Looking for some clarifications

I'm hoping someone can take a moment to help clarify something for me.

I'm working on a homework assignment for my class, where we are doing the same model in both Arena and Flexsim. However, I'm getting pretty different results from the two, and I'm trying to make sure that I'm not confusing the creation distribution.

The setup is as follows:
Two items, each has it's own arrival source, it's own queue, then they share a processor, each item having it's own processing time. From there, the items go to a second queue for a second shared processor before exiting the system.

Item One arrives with a lognormal distribution wtih a log mean of 11.5 hours, and standard deviation of 2 hours. Processing time is triangular distribution of 5,6, and 8 hours.

Item Two arrives with an exponential distribution with a mean of 15.1 hours, and a processing time using a triangular distribution of 3,7, and 8 hours.

The second process is a triangular distribution of 4,6 and 8 hours.

The system is run for 5000 hours, and the performance measures are the average number in the seperated queues and the total time in system.

Now, if I'm reading the distributions right, my commands should look like as follows:
lognormal2(11.5,2,1,0)
exponential(15.1,0,0)
triangular(5,8,6,0) (just doing the first one here)

Now here's where the weirdness comes in. In Flexsim, my average queue content for items 1 and 2 bounce around from 0.10 to 0.20. However, in Arena, the are 1.80 for Item 1 and 1.50 for Item 2. And obviously the average total time in system is much higher for Arena.

I'm doing the first processor as a case by value, to return the triangular distributions.

But is there something I'm missing here? Am I reading this drop down incorrectly? The reason I ask is I've never worked with distributions as location, scale, shape or stream (honestly, I'm still looking this up. I've never heard of the stream of a distribution, or if I did, it never registered), and Arena asks for things in the terms I'm used to (mean, standard deviation, etc)

Thanks in advance for any clarifications or help.

To further clarify, I ran the models again in both simulations, this time increasing the run time to 50000 hours. Flexsim had both queue contents level off at 0.09, and Arena had 1.91 for Item 1 average content and 1.60 for Item 2.

So this implies to me that I'm misreading how the distributions operate in Flexsim.

In lognormal2, the first parameter is an extra position/offset not to be confused with the lognormal's location parameter. The log mean (also known as location) is the second parameter while the log stardard deviation (scale) is the third.

I think you need lognormal2(0,11.5,2) or if 2 was the actual standard deviation before the data was transformed, lognormal2(0,11.5,log(2)).

I'd run that in a script or write values to a table to check that you're getting what you expected.

Similarly you'd need exponential(0,15.1) for item2.


By the way I've omitted stream here as that's a separate point. You use that to say which progression of pseudo random numbers you want to use - in particular to make sure samples are independent when you make changes to the model and any distributions.

Hmm.. Ok, so I was right the first time. That's what I get for second guessing myself. But if that's the case, I can't seem to figure out why they're vastly different. Running a 50000 hour test resulted in 0.08 average queue content for Item 1 and 0.32 for Item 2.

Switch to the suggested (0,11.5,log(2)) resulted in 0.2 and 1.42 for the contents. Still far off from what Arena is putting out. But that at least eliminates one possible problem.

To be honest, I'm not sure with would be appropriate, the 2 or log(2), as it was just given to us as the log standard deviation of 2.0 hours. I think it's just 2, since it has the note that the values are the mean and standard deviation of the lognormal random variable itself..

Thanks for the clarification!

Just some ideas - did you change the processor to pull the longest waiting?

Also - how did you get your values? I noticed you'd chosen lognormal2 which is ExpertFit compatible. If you've used a different technique you might want to try lognormal.

Correct, the processor is pulling longest waiting. I figured that one out pretty quite when I saw one queue on another problem building up.

All the values were given as part of the problem statement, so it wasn't derived from a set of data we had available.