The trend towards multi-core and many-core CPUs is forever changing the
composition of the Beowulf cluster. The modern Beowulf cluster is now a
heterogeneous cluster of single core, multi-core, and even many-core processors.
Distributed discrete event simulations which are common tasks for such clusters
are in the position now to take full advantage of this new environment by moving
away from the single heavy-weight processes and converting to a many thread
processes capable of fully utilizing a multi-core CPU.
This research focuses on expanding the current warped discrete event
simulation kernel to take full advantage of these heterogeneous Beowulf
clusters. Through this research, multiple lock-free data structures, as well as
shared memory discrete event simulation algorithms were implemented and
analyzed. In the analysis of the first Threaded Warped implementation,
named ThreadedWarped, several key weaknesses are defined which caused
an actual worsening of the performance versus that of the single heavy-weight
process implementation. With solutions to some of these weaknesses given,
ThreadedWarped makes for a solid foundation on which to continue
research in this area of optimistic distributed shared memory parallel discrete
event simulation.