Multi Core Processor

It has been said before that Flexsim can only truely work on a single core due to nature of it's process.

BUT

If we can develop the DS version of Flexsim that can use the spare capacity of multiple PC's then can we not have a version that uses multiple cores inside just one PC?

Has this thought gone through anyone else's head.......

Alistair,

First we need to understand what we mean by discrete event simulation. This basically means that you have a series of seperate (discrete) events that are simulated at specific times. Flexsim does this by keeping a list of events that need to occur at given times and then simulating those at the times specified. You can get a feel for this by hitting the step button to step though the event list instead of the run button.

From a distance it seems simple enough, just send different events to different cores/processors and problem solved. Eventually however, you will stumble upon the facts that you can not simultaneously calculate two events for the following reasons:

1. Event A might create more events that happen before Event B (We'll call these #1 Events).
2. Event B might depend upon Event A or any of the #1 Events that Event A created. Therefore, that Event would need to be calculated before Event B.
3. There is no way of knowing if an Event will be dependent upon other Events or if an Event will create other Events before the Event is calculated.

Now we come to the question: "Well then how does DS do it?" To be honest I have no clue as to exactly how Eamonn has programmed DS. I do know that however he has done it, it has required changes that make it impossible to use a Flexsim model in DS or a DS model in Flexsim. The two technologies are drastically different.

When I have had this (Multi Core with Flexsim) conversation with others the only answer we have come up with is to somehow split off the parts of a model that don't interact with each other or interact very infrequently and run each part on a different core. You can then sync these parts every so often so that the simulation looks like it is running at the same speed. However, this solution would be exceptionally difficult to implement and would require our users to learn an almost entirely different way of modeling because they would have to think about how to split their models up. And that is sounds a whole lot like the conversion to DS .

I know that this answer is long and others here might be able to get you a little better answer or correct anything I have posted here but I felt that the simple answer wasn't what you were looking for. I do know that we are actively looking for ways to use multiple cores and we will eventually make use of them, we just aren't there yet.

I hope this helps,
Brandon