High Performance Computing and Grid in Asia Pacific Region, International Conference on (1997)
Apr. 28, 1997 to May 2, 1997
This paper proposes a new calendar queue which can improve the conventional calendar queue's performance over uneven event distributions. A calendar queue is a multi-list priority queue frequently employed in discrete event simulations as the global event list since its performance shows O(1) time complexity. For O(1) performance, calendar queues maintain only a small number of events at each list of their multi-list by constantly adjusting their multi-list size depending on the number of enqueued events and redistributing events over the newly resized multi-list. Calendar queues, however, perform poorly over skewed event distributions. Our proposed calendar queue can reduce this conventional calendar queue's sensitivity to event distributions by adding two new mechanisms; The first mechanism constantly measures the event distribution and according to the measured metrics reconfigures calendar queues' multi-list to maintain O(1) performance even for uneven distributions. The second mechanism adopts an additional data structure to save the time wasted for frequent resizing calendar queues. Our conducted experiments show that our calendar queue can achieve more than 10-time speedup for uneven event distributions while maintaining the same performance for other distributions.
S. H. Oh, "Dynamic Lazy Calendar Queue : An Event List for Network Simulation," High Performance Computing and Grid in Asia Pacific Region, International Conference on(HPCASIA), Seoul, Korea, 1997, pp. 254.