Notes on Ciw’s Mechanisms¶
Ciw uses the event scheduling approach [SW14] , similar to the three phase approach. In the event scheduling approach, three types of event take place: A Events move the clock forward, B Events are prescheduled events, and C Events are events that arise because a B Event has happened.
Here A-events correspond to moving the clock forward to the next B-event. B-events correspond to either an external arrival, a customer finishing service, or a server shift change. C-events correspond to a customer starting service, customer being released from a node, and being blocked or unblocked.
In event scheduling the following process occurs:
Initialise the simulation
A Phase: move the clock to the next scheduled event
Take a B Event scheduled for now, carry out the event
Carry out all C Events that arose due to the event carried out in (3.)
Repeat (3.) - (4.) until all B Event scheduled for that date have been carried out
Repeat (2.) - (5.) until a terminating criteria has been satisfied
In Ciw, Type I blocking (blocking after service) is implemented for restricted networks.
After service, a customer’s next destination is sampled from the transition matrix. If there is space at the destination node, that customer will join the queue there. Else if the destination node’s queueing capacity is full, then that customer will be blocked. That customer remains at that node, with its server, until space becomes available at the destination. This means the server that was serving that customer remains attached to that customer, being unable to serve anyone else until that customer is unblocked.
At the time of blockage, information about this customer is added to the destination node’s
blocked_queue, a virtual queue containing information about all the customers blocked to that node, and the order in which they became blocked.
Thus, the sequence of unblockages happen in the order which customers were blocked.
Circular blockages can lead to deadlock.
In discrete event simulation, simultaneous event are inevitable. That is two or more events that are scheduled to happen at the same time. However due to the nature of discrete event simulation, these event cannot be carried out computationally at the same time, and the order at which these events are computed can greatly effect their eventual outcome. For example, if two customers are scheduled to arrive at an empty M/M/1 queue at the same date: which one should begin service and which one should wait?
In Ciw, to prevent any bias, whenever more than one event is scheduled to happen simultaneously, the next event to be computed is uniformly randomly selected from the list of events to be undertaken.