Thanks:  0
Likes:  0
Dislikes:  0

# Thread: Formula for Scaling Users without think times?

1. ## Formula for Scaling Users without think times?

Is there a way to understand how concurrent Vusers (no think times) scale to real world users?

2. ## Re: Formula for Scaling Users without thinktimes

Not easily....

You can estimate after the fact. You would need production usage information as to how the app is used generally as a day in the life of. You would need to know the transactions, transaction rates, and hits. Then you could get an approximation. But then again, there is a lot more here than meets the eye.

What exactly are you trying to do?

3. ## Re: Formula for Scaling Users without thinktimes

I thought I just replied to this but it didn't show. Anyway..

I test with up to 100 concurrent users (license capacity) but my application is in use by hundreds more. So how can I take the metrics returned from 100 concurrent user load tests, add some numbers, divide some numbers, shake it up and spit out, that "100 concurrent users represents x real world users"

4. ## Re: Formula for Scaling Users without thinktimes

1) Run your 100 concurrent virtual users test.
3) Run your test again, this time using more and more real users, until the load on your system is similar.

Use the factor you have learned in future experiments (but don't expect it to be too precise).

5. ## Re: Formula for Scaling Users without thinktimes

Assuming this is a web app, here is the most you can say, "100 think-time-less concurrent users represents approximately 100 think-time-less concurrent real world users". Again, you could estimate as I already stated but then again, there is a lot more here than meets the eye.

This is a matter of non-linear math. The nature of networks, caching mechanisms throughout an architecture, the possibility of disk i/o at any time, and any number of i/o buffer sizes of various quantities, assure that such an extrapolation will be roughly accurate only by accident.

The best thing to do is buy more licenses. In the end, you can justify the cost. Since your company is a VAR, for your client you can get licenses dirt cheap - correct? [img]/images/graemlins/wink.gif[/img] Vuser days?

6. ## Re: Formula for Scaling Users without thinktimes

I have a 100 user license capacity Jake

7. ## Re: Formula for Scaling Users without thinktimes

Thanks for the replies, I've been trying to determine a method for designing a predictive model using the data from my tests, but seems like quite a bit more effort than I am able to bill my clients for. =)

8. ## Re: Formula for Scaling Users without thinktimes

If you come up with a method it will be an instant hit. It would seem that a tool such as this could be quickly adapted to extrapolating performance on platforms of varying computing horsepower. You can sell it and retire from the profits.

9. ## Re: Formula for Scaling Users without thinktimes

Jake, that is very interesting. I've been pondering topics for my master's thesis and maybe this is the direction I should think carefully about. There are plenty of capacity modeling applications on the market, but none that I know of that directly addresses the question I am asking about. If I make some progress on this, I'll update the board with progress.

10. ## Re: Formula for Scaling Users without thinktimes

[ QUOTE ]
1) Run your 100 concurrent virtual users test.
3) Run your test again, this time using more and more real users, until the load on your system is similar.

Use the factor you have learned in future experiments (but don't expect it to be too precise).

[/ QUOTE ]

Depending on the amount of time you wish to put into this, I would consider repeating this experiment a number of times to see what type of variance you get. Real usets do not behave in a uniform fashion like virtual users, and 100 is a pretty small number. Assuming the relationship is linear (dangerous), I would try repeating the experiment 5 times, discarding the lowest and highest, and taking the factor as the average of the remaining 3. You might want to also repeat this with at least two other sets of users, say 20 and 60, to give some indication that the relationship is actually linear. The variation in results will also give you a fair idea how reliable the factor you have created will be.

Don't assume your factor will scale well to numbers greater than 100, as extrapolative models are prone to many problems, and need a heck of a lot more input data to have any chance of being correct.

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts
•
Search Engine Optimisation provided by DragonByte SEO v2.0.36 (Pro) - vBulletin Mods & Addons Copyright © 2016 DragonByte Technologies Ltd.