Conservative Parallel Discrete Event Simulation Research Articles

Synchronization of Processes in Parallel Discrete Event Simulation

The method of parallel discrete event simulation (PDES) is one of promising methods for effective use of supercomputer systems. In this paper, the results of the analysis of synchronization models for processor elements are presented when calculations are performed by the PDES method. An analogy between the change in the local time profile of processor elements and the increase of the sample surface during molecular epitaxy is used for the classification of synchronization models. Two important classes of such models are the models of conservative PDES and optimistic PDES. The conservative model belongs to the universality class of the Kardar–Parisi–Zhang (KPZ) equation, and the optimistic model belongs to the universality class of directed percolation (DP) problems. We present the results of the analysis of synchronization models in the case when the graph of message exchange between processor elements belongs to the class of small-world networks.

Synchronization of conservative parallel discrete event simulations on a small-world network.

We examine the question of the influence of sparse long-range communications on the synchronization in parallel discrete event simulations. We build a model of the evolution of local virtual times in a conservative algorithm including several choices of local links. All network realizations belong to the small-world network class. We find that synchronization depends on the average shortest path of the network. The time profile dynamics are similar to the surface profile growth, which helps to analyze synchronization effects using a statistical physics approach. Without long-range links of the nodes, the model belongs to the universality class of the Kardar-Parisi-Zhang equation for surface growth. We find that the critical exponents depend logarithmically on the fraction of long-range links. We present the results of simulations and discuss our observations.

Simulation of virtual time profile in conservative parallel discrete event simulation algorithm for small-world network

We simulate model for evolution of local virtual time profile in conservative parallel discrete event the simulation (PDES) algorithm with long-range communication links. The main findings of simulation are that i) growth exponent depends logarithmically on the concentration p of long-range links; ii) utilisation of processing elements time decreases slowly with p. Thismeans that the conservative PDES with long-range communication links is fully scalable.

New advances in large‐scale distributed simulation and real‐time applications

New advances in large‐scale distributed simulation and real‐time applications

Measuring the Efficiency of Parallel Discrete Event Simulation in Heterogeneous Execution Environments

This paper deals with the experimental testing and verification of the earlier proposed load balancing and coupling factor criteria for the conservative parallel discrete event simulation in heterogeneous execution environments whether they can ensure a good speedup. The definition of the relative speedup is extended to the heterogeneous systems in a natural way. This definition is used to measure the efficiency of the parallel simulation executed by heterogeneous systems. A closed queueing network is used as the simulation model. It is demonstrated on the example of a heterogeneous system containing 134 CPU cores of 6 different types of computers that a good speedup can be achieved using the load balancing and coupling factor criteria. It is shown that the extension of the definition of the relative speedup to the heterogeneous systems made it easy to judge the speedup of parallel discrete event simulation in heterogeneous execution environments.

Parallel discrete event simulation on desktop grid computing infrastructures

In contrast to traditional Parallel Discrete Event Simulation (PDES) systems using dedicated machines, an alternative master/worker approach for high throughput execution on shared computing platforms is proposed. Master/worker systems offer potential advantages over traditional PDES systems such as better support for heterogeneous machines, client-driven load balancing, fault tolerance, and execution on non-dedicated machines. We examine feasibility of executing PDES codes on master/worker systems. We characterise application properties that are important for efficient execution and define suitable performance metrics. The Aurora Parallel and Distributed Simulation System utilising the master/worker approach is described along with performance measurements in executing conservative PDES codes.

Algorithmic scalability in globally constrained conservative parallel discrete event simulations of asynchronous systems.

We consider parallel simulations for asynchronous systems employing L processing elements that are arranged on a ring. Processors communicate only among the nearest neighbors and advance their local simulated time only if it is guaranteed that this does not violate causality. In simulations with no constraints, in the infinite L limit the utilization scales [Korniss et al., Phys. Rev. Lett. 84, 1351 (2000)]; but, the width of the virtual time horizon diverges (i.e., the measurement phase of the algorithm does not scale). In this work, we introduce a moving Delta-window global constraint, which modifies the algorithm so that the measurement phase scales as well. We present results of systematic studies in which the system size (i.e., L and the volume load per processor) as well as the constraint are varied. The Delta constraint eliminates the extreme fluctuations in the virtual time horizon, provides a bound on its width, and controls the average progress rate. The width of the Delta window can serve as a tuning parameter that, for a given volume load per processor, could be adjusted to optimize the utilization, so as to maximize the efficiency. This result may find numerous applications in modeling the evolution of general spatially extended short-range interacting systems with asynchronous dynamics, including dynamic Monte Carlo studies.

Suppressing roughness of virtual times in parallel discrete-event simulations.

In a parallel discrete-event simulation (PDES) scheme, tasks are distributed among processing elements (PEs) whose progress is controlled by a synchronization scheme. For lattice systems with short-range interactions, the progress of the conservative PDES scheme is governed by the Kardar-Parisi-Zhang equation from the theory of nonequilibrium surface growth. Although the simulated (virtual) times of the PEs progress at a nonzero rate, their standard deviation (spread) diverges with the number of PEs, hindering efficient data collection. We show that weak random interactions among the PEs can make this spread nondivergent. The PEs then progress at a nonzero, near-uniform rate without requiring global synchronizations.

Going through Rough Times: from Non-Equilibrium Surface Growth to Algorithmic Scalability

AbstractEffcient and faithful parallel simulation of large asynchronous systems is a challenging computational problem. It requires using the concept of local simulated times and a synchronization scheme. We study the scalability of massively parallel algorithms for discrete-event simulations which employ conservative synchronization to enforce causality. We do this by looking at the simulated time horizon as a complex evolving system, and we identify its universal characteristics. We find that the time horizon for the conservative parallel discrete-event simulation scheme exhibits Kardar-Parisi-Zhang-like kinetic roughening. This implies that the algorithm is asymptotically scalable in the sense that the average progress rate of the simulation approaches a non-zero constant. It also implies, however, that there are diverging memory requirements associated with such schemes.

Going Through Rough Times: from Non-Equilibrium Surface Growth to Algorithmic Scalability

ABSTRACTEfficient and faithful parallel simulation of large asynchronous systems is a challenging computational problem. It requires using the concept of local simulated times and a synchronization scheme. We study the scalability of massively parallel algorithms for discrete-event simulations which employ conservative synchronization to enforce causality. We do this by looking at the simulated time horizon as a complex evolving system, and we identify its universal characteristics. We find that the time horizon for the conservative parallel discrete-event simulation scheme exhibits Kardar-Parisi-Zhang-like kinetic roughening. This implies that the algorithm is asymptotically scalable in the sense that the average progress rate of the simulation approaches a non-zero constant. It also implies, however, that there are diverging memory requirements associated with such schemes.

Conservative Parallel Simulation of a Large Number of Processes

In this paper we evaluate parallel simulation based on conservative parallel discrete-event simulation algorithtms. We focus on optimization techniques for simulation models consisting of a large set of elements or processes. One of the main problems during simulation of these large models is the synchronization of elements located on the same processor. While advanced algo rithms are often very time consuming, simple methods fail to exploit the parallelism within the model. Here we describe methods which exploit the parallelism within a model while adding only moderate overhead to the simulation. Finally, we present some simulation results based on our methods.

Scalability Analysis for Conservative Simulationof Logical Circuits

We investigate conservative parallel discrete event simulations for logical circuits on shared‐memory multiprocessors. For a first estimation of the possible speedup, we extend the critical path analysis technique by partitioning strategies. To incorporate overhead due to the management of data structures, we use a simulation on an ideal parallel machine (PRAM). This simulation can be directly executed on the SB‐PRAM prototype, yielding both an implementation and a basis for data structure optimizations. One of the major tools to achieve these optimizations is the SB‐PRAM′s hardware support for parallel prefix operations. Our reimplementation of the PTHOR program on the SB‐PRAM yields substantially higher speedups than before.

Dag consistent parallel simulation

We present a novel approach to parallel discrete event simulation based on the Cilk model of multithreaded computation. Cilk's runtime system not only manages the low-level aspects of program execution, but also provides the user with an algorithmic model of performance which can be used to predict the execution time of a parallel simulation. Moreover, a Cilk application can ``scale down'' to run on a single processor with nearly the same performance as that of serial code.A conservative parallel discrete event simulation algorithm has been developed in which communication between logical processes is achieved using Cilk's virtual memory model, dag consistent shared memory. The simulation executes in cycles, where each cycle involves a divide and conquer computation. Although local lookahead information can be exploited, the algorithm is robust in that it also calculates a global simulation time for each cycle. It can therefore be used for applications where zero lookahead may occur.

Improving conservative VHDL simulation performance by reduction of feedback

This paper describes two forms of feedback in the simulation runtime of VHDL circuits that greatly influences performance. While circuit feedback and strongly connected components have been observed and documented as detrimental influences to conservative parallel discrete event simulation (PDES) efficiency, that influence has never been quantified. Moreover, in this study, the phenomenon of induced feedback was observed to diminish speedup to the same degree as explicit feedback. In this paper the influence of feedback on simulation runtime is analyzed and an algorithm for its elimination is presented. In addition, a metric for the quantification of feedback is introduced. By measuring feedback, it is possible to balance its influence on simulation runtime with that of other factors (e.g. load balance, number of processors, machine granularity, etc. ) through the use of a cost-based partitioning approach. This paper reports significant improvements in runtime for three circuits due to the prevention of feedback using the partitioning algorithm presented. In addition, strong correlation between the feedback metric and conservative parallel simulation overhead is demonstrated.

Conservative circuit simulation on shared-memory multiprocessors

We investigate conservative parallel discrete event simulations for logical circuits on shared-memory multiprocessors. For a first estimation of the possible speedup, we extend the critical path analysis technique by partitioning strategies. To incorporate overhead due to the management of data structures, we use a simulation on an ideal parallel machine (PRAM). This simulation can be directly executed on the SB-PRAM prototype, yielding both an implementation and a basis for data structure optimizations. One of the major tools to achieve these is the SB-PRAM's hardware support for parallel prefix operations. Our reimplementation of the PTHOR program on the SB-PRAM yields substantially higher speedups than before.

BULK SYNCHRONOUS PARALLEL ALGORITHMS FOR CONSERVATIVE DISCRETE EVENT SIMULATION∗

All the parallel discrete event simulation algorithms developed so far have been designed to suit a specific parallel model (e.g., a PRAM model, a MP-RAM model, etc.). This paper presents several versions of conservative parallel discrete event simulation algorithms developed around a unifying model for general purpose parallel computer design and programming, namely around the Bulk Synchronous Parallel (BSP) model. The new algorithms are analysed in terms of the BSP model parameters and the effectiveness of simulators based on these algorithms is evaluated. The performance achieved even for a loosely coupled distributed system is comparable with that reported in previous research work, while the generality of the BSP model provides portability to the new approaches.

Parallelism in sequential multiprocessor simulation models

The design and analysis of multiprocessor simulation models represents a complex and computationally demanding application that is a candidate for parallel simulation. This paper examines the application of conservative parallel discrete event simulation on a set of existing “real-world” models created over the years with no thought given to the parallel execution. These models are based on a subset of Petri Nets known as Marked graphs . The results of the study of the transparent and automatic parallel execution of the simulation of these models will be presented. Focus will be on the amount of inherent parallelism in such models, means of exploiting it, obstacles prohibiting parallelization, and sensitivity to different system and model parameters.

The cost of conservative synchronization in parallel discrete event simulations

This paper analytically studies the performance of a synchronous conservative parallel discrete-event simulation protocol. The class of models considered simulates activity in a physical domain, and possesses a limited ability to predict future behavior. Using a stochastic model, it is shown that as the volume of simulation activity in the model increases relative to a fixed architecture, the complexity of the average per-event overhead due to synchronization, event list manipulation, lookahead calculations, and processor idle time approaches the complexity of the average per-event overhead of a serial simulation, sometimes rapidly. The method is therefore within a constant factor of optimal. The result holds for the worst case “fully-connected” communication topology, where an event in any other portion of the domain can cause an event in any other protion of the domain. Our analysis demonstrates that on large problems—those for which parallel processing is ideally suited— there is often enough parallel workload so that processors are not usually idle. It also demonstrated the viability of the method empirically, showing how good performance is achieved on large problems using a thirty-two node Intel iPSC/2 distributed memory multiprocessor.