Impact of adding aggressiveness to a non-aggressive windowing protocol

In this paper we summarize the results of our theoretical investigation into the costs and benefits of extending the conservative simulation window established in a non-aggressive windowing algorithm. There are two primary cost incurred by the non-aggressive algorithm: the cost of global synchronization and the cost of blocking due to pessimistic synchronization constraints. As the conservative simulation window is extended processors are required to synchronize less often and parallelism is increased. However, the increased aggressiveness increases the costs associated with state saving and rollbacks. This is the fundamental trade-off we capture analytically.