Rate Tables vs Manual Arrival Rate Changes

[akramnajjar]

In the book by SMITH/STURROCK/KELTON, in Section 7-3, the authors warn against manually changing the arrival rate from one period to the other. They state that it will give wrong results at the transitions. The recommendation is to use Rate Tables.

Is my guess correct ? Say we have 2 periods of 1 hour each, one with mean = 100 and one with mean = 200.

If we use Rate Tables, Simio will use the rate 100 in the first hour BUT will only generate arrivals during the first hour. 

If we use a manually set rate and change it to 200 after 1 hour, it is possible that in the last few minutes of the first hour, entities might be generated that arrive in the second hour.

If not correct, what is the real reason for the author's warning?

Thank you

[gocken]

from exponential distribution you can get varying numbers in a very very wide range (for example, if you are using exponential distribution for generating interarrival times in one draw (random number generation) you can get 1 (say hour) and in another draw you can get 100 hours). The authors emphasizes that this situation can cause improper results in simulations due to discrete event simulation time advancement mechanism.

Assume that you have periods each length is 20 hours. Note that, in discrete event simulation, i.e., the first event plans the next event. So, assume that in the first  period (within the first 20 hour) the first event will be created at exactly time 1 (hour) (say an arrival event). After its creation, this entity plans the next arrival (thus an interarrival time will be generated from an exponential dist), i.e., the second entity will arrive to the system at time 101 (hour, 1+100). this means that you will have only one arrival in the first period and no arrivals in 2th,3th,4th and 5th periods.
the authors says that this is an inconvenience and arrival should be planned in each period, seperately. By this way, in each period you will have at least one arrival. For this to happen you can use rate tables.  

[akramnajjar]

Thanks ...