Wentong Cai

Time-space consistency in large-scale distributed virtual environments

Maintaining a consistent view of the simulated world among different simulation nodes is a fundamental problem in large-scale distributed virtual environments (DVEs). In this paper, we characterize this problem by quantifying the time-space inconsistency in a DVE. To this end, a metric is defined to measure the time-space inconsistency in a DVE. One major advantage of the metric is that it may be estimated based on some characteristic parameters of a DVE, such as clock asynchrony, message transmission delay, the accuracy of the dead reckoning algorithm, the kinetics of the moving entity, and human factors. Thus the metric can be used to evaluate the time-space consistency property of a DVE without the actual execution of the DVE application, which is especially useful in the design stage of a DVE. Our work also clearly shows how the characteristic parameters of a DVE are interrelated in deciding the time-space inconsistency, so that we may fine-tune the DVE to make it as consistent as possible. To verify the effectiveness of the metric, a Ping-Pong game is developed. Experimental results show that the metric is effective in evaluating the time-space consistency property of the game.

A scalable architecture for supporting interactive games on the internet

This paper presents a scalable architecture for supporting large-scale interactive Internet games. In order to support a large number of participants and to divide the workload, the virtual world is divided into partitions. Each partition is then assigned to a server. A client (i.e., a player or a participant) will join a server according to the position of the avatar it controls. Compared to a centralized architecture, this distributed client-server architecture is more scalable. In addition, compared to a fully distributed, peer-to-peer architecture, it also provides a means for detecting cheating in distributed games. Since interactions and accounting information must be forwarded directly to one of the servers for qualification and verification, cheating amongst distributed players of the game will be minimized. To support secured communication for interactions and accounting information as well as to speedup periodic update messages (e.g., position updates), a hybrid communication scheme using both TCP and IP multicast is used between clients and the associated server. The communication among servers is enabled by the Run-Time Infrastructure (RTI) services. The High Level Architecture (HLA) Data Distribution Management (DDM) is employed to limit the amount of communication between the servers. In addition, the Ownership Management (OM) is also employed to implement the need for transferring the avatars between servers. In this paper, the design detail of the architecture will be presented. An experimental interactive Internet game realized using the architecture will be also described in the paper.

A Cost Calculus for Parallel Functional Programming

Building a cost calculus for a parallel program development environment is difficult because of the many degrees of freedom available in parallel implementations and because of difficulties with compositionality. We present a strategy for building cost calculi for skeleton-based programming languages which can be used for derivational software development and which deals in a pragmatic way with the difficulties of composition. The approach is illustrated for the Bird-Meertens theory of lists, a parallel functional language with an associated equational transformation system.

A load management system for running HLA-based distributed simulations over the grid

Running a large-scale distributed simulation may need a large amount of computing resources at geographically different locations. These resources may be from different organizations. The simulation may run for a long period of time and the availability and amount of computing resources available may change during the course of simulation execution. Therefore, coordinating and managing resources for distributed simulation to complete the simulation efficiently and effectively is a critical issue. This paper describes a load management system for HLA-based distributed simulation. The system is constructed on top of a grid computing environment supported by Globus. The overall structure of the system is presented in the paper and how the system saves and restores a federate is also discussed in detail.

Hierarchical federations: an architecture for information hiding

In a distributed simulation, simulation components of various types are executed at geographically different locations, forming a simulation federation to create a common virtual environment. Under the High Level Architecture (HLA), information that will be produced and consumed by a simulation component is defined in its object model, and how that information is produced and consumed is well encapsulated inside the simulation components implementation. However in the current implementation of the HLAs Runtime Infrastructure (RTI), information hiding between groups of simulation components in a simulation federation is not addressed. The authors discuss how hierarchical federation architectures can be used to tackle this problem. The hierarchical federation architecture adopted in the paper differs from the existing architectures in that it is based on a hybrid approach for interoperability between simulation federations. To demonstrate information hiding using the architecture, a distributed semiconductor supply-chain simulation is also described.

Federate migration in HLA-based simulation

The High Level Architecture (HLA) promotes simulation interoperability and reusability, making it an ideal candidate to model large-scale systems. However, a large-scale simulation running in a distributed environment is often affected by the imbalance of load level at different computing hosts. Migrating simulation components from heavily-loaded hosts to less-loaded ones can solve the problem. Protocols to migrate HLA simulation federates have been introduced by various research projects. However, existing protocols achieve migration by using either federation wide synchronization or a third party host, such as FTP servers, to handle the saving and restoring of migration states. We have previously developed a framework to execute HLA-based simulations in the Grid environment with migration support as a prominent design feature. Based on the framework, a federate migration protocol that bypasses the shortcomings identified above has been developed and better migration performance is achieved. To eliminate message loss during the migration process, a counter mechanism is employed. Studies also show that federate join time contributes significantly to the migration overhead. Therefore, our protocol overlaps the simulation execution of the federate to be migrated and the join federation operation performed by the restarting federate at the migration destination. In this paper, the development of our protocol is described and experimental results with comparison to the approach using federation wide synchronization are discussed.

Efficient parallel algorithms for tree accumulations

Abstract Accumulations are higher-order operations on structured objects; they leave the shape of an object unchanged, but replace elements of that object with accumulated information about other elements. Upwards and downwards accumulations on trees are two such operations; they form the basis of many tree algorithms. We present two E REW P RAM algorithms for computing accumulations on trees taking O(log n) time on O(n/log n) processors, which is optimal.

Algorithms for HLA-based distributed simulation cloning

Distributed simulation cloning technology is designed to analyze alternative scenarios of a distributed simulation concurrently within the same execution session. One important goal is to optimize execution by avoiding repeated computation among independent scenarios. Our research is concerned with the cloning of High Level Architecture (HLA)-based distributed simulations; a federate may spawn clones to explore different scenarios at a decision point. This article introduces the cloning mechanism and the supporting infrastructure. When enabling cloning, our approach ensures the state consistency and supports user transparency and reusability of federate codes. When a federate clones, it is desirable to replicate only those federates whose states will be affected while the rest are shared among the old and new scenarios. This article discusses the theory and issues involved in such an incremental cloning mechanism, which guarantees accurate sharing and initiates cloning only when absolutely necessary. Experiments have been carried out to compare the performance of entire cloning and incremental cloning mechanisms. Experimental results indicate that the proposed approach provides correct cloning and can significantly reduce the execution time for evaluating different scenarios of a distributed simulation. Moreover the incremental cloning mechanism significantly surpasses entire cloning in terms of execution efficiency.

Adapting a supply-chain simulation for HLA

Supply-chain management covers the planning and management of material and information from the manufacturer through the distributors, and finally to the customer. With the globalization of markets, the optimization of supply-chain management becomes more and more important. Simulation of supply-chains can help in the optimization process by evaluating the impact of alternative policies. To support the reusability of existing simulation models in a supply-chain simulation, a common standard is required for enabling the interoperability amongst different simulation programs through a well-defined interface. The High Level Architecture (HLA) is an architecture for reuse and interoperation of simulations. We report our experiences on employing the HLA to support reusability and interoperability in semiconductor supply-chain simulation. Our experiments show that by fine-tuning the integration of the application with the HLA Run-Time Infrastructure (RTI), considerable performance improvements can be achieved.

An alternative time management mechanism for distributed simulations

Over the past few years, there has been a keen interest in the management of time in distributed simulation environments. Previous emphasis in time management (TM) services has been based on time stamp ordering, which is both computation and bandwidth intensive. This article discusses an alternative approach to time management based on causal ordering. Traditional causal ordering protocols incur a large amount of communication overhead, which is generally of the order of N2 for a distributed system of N processes. A new causal ordering protocol proposed by the authors, the Modified Schiper-Eggli-Sandoz (MSES) protocol, is presented in this article. This new protocol minimizes the control information overhead of causal ordering by using the direct dependency tracking technique. The MSES protocol works well in both unicast and multicast environments, without relying on information about the underlying network topology and communication pattern among the processes of the distributed system. The MSES protocol has been successfully implemented as a middleware on top of DMSO RTI. Experiments have been conducted to benchmark the performance of the new time management mechanism with respect to the existing TM mechanisms available in DMSO RTI. The simulation scenarios of the experiments vary with different degrees of inter-federate dependency and federate event granularities. The ordering limitations of the causality based TM mechanism are addressed in this article and the trade-off of the degree of event ordering and execution speed of simulations is discussed.