JiST: Embedding simulation time into a virtual machine

Since progress in many avenues of science depends heavily on simulated results, discrete event simulators have been the subject of much research into their efficient design and execution. This paper introduces JiST, a Java-based simulation framework that executes discrete event simulations both efficiently and transparently. Our system differs from existing work in that it embeds simulation time semantics into the Java execution model, but does so without inventing a new language, without requiring a specialized compiler and without utilizing a custom runtime. The result is a flexible simulation environment that allows sequential simulation execution and also transparently supports both parallel and optimistic execution with automatic checkpointing and rollback. The JiST approach uses a convenient single system image abstraction across a cluster of nodes, that allows for dynamic network and computational load-balancing and fine-grained migration of simulation state. The system provides standard benefits that the modern Java runtime affords, such as type-safety, garbage collection and portability. Nevertheless, JiST performs well, either matching or exceeding the performance of ns2 and the highly optimized GloMo-Sim runtime in both throughput and memory consumption. We illustrate the practicality of the JiST framework by applying it to the construction of SWANS, a scalable wireless ad hoc network simulator