A Null Message Count of Conservative Parallel Simulation
Ha Yoon Song - Hongik University
Sung Hyun Cho - Hongik University
Sang Yong Han - Seoul National University
Extensive research has been done on the performance of simulation systems. A
conservative simulation system has been widely used as one of the scalable
high-performance simulation system. In this paper, a performance analysis of
the conservative parallel simulation protocol is presented. We analyze several
performance factors for the conservative simulation of a simulation model with
an arbitrary number of logical processes. The analysis probabilistically
identifies the critical path in a conservative model, and also estimates the
number of null messages, which is the major overhead factor of a conservative
simulation. Apart from the factors of hardware platform on which a simulation
system is running, null message count is completely based on logical factors
such as factors of simulation topology and simulation protocol. Null message
count can be estimated in terms of lookahead of logical processes, simulation
time, and the connectivity of simulation topology.  Several verifications with
experimental results have been presented to show the validity of the null
message count model. Identifying the null message count and analyzing it with
several factors from the conservative simulation protocol and simulation model
will show a new idea to revise the simulation model in order to minimize the
number of null messages and to improve the performance of a conservative
parallel simulation.
Last update: Wed Jun 12 14:26:54 2002 WEST