Adaptively Moving Time-Ceiling In a naively implemented TW, a message is evaluated as soon as it arrives at its destination gate, even if its time-stamp is extremely large. Such a message, however, will probably be rolled back later and thus the evaluation may cause other side-effects. Moving Time-Ceiling (MTC) gives the upper limit of time for suppressing both the evaluation and generation of messages with extremely large time- stamps. MTC contributes to reducing the frequency of rollback. MTC is sometimes updated into the future. However, it is quite difficult to statically determine the optimal interval between the old MTC position and the new. If the interval is too short, many processors might idle although the rollback frequency would be greatly reduced. Conversely, if MTC jumps to the distant future, the rollback frequency will hardly be reduced at all. In our simulator, the MTC interval is adaptively determined according to the rollback frequency and the effect of the previous intervals on performance (Adaptively Moving Time-Ceiling). AMTC can exploit the entire potential of parallel machines without depending on target circuits or simulation con- ditions.
![]() Figure 3: Adaptively Moving Time-Ceiling |