I am having some problems with the setup of a very simple queue. I can't seem to get the average queue time to show any sensible values. The average queue time ends up around 1E-5 on every simulation. The average service time around 300 seconds and average queue length of 30 people. There are two classes defined which are treated in the same queue. It doesn't matter if I measure one class or the other, or both classes.
The intensity of arrival is defined in seconds and service time in seconds.
What could be the error? Are there any good ways of finding out what could be error?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi, several parameters are missing. Are the service times of the two classes the same (300s)? What about the two arrival rates? Are the distributions exponential? Just to make a simple example, let me consider the parameters you gave as a single class model. With N_queue=30 and S=300s you have about R=9615s and Queue_time=R-300s, so very different from the value you got. The arrival rate lambda is 0.003239j/s.
Why you do not start to model a single class queue and then you increase the number of classes later?
Best regards, Giuseppe
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
The classes have fairly the same arrival rates around 0.1j/s and we assumed expontential. Forgot to add that we had 100 servers. I tried a simpler example and decreased the number of servers to around 30. Now we get a reasonable queue time around 60 seconds. The problem lied in that there were to many servers to get any queue time at all.
Thanks for the quick response.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
pay attention because looking at your data it seems to me that your system is in saturation. Indeed, using a very rough approximation of your model, i.e., splitting evenly your arrival rate of 1j/s between 30 independent queues M/M/1 with S=300s we obtain that the utilization of each station is 1. Of course, M/M/30 provides better performance than 30 M/M/1 in parallel, but still you should be very close to saturation. Furthermore, you also say that you are considering two classes of jobs.
Suggestion: you may use also JMCH since it solve analytically M/M/c stations
Best regards
Giuseppe
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
I am having some problems with the setup of a very simple queue. I can't seem to get the average queue time to show any sensible values. The average queue time ends up around 1E-5 on every simulation. The average service time around 300 seconds and average queue length of 30 people. There are two classes defined which are treated in the same queue. It doesn't matter if I measure one class or the other, or both classes.
The intensity of arrival is defined in seconds and service time in seconds.
What could be the error? Are there any good ways of finding out what could be error?
Hi, several parameters are missing. Are the service times of the two classes the same (300s)? What about the two arrival rates? Are the distributions exponential? Just to make a simple example, let me consider the parameters you gave as a single class model. With N_queue=30 and S=300s you have about R=9615s and Queue_time=R-300s, so very different from the value you got. The arrival rate lambda is 0.003239j/s.
Why you do not start to model a single class queue and then you increase the number of classes later?
Best regards, Giuseppe
Hi,
The classes have fairly the same arrival rates around 0.1j/s and we assumed expontential. Forgot to add that we had 100 servers. I tried a simpler example and decreased the number of servers to around 30. Now we get a reasonable queue time around 60 seconds. The problem lied in that there were to many servers to get any queue time at all.
Thanks for the quick response.
Hi,
pay attention because looking at your data it seems to me that your system is in saturation. Indeed, using a very rough approximation of your model, i.e., splitting evenly your arrival rate of 1j/s between 30 independent queues M/M/1 with S=300s we obtain that the utilization of each station is 1. Of course, M/M/30 provides better performance than 30 M/M/1 in parallel, but still you should be very close to saturation. Furthermore, you also say that you are considering two classes of jobs.
Suggestion: you may use also JMCH since it solve analytically M/M/c stations
Best regards
Giuseppe