بررسی تاثیر خدمات مدیریت زمان بر موتور شبیه سازی شبکه های پتری بر اساس HLA

Distributed simulation of Stochastic Petri Nets (SPN) has been considered for years as an attractive idea for increasing the effectiveness of execution of complex models. SPN concept exhibits an inherent parallelism and it was expected that exploiting this property will result in a remarkable gain in speedup. The research has focused on methods for partitioning a single model in order to distribute submodels over several nodes and on algorithms of controlling communication between submodels as well [1], [6], [7], [23] and [27]. However, the results were rather not encouraging in terms of obtained speedup due to the overhead introduced by the controlling algorithms. Moreover comparisons of algorithms, independent from simulation platforms and communication methods used, were very difficult because of the lack of appropriate metrics of speedup. New technologies (fast networking, WEB) however put the task of distributed simulation of SPN again on the list of attractive research topics. This time the aim of distributed execution has changed a little bit. The emphasis is no longer on speedup (although still important), rather on the possibility of the construction and execution of complex models from heterogeneous components developed and run on different servers. We have developed a Simulation Engine for SPN using the new technology offered by High Level Architecture (HLA). We are interested in studying the impact of HLA on the effectiveness of distributed simulation. We also believe, that the primary importance of HLA could be in using it as a reference platform for the comparison of algorithms for partitioning models and distributing them among the network. Generally the main difficulties in the area of distributed simulation [9], [10] and [21] are to be found in both the management of events without causality violation and communication between the distributed components which compose the simulation model. To cope with these difficulties among others, the HLA concept has been developed by the Defence Modelling and Simulation Office (DMSO) of the Department of Defence (DoD). The main aim of HLA is to support interoperability [26] between simulations and reusability of components. HLA is a component integration standard for distributed simulation and now an Institute of Electrical and Electronics Engineers (IEEE) standard No. P1516 by the Simulation Interoperability Standards Organization (SISO) since October 21, 2000.

We have described an HLA-based Simulation Engine which has been developed for experimental and research purposes. The engine can be used to build a HLA federation composed of any SPN-Federates. SPN-Federates are obtained either by decomposing a single SPN-model or by using existing SPN-models which can then be aggregated to a single HLA federation. RTI time management services have been applied for the implementation of this engine. For communication we have developed two types of messages, TSO and RO. These messages are passed from one SPN-federate to another over RTI. A SPN-federate in a given SPN-federation (by means of this engine) operates time-regulating and time-constrained. A Manager federate is always provided in this engine to control the synchronization start point of the SPN-federation. The tests and experiments show that the HLA-based engine we described matches the HLA policy properly (FEDEP, HLA-rules, OMT, RTI, etc). The degradation of the distributed simulation speed against the sequential simulation still remains an open question. The subnet topologies we tested demonstrate that the RO messages had not induced a negative effect in the simulation run time for distributed simulation. It remains a challenge for future research to look for methods which would enhance the speedup of the simulation even though the acceleration is not the main goal of the HLA. Our approach provides an easy way of constructing and simulating different SPN partitions of a SPN model using HLA infrastructure. We observe that HLA is a convenient technology to test and compare diverse decomposition algorithms. One of the lessons we learned developing this simulation engine is a way to cope with the causality violation which was due to the small look-ahead. A correct setting of the look-ahead results in more parallelism in the distributed simulation.