next up previous
Next: Linkages between PRC, Quantum Up: Research Thrusts for Reverse Previous: New Modeling Techniques

Comparison and Contrast of Performance

In the third research thrust area, we propose to conduct a performance comparison of PRC to TiPRC and conservative synchronization strategies from both a theoretical model prospective as well as experimental. On the theoretical model side, we are curious to know if Lipton and Mizell's model [69] under PRC conditions is still valid. Recall from the Section 2 that Lipton and Mizell's model showed that Time Warp can outperform the Chandy/Misra algorithms by an arbitrary amount. We anticipate showing that PRC like Time Warp can outperform Chandy/Misra/Bryant by an arbitrary amount.

In addition to Lipton's model, we propose to understand how PRC models fit within the Discrete-Event Dynamical Systems (DEVS) formalism [110,109,38]. It has been shown that any causal dynamical system which has piece-wise constant input and output segments can be represented by DEVS. DEVS models are constructed in a hierarchical fashion by interconnecting components which are DEVS models. The specification of the interconnection is provided in the form of a coupled model.

We also propose to explore the question of the PRC stability relative to TiPRC and conservative techniques [54] as well as understand the implications risk. We are interested to understand if PRC models are prone to rollback inconsistency and stale state problems as shown by David Nicol's now famous ``Dark Side of Risk'' paper [76]. Based on Nicol's own theorems, if a PRC model can be constructed such that it does not generate anti-messages, as in the example in Figure 2, it could potentially be demonstrated that PRC models will not suffer from rollback inconsistency.

And finally, we propose to perform a series of experiments to understand the performance implications of both individual PRC simulation runs as compared to TiPRC and conservative techniques as well as from a design of experiment prospective. Meaning, because of the PRC's capabilities, how many less experiments will have to be run as compared to TiPRC and conservative parallel simulations.


next up previous
Next: Linkages between PRC, Quantum Up: Research Thrusts for Reverse Previous: New Modeling Techniques
Christopher D. Carothers 2002-03-07