I was reading this blog again because I needed to test a login function, so thanks for that. It was very helpful.

I have two comments:

– not only does the numbering end at 18, you also seem to jump from 5 number 5 to eight. So actually there are only 16 ways to test the login function.
- Why not name it 16 ways to test the login function. I don’t agree with you that 20 sounds better. I think 16 ways to test a login function sounds just as good or even better. When I was searching this blog with google to use it today, I couldn’t remember how may ways there were, if it had been 16 I would certainly have remembered.

Hello Thanks heaps for this indeed!… if anyone else has anything, it would be much appreciated. Great website Super Piano Links http://www.en.Grand-Pianos.org Enjoy!…

Thanks for this useful information. I never heard this theory but is sounds very plausible. This is in any case a more valuable method than just guess based of feelings.

Do you know where we can find more information about this topic?

There’s a better answer to this.

We want to close a session as soon as we can “guess” that a user has gone away. We don’t have absolute certainty, but we should be able to make statements of the nature “In [confidence level] cases where the user makes no requests in [time interval], the user has departed”.

Choose your confidence level, model the usage pattern, choose the corresponding session duration.

The usage pattern can be determined from real data (assuming you have enough activity history to mine); but lacking that you can make some pretty reasonable estimates by assuming a Poisson distribution is on play. If the average rate of requests is one unit of time, then if you close a session after 2.3 units you’ll be right to do so 90% of the time. 4.6 units gives you the correct choice 99% of the time, 6.9 units gives you three nines, etc.

20 minutes is about right for user requests that are 6 minutes apart, on average, if you need 99% confidence. Or 90 seconds apart, if you need .999999.