Sources of Error

The sources of error in the simulation model and its implementation are:

1. The use of runtime measurement to predict the execution time of local code blocks. As discussed in Section 4.4, runtime measurement is less than accurate since it cannot account for the cache perturbation effect by the simulator code and data, among other factors. Privatization also takes away from the accuracy of runtime measurement. On the whole, this problem is more an implementation problem than an inaccuracy in the simulation model.
2. The analytic model of the communication protocol assumes a fixed transmission delay for each message, and hence does not take into account software and hardware queueing delays. This problem is an inaccuracy of the simulation model.
3. The communication model does not charge a sender, in terms of simulation time, for sending a message. Consequently, subsequent events may be assigned a lower simulation timestamp than they should. This problem is also an inaccuracy of the simulation model.
4. The communication model charges a receiver for the transmission delay of a message only at the point of message acceptance, not at the point of message reception. For example a message with a receive timestamp 50, which is accepted by a receive statement at time 150, would have incurred some delay at the receiver even at time 50. Consequently, events after time 50 will be assigned a lower timestamp than they should. This is also an inaccuracy of the simulation model.