In this project, you are asked to build a parallel queuing network simulator on a network of workstations implementing either of the deadlock detection algorithms for communication deadlocks which we discussed in class. A queuing network simulator is consists of a collection of discrete event simulators which pass events between themselves. Each of the discrete event simulators constitutes a process in the distributed simulation. You will make use of a blocking (i.e. conservative) algorithm in the queueing network simulation. This means that if one of the queues at a process is empty, then the process blocks itself until it receives a message in the empty queue. This blocking is the cause of the deadlocks which occur in the simulation. This is why you will be using the deadlock detection algorithm.

Here is how you organize the project. First you write a simulation without any message passing at all, i.e. a one node simulation. Then you include message passing into the picture. Then you include the knot detection algorithm into the picture (last but not least!!!).

How do you write the one node simulation? Simple, you create events and routines which proess the events.You put the events into a priority queue ordered by their time-stamps.You will need two events, send_message, receive_message, to start with. The routine which processes the send_event figures out how much (logical) time to add to the time-stamp of the event which it selects. The time that it adds is a constant (we will discuss the details of how to figure out this constant in class).

How do we extend this to multiple workstations? First, we use MPI to send messages (which contain events) between the workstations. Then we implement the blocking algorithm, and finally we implement the knot detection algorithm. When the time comes, we will briefly discuss MPI in class. Reading materials are indicated on the course home-page. (P> The trick to doing this project is to follow the above recipe for doing it, i.e. first the single proces simulator, then the multiple process simulator. We will devote class time to each part of the project as the semster progresses.

Last but not least, the project is due on Friday, November 27 at 5 PM. You will have to mail in the (well commented) source to the TA. We will make appointments to demonstrate the program to the TA. We will sort out these details as the time approaches.