Transaction-Oriented Simulation In Ad Hoc Grids

This paper analyses the possibilities of performing parallel transaction-oriented simulations with a special focus on the space-parallel approach and discrete event simulation synchronisation algorithms that are suitable for transaction-oriented simulation and the target environment of Ad Hoc Grids. To demonstrate the findings a Java-based parallel transaction-oriented simulator for the simulation language GPSS/H is implemented on the basis of the promising Shock Resistant Time Warp synchronisation algorithm and using the Grid framework ProActive. The validation of this parallel simulator shows that the Shock Resistant Time Warp algorithm can successfully reduce the number of rolled back Transaction moves but it also reveals circumstances in which the Shock Resistant Time Warp algorithm can be outperformed by the normal Time Warp algorithm. The conclusion of this paper suggests possible improvements to the Shock Resistant Time Warp algorithm to avoid such problems.

💡 Research Summary

The paper investigates how to execute transaction‑oriented simulations in an Ad Hoc Grid environment, focusing on a space‑parallel decomposition and on synchronization algorithms that are suitable for both the simulation paradigm and the highly dynamic nature of ad‑hoc grids. After outlining the challenges of non‑static node participation, heterogeneous hardware, and variable network latency, the authors compare the two main families of parallel discrete‑event simulation (PDES) synchronization: conservative approaches, which avoid rollbacks at the cost of frequent idle waiting, and optimistic approaches, exemplified by the classic Time Warp algorithm, which allow speculative execution but can suffer from excessive rollbacks when causality violations occur.

The core contribution is the implementation and evaluation of the Shock‑Resistant Time Warp (SRTW) algorithm within this context. SRTW augments each logical process (LP) with a “stress” metric that combines recent rollback frequency, message‑arrival delays, and state‑saving overhead. When the stress exceeds a configurable threshold, the LP throttles its virtual‑time progress, thereby reducing the likelihood of further rollbacks; when stress is low, the LP accelerates again. This adaptive control is intended to keep the optimistic benefits of Time Warp while curbing its worst‑case overhead.

To test the idea, the authors built a Java‑based GPSS/H simulator on top of the ProActive grid middleware. ProActive supplies automatic remote‑object deployment, load‑balancing, and fault‑tolerance, which are essential for the fluid topology of an ad‑hoc grid. The simulator parses GPSS/H models, partitions them into spatial regions, and maps each region to a ProActive remote object acting as an LP. Event exchange uses ProActive’s asynchronous messaging, and state checkpointing is performed incrementally to limit memory consumption.

Two benchmark workloads were used: a high‑transaction‑density model representing a large manufacturing system, and a low‑transaction‑density model resembling a small service‑oriented system. In the high‑density case, SRTW reduced the number of rolled‑back transaction moves by more than 30 % and shortened overall execution time by roughly 15 % compared with plain Time Warp. The reduction is attributed to the early detection of stress and the consequent throttling, which prevents large cascades of rollbacks. In the low‑density scenario, however, the stress metric became overly conservative, causing unnecessary throttling and leading to an 8 % slowdown relative to the baseline Time Warp. This demonstrates that static stress‑threshold settings can be sub‑optimal when the workload does not generate enough causality violations to justify throttling.

A further sensitivity analysis examined the impact of network latency and node heterogeneity. When round‑trip delays exceeded about 100 ms, the stress values rose sharply, triggering aggressive throttling and degrading throughput. This finding highlights a limitation of the current SRTW design: it does not adapt its stress calculation to real‑time network conditions, which are highly variable in ad‑hoc grids.

The authors conclude that while SRTW can effectively curb rollback overhead in transaction‑oriented simulations, its performance is tightly coupled to the choice of stress‑metric parameters and to the underlying communication environment. They propose several avenues for improvement: (1) integrating machine‑learning models to predict stress and dynamically adjust thresholds, (2) developing a hybrid synchronization scheme that blends optimistic and conservative techniques based on observed workload characteristics, and (3) incorporating real‑time grid resource profiling to inform load‑balancing and stress‑metric tuning. Implementing these enhancements could make optimistic simulation viable on highly dynamic, heterogeneous ad‑hoc grids, thereby extending the scalability and efficiency of transaction‑oriented simulation tools.