Combining optimism limiting schemes in time warp based parallel simulations

The Time Warp protocol is considered to be an effective synchronization mechanism for parallel discrete event simulation (PDES). However, it is widely recognized that it suffers over-optimistic behavior on the part of the simulation processes that may be very harmful for performance. In current literature, two techniques have been used to counteract this problem - (i) throttling of over-optimistic processes and (ii) scheduling or load balancing. However, study of these techniques has been primarily done in isolation. We demonstrate using a parameterized simulation model of Time Warp that an appropriate combination of throttling and global scheduling using LP migration can be very beneficial for performance compared to any one of these schemes acting in isolation. This study forms the basis of the design of more powerful control schemes that use a combination of multiple techniques.