240 likes | 451 Views
A Generic Architecture for Large-Scale Distributed Simulations. Stephen J. TURNER School of Computer Engineering, Nanyang Technological University, Singapore 639798 ASSJTurner@ntu.edu.sg. Overview. Parallel and Distributed Simulation Example Application Areas Research Issues
E N D
A Generic Architecture for Large-Scale Distributed Simulations Stephen J. TURNER School of Computer Engineering, Nanyang Technological University, Singapore 639798 ASSJTurner@ntu.edu.sg
Overview • Parallel and Distributed Simulation • Example Application Areas • Research Issues • A Generic Architecture • Federating Parallel Simulators • HLA vs. Customized Protocol • Hierarchical Federations Architecture • Conclusions and Future Research
Parallel and Distributed Simulation • Parallel Discrete Event Simulation • Aims to reduce the execution time of large discrete event simulations. • The simulation model is partitioned into a number of Logical Processes (LPs) that are executed in parallel. • As each LP has its own event list, a synchronization protocol is required to preserve causality: • A conservative protocol strictly avoids the occurrence of any causality error. • An optimistic protocol detects and recovers from causality violations.
Parallel and Distributed Simulation • Distributed Simulation • Provides a way of linking simulation components (federates) of various types at possibly different locations to create a common virtual environment (federation). training federate: real-time execution constructive federate: time-stepped execution live component: real-time execution with hard deadlines constructive federate: event driven execution
Parallel and Distributed Simulation • Causality Violation real world simulated world event message “fire” Simulator A (artillery unit) the observer should see the artillery unit fire before the target is destroyed. Simulator B (target) “target destroyed” Simulator C (observer) Time (wallclock time)
Example Application Areas • Battle Simulation • Links different types of forces at multiple physical locations to create a realistic and complex virtual world. • Multi-player Internet Games • Requires massive multi-player (~10,000) virtual world. • Air Traffic Control • Simulates airports and airspace sectors to provide “faster than real-time” simulation for alternative scenario analysis. • Supply Chain Management • Covers the planning and management of material and information flow, from the manufacturer through the distributors to the customer.
Inventory information Wafers Example Application Areas • Supply Chain Management • With the globalization of markets, factories and distribution centres in a supply chain simulation may be dispersed across many different countries. Shipment to Customers Wafer Fab 1 ICs Wafer Fab 2 Assembly & Test Wafer Fab 3
Research Issues • Fast Execution • The execution time of large-scale simulations may be unacceptably large due to the detail and complexity. • Fast simulation is required for “what if” and alternative scenario analysis. • Reuse and Interoperability • Large-scale simulations are constructed by linking together existing simulation models to form a simulation federation. • These component models may have been implemented using different languages or packages and developed for different hardware platforms.
Research Issues • Geographical Distribution • A large-scale simulation may involve linking a number of simulation components that are geographically distributed. • Scalability • As the number of simulation components and the size of the network increases, the run-time system should be able to handle the communications effectively. • Data Security • A group of simulation components may need to share some sensitive information with each other while hiding that data from other simulation components in the federation.
A Generic Architecture • Generic Architecture • A generic architecture for large-scale distributed simulation is being developed to investigate these research issues. • As these research issues are common to many application areas, the architecture is not restricted to any particular application. • The research issues are addressed by: • Federating Parallel Simulators • High Level Architecture • Hierarchical Federations Architecture
Federating Parallel Simulators • Hybrid Distributed/Parallel Simulation • This is a distributed simulation architecture where one or more simulation components (federates) is itself partitioned into LPs which are executed in parallel. • Addresses research issue of fast execution. workstation Our distributed supply chain simulation can be speeded up by executing the Assembly and Test facility as a parallel federate workstation workstation multiprocessor multiprocessor
HLA vs. Customized Protocol • Option 1: High Level Architecture (HLA) for Simulation • HLA is designed to support reuse and interoperability of simulation models through its rules, interface specification and object model template.
HLA vs. Customized Protocol • Features of High Level Architecture • Each federate has a simulation object model (SOM) that defines the data it is willing to share with other federates. • The federation (set of federates) has a common federation object model (FOM). • With its capability defined by its SOM, a federate may be reused in different federations. • HLA is designed to support distributed simulations linking the federates of a federation over a LAN or the Internet. • Time Management can be used to ensure the correct ordering of events. • HLA is an IEEE (1516) and OMG standard.
HLA vs. Customized Protocol • Option 2:Customized Distributed Simulation Protocol • As each federate has its own simulation time, a synchronization protocol is required to preserve causality. • A customized distributed simulation protocol can be developed based on an existing protocol such as the conservative “null message” approach. • A standard message passing library such as MPI can be used for communication over a LAN or Internet. • Encapsulation of information within the federate can be achieved by specifying the interactions and data that can be sent as external events.
Simulation over the Internet • Simulation using HLA-RTI between Singapore and UK • Three Sun workstations at NTU + SGI & Sun at Oxford.
Hierarchical Federations Architecture • Cluster Based Architecture • In many distributed simulation applications, the individual federates are found to be organized into groups. • Communication traffic within a group is generally higher than that between groups, due to closer physical and logical proximity. • Related federates can therefore be grouped into clusters, where each cluster is supported by a high-speed communication link. • Each cluster has its own RTI session, with an application gateway connecting it to other clusters.
Scalability • Gateway Data Filtering • The transmission of irrelevant data between clusters can be avoided since data filtering algorithms can be implemented efficiently at the individual gateways. • Gateway Packet Bundling • Packet bundling techniques can be implemented at the gateways to reduce the bandwidth requirements. • Time Management • Hierarchical federations can provide more efficient time management as the federates are more loosely synchronized and lookahead restrictions may be relaxed.
Reuse and Interoperability • Heterogeneous Federations • A cluster based approach can support heterogeneous federations, each with its own FOM: • Different federations/clusters may have FOMS at different levels of resolution. • Heterogeneous FOMS allow interoperability with legacy simulations, where it is infeasible to develop new FOMS. • Heterogeneous RTIs • Different RTI implementations may be used for different clusters within the hierarchy: • A single RTI might not support all the hardware platforms used. • Some clusters may benefit from specialized RTIs.
Data Security • Information Hiding • A gateway can provide information hiding, by filtering out sensitive data that should not be transmitted to other clusters. • While the HLA allows data to be encapsulated within the SOM of a single federate, it does not provide a mechanism whereby a subset of federates may share information. • Hierarchical federations allow a group of federates within a cluster to share sensitive information without making this visible to federates outside the group. • For information hiding, a hybrid gateway/proxy architecture has been developed.
Conclusions • Federating Parallel Simulators • Can provide a solution to the requirement of fast execution. • High Level Architecture • Facilitates reuse and interoperability of component models and supports geographically distributed simulations. • Hierarchical Federations Architecture • Improves the scalability of distributed simulations through bandwidth reduction and improved time management. • Increases the reusability and interoperability through heterogeneous federations and RTIs. • Provides data security through information hiding.
Future Research • Consistency • Maintaining a consistent view of the virtual world. • Characterization of inconsistency and the development of techniques for latency hiding. • Time Management • Efficient simulation even with zero or very small lookahead. • Development of alternative mechanisms, e.g. causal ordering. • Reuse and Interoperability • Tools to support reuse and interoperability at the semantic level. • Automatic generation of gateways in hierarchical federations. • Verification and validation of hierarchical federations.
The End Thank You! HLA-RTI Internet Questions? Further Information htttp://www.ntu.edu.sg/home/ASSJTurner