier.arcal@gmail.com
Madrid-Spain