ABSTRACT
Real-time simulation is characterized by the strong constraints that must be respected regarding the time spent simulating an activity. Usually we can run simulation faster than real time, and we synchronize with virtual time with techniques that are now mastered by many engineers [1]. Sometimes, the simulation can execute at a higher rate than the actual system evolves; in this case, all that is necessary is delaying the simulation so that virtual time stays synchronized to 98the expected wall clock real time. However, when the simulation involves heavy computation or the simulated activities are very short-lived, meeting the deadlines may imply optimizing certain parts of the simulation to reduce its execution time. A simulation run that is slower than real time can occur when dealing with detailed physical models. For instance, we experienced this when simulating the details of particles for physical studies in nuclear medicine [2,3]. In these works, we faced modeling cases where it was impossible to meet the real-time constraints. Medical scanning experiments that last 20–30 minutes can be simulated in a few seconds with analytical models. However, the results obtained can be very imprecise for small tumors (up to 100% error for tumors lesser than 1 cm in diameter). To be more precise, we must build 3D spatial stochastic models, which run for more than 9 hours on our current computer, and many replicates are necessary to reduce the error (less than 10% error achieved in 3 days of computing using the European Computing Grid) [4].
