Experimenter is not coming up with results

Question

question

Fernando Q3 asked Oct 02 2021 at 4:19 AM Ryan Clark commented Oct 12 2021 at 9:42 PM

Experimenter is not coming up with results

1er Caso de Experimentación 21-2 Network V4.fsm 1er Caso de Experimentación 21-2 Network V5.fsm

Hi,

I am just working with more than 10 parameters but the experimenter does not show up results properly. I attached two models with different parameters in each of one. Only the content at "queue D" seems to work out. ¿What could be the cause?

Software Version:

FlexSim 21.2.3

experimenter crashing

1er-caso-de-experimentacion-21-2-network-v4.fsm (232.2 KiB)

1er-caso-de-experimentacion-21-2-network-v5.fsm (269.8 KiB)

· 1

Answer 1 · 2021-10-04T07:43:49Z

Felix Möhlmann answered Oct 04 2021 at 7:43 AM Fernando Q3 commented Oct 05 2021 at 7:42 PM

Your model is reaching a deadlock state before reaching the warmup time. The mill and drill want to send the item back to "queue t", but it has reached the maximum content. As a result no more items are processed. This happens in both models, but later if the maximum content is larger.

When the warmup time is reached the statistics are reset and because nothing is happening in the model, almost no new values are generated.

As a possible workaround I use a custom pull requirement in the "queue t" to limit the items it can take from the buffers to the value set in the parameters, but always let items from the mill and drill enter. (Parameter is not directly linked to maximum content anymore and the max. content is set at 100)

1er-caso-de-experimentacion-21-2-network-v4_workaround.fsm

1633333419510.png (214.0 KiB)

1er-caso-de-experimentacion-21-2-network-v4-workar.fsm (1.4 MiB)

· 4

Fernando Q3 commented · Oct 05 2021 at 5:08 PM

Thanks a lot. However, this experiment was focused to set how large queues need to be in order to avoid blocking o deadlock state as you call it. So, as large the simulator tries, some replications should less o zero blocking, and it does not happen. I figured out that the node value "performuse batching" on the tree does not work properly, but "do batching" does by clicking with the sample.

0 ·

Felix Möhlmann Fernando Q3 commented · Oct 05 2021 at 5:43 PM

If the average input rate is higher than the throughput the machines can handle, the queue will always eventually fill up and lock the model. You won't find a value for the maximum content that allows the model to run indefinitely.

I just set the max. content to 1000. At a time of 20000min (the currently set stop time of the experimenter), there were just over 400 items in the queue. At 40000min the model locked again.

Changing the value of "usebatching" in a queues node to something other than zero does correctly activate the batching function for me. Maybe you forgot to adjust the "On Set" option to update the reference when the value is changed? That only happends automatically if you choose "Set Node Value" when sampling the node.

0 ·

1633455722245.png (14.6 KiB)

Felix Möhlmann Felix Möhlmann commented · Oct 05 2021 at 7:30 PM

After checking the numbers; yes, the model is indeed balanced on average. My impression fooled me, sorry for that.

It seems that the needed values to consistently avoid locking the model are higher than the currently set parameter limits in your models.

I also still don't experience any issues with setting the node value. If I double the max. content of the queues in your v5 model and measure the average staytime of items in the buffer queues, I get results that are consistent with the chosen parameters for each scenario.

(Here: staytime in scenario 2 three times higher than in scenario 3 due to waiting for three further items arriving instead of one. Also note that there are still some failed replications)

There seems to be nothing wrong your models.

1er-caso-de-experimentacion-21-2-network-v5_1.fsm

0 ·

1633462069005.png (39.3 KiB)

1er-caso-de-experimentacion-21-2-network-v5-1.fsm (3.4 MiB)

Fernando Q3 commented · Oct 05 2021 at 6:15 PM

Felix, the system is steady state, so it reaches a max content without blocking. I had tried "binary" with the right set node value reference but it did not work out. That is why I published my question.

0 ·

question