This paper addresses the issue of efficient and accurate performance prediction of large-scale message-passing applications on high performance architectures using simulation. Such simulators are often based on parall...
详细信息
ISBN:
(纸本)076951104X;0769511058
This paper addresses the issue of efficient and accurate performance prediction of large-scale message-passing applications on high performance architectures using simulation. Such simulators are often based on parallel discrete event simulation, typically using the conservative protocol to synchronize the simulation threads. The paper considers how a compiler can be used to automatically extract information about the lookahead present in the application, and how this can be used to improve the performance of the null protocol used for synchronization. These techniques are implemented in the MPI-Sim simulator and dHPF compiler, which had previously been extended to work together for optimizing the simulation of local computational components of an application. The results show that the availability of lookahead information improves the runtime of the simulator by factors of magnitude, with 30-60% improvements being typical for the real-world codes. The experiments also show that these improvements are directly correlated with reductions in the number of null messages required by the simulations.
Real-time distributedsimulations, such as on-line gaming or military training simulations are normally considered to be non-deterministic. Analysis of these simulations is therefore difficult depending solely on logg...
详细信息
ISBN:
(纸本)0769506771
Real-time distributedsimulations, such as on-line gaming or military training simulations are normally considered to be non-deterministic. Analysis of these simulations is therefore difficult depending solely on logging and runtime observations. This paper explores art approach for removing one major source of nondeterminism in these simulations, thereby allowing repeatable executions. Specifically, we use a synchronization protocol to ensure repeatable delivery of messages. Through limited instrumentation of the simulation code, we maintain a virtual time clock, by which message delivery! is governed. The additional overhead imposed by the scheme is shown to be reasonable, although additional reductions in this overhead are anticipated The results are demonstrated in the context of a simple combat model whose only source of non-determinism is communications latency. The simulation is shown to be tirade repeatable, and the perturbation on the execution compared to the nonrepeatable execution small. The paper is one step in bridging the gap between the traditional PDES perspective and real-time simulation world.
If a model shall be executed in a parallel, distributed instead of a sequential manner typically the entire simulation engine has to be exchanged. To adapt the simulation layer more easily to the requirements of a con...
详细信息
ISBN:
(纸本)0769521118
If a model shall be executed in a parallel, distributed instead of a sequential manner typically the entire simulation engine has to be exchanged. To adapt the simulation layer more easily to the requirements of a concrete model to be run in a specific environment a component based simulation layer has been developed for JAMES. A set of different simulator components demonstrates that a component-based design facilitates the exchange of simulators and their combination.
With fixed lookahead information in a simulation model, the overhead of asynchronous conservative parallelsimulation lies in the mechanism used for propagating time updates in order for logical processes to safety ad...
详细信息
ISBN:
(纸本)076951104X;0769511058
With fixed lookahead information in a simulation model, the overhead of asynchronous conservative parallelsimulation lies in the mechanism used for propagating time updates in order for logical processes to safety advance their local simulation clocks. Studies have shown that a good scheduling algorithm should preferentially schedule processes containing events on the critical path. This paper introduces a lock-free algorithm for scheduling logical processes in conservative parallel discrete-event simulation on shared-memory multiprocessor machines. The algorithm uses fetch&add operations that help avoid inefficiencies associated with using locks. The lock-free algorithm is robust. Experiments show that, compared with the scheduling algorithm using locks, the lock-free algorithm exhibits better performance when the number of logical processes assigned to each processor is small or when the workload becomes significant. In models with large number of logical processes, our algorithm shows only modest increase in execution time due to the overhead in the algorithm for extra bookkeeping.
This paper illustrates the effects of the Hyper-Threading processor technology on the runtime performance of a parallel and distributedsimulation middleware. A preliminary analysis of the middleware design and execut...
详细信息
ISBN:
(纸本)0769526977
This paper illustrates the effects of the Hyper-Threading processor technology on the runtime performance of a parallel and distributedsimulation middleware. A preliminary analysis of the middleware design and execution parameters is given to identify the tuning parameters and to evaluate the scalabilility of parallelsimulation. A real testbed scenario has been illustrated, based on the ARTIS parallel and distributedsimulation middleware. The experimental analysis has provided some interesting guidelines about the way to adapt the parallel and distributedsimulation middleware to Hyper-Threading and to increase the execution speed of the simulation.
This paper visits issues that recur in consideration of simulation time-stamps, in the context of building very large simulation models from components developed by different groups, at different times. A key problem ...
详细信息
ISBN:
(纸本)0769506771
This paper visits issues that recur in consideration of simulation time-stamps, in the context of building very large simulation models from components developed by different groups, at different times. A key problem here is "safety", loosely defined to mean that unintended model behavior does not occur due to unpredictable behavior of timestamp generation and comparisons. We revisit the problems of timestamp format and simultaneity, and then turn to the new problem of timestamp inter-operability. We describe how a C++ simulation kernel can support the concurrent evaluation of submodels that internally use heterogeneous timestamps, and evaluate the execution time costs of doing so. We find that use of a safe timestamp format that explicitly allows different timescales costs less than 10 over a stock 64-bit integer format, whereas support for completely heterogeneous timestamps can cost as much as 50% in execution speed.
This paper presents the tolerant, hybrid synchronization schema and its benefits for the parallel and distributedsimulation of interconnected computer networks. The hybrid schema combines conservative and optimistic ...
详细信息
ISBN:
(纸本)0818679654
This paper presents the tolerant, hybrid synchronization schema and its benefits for the parallel and distributedsimulation of interconnected computer networks. The hybrid schema combines conservative and optimistic synchronization approaches by using lookahead for scheduling special events and using the flexibility of Time Warp in certain cases. In addition to these classical approaches the introduction of the ''tolerance'' allows the distributed modules to simulate further ahead than guaranteed by the conservative synchronization schema. This results in significantly smaller simulation runtimes and many other benefits.
In a large-scale distributedsimulation with thousands of dynamic objects, efficient communication of data among these objects is an important issue. The broadcasting mechanism specified by the distributed Interactive...
详细信息
ISBN:
(纸本)0769506771
In a large-scale distributedsimulation with thousands of dynamic objects, efficient communication of data among these objects is an important issue. The broadcasting mechanism specified by the distributed Interactive simulation (DIS) standards is not suitable for large scale described simulation. In the High Level Architecture (HLA) paradigm, the Runtime Infrastructure (RTI) provides a set of services, sich as data distribution management (DDM) among federates. The goal of the DDM module in RTI is to make the data communication more efficient by sending the data only to those federates that need the data, as opposed to the broadcasting mechanism employed by DIS. Several DDM schemes have appeared in the literature. In this paper we discuss grid-based DDM and develop a DDM model that uses grids for matching the publishing/subscription regions, and for data filtering. We show that appropriate choice of the grid-cell size is crucial in obtaining good performance. We develop an analytical model and derive a formula for identifying the optimal cell size in grid-based DDM.
Recently a new class of synchronization algorithms for parallel discrete event simulation has been proposed, namely the near perfect state information algorithms, which are based on a notion of error potential to cont...
详细信息
ISBN:
(纸本)076951104X;0769511058
Recently a new class of synchronization algorithms for parallel discrete event simulation has been proposed, namely the near perfect state information algorithms, which are based on a notion of error potential to control the optimism of event execution. An algorithms of this class, called elastic time algorithm (ETA), has been instantiated. In this algorithm, the error potentials computed using temporal information (next event timestamp, simulation clocks etc.) and is then translated into event execution delay based on a constant factor. In this paper we present a scaled version of ETA (SEAT), in which the error potential is translated into event execution delay based on both a constant factor and an additional scaling factor determined dynamically as a function of the event granularity. We have implemented versions of ETA and SETA for a cluster of PCs connected by a Myrinet switch and we have established in an empirical study that SETA outperforms ETA if there is difference in the granularity of different event types.
parallelsimulation and distributedsimulation sometimes appear to be two different worlds. Where parallelsimulation aims at increasing the speed of a single model by distributing it over more processors, distributed...
详细信息
ISBN:
(纸本)0769521118
parallelsimulation and distributedsimulation sometimes appear to be two different worlds. Where parallelsimulation aims at increasing the speed of a single model by distributing it over more processors, distributedsimulation looks at ways to link entire models or federates that run on different computers. In the case of distributedsimulation, the models themselves are hard to distribute, and often they each run on one processor as a monolithic model. This paper advocates to build the more traditional simulation models in such a way, that they can be easily distributed. As simulationists, we can learn from component-based theory from the software engineering field to prepare our models for distribution, and parts of our models for reuse. The results of several projects show that componentizing simulation models can have many advantages. The results also show that it is not easy to create models in a componentized way, and that current methods for simulation model building should be adapted.
暂无评论