P. M. Dickens, P. Heidelberger, and D. M. Nicol. Parallelized direct execution simulation of message-passing parallel programs. IEEE Trans. on Parallel and Distributed Systems, 7(10):1090-1105, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... [9, 13, 14] However, even with the use of abstract models and direct execution, sequential program simulators tended to be slow with slowdown factors ranging from 2 to 35 for each process in the simulated program [9] Several recent efforts have been exploring the use of parallel execution [10, 17, 18, 24, 25, 28, 29] to reduce the model execution times, with varying degrees of success. In order to have multiple simulation processes and maintain accuracy, simulations use protocols to synchronize the processes. One of the widely used protocols is the Quantum protocol, which lets the processes compute for a ....

....synchronizing them. In general, synchronous simulators that use the quantum protocol must trade off simulation accuracy with speed (frequent synchronizations slowdown the simulation, but synchronizing less frequently introduces errors, by possibly executing statements out of order) Both LAPSE [17, 18] and Parallel Proteus use some form of program analysis to increase the simulation window beyond a fixed quantum. MPI Sim uses parallel discrete event simulation with the conservative protocol [25, 28] Supported protocols include the Null Message Protocol (NMP) 11] the Conditional Event ....

P. M. Dickens, P. Heidelberger, and D. M. Nicol. Parallelized direct execution simulation of message-passing parallel programs. IEEE Transactions on Parallel and Distributed Systems, 7:1090--1105, 1996.

....Moreover, the model is quite accurate in predicting the relative performance of the three parallelization techniques. 1. INTRODUCTION Predicting performance of a program on a parallel machine has always been a hard problem. A variety of approaches have been taken, including detailed simulations [5], profiling [15] analytical modeling [21] and micro benchmarking [17] Profiling and simu # This research was supported by NSF CAREER award ACI9733520, NSF Grant ACR 9982087, and NSF Grant ACR0130437 lation based approaches can be quite time consuming, but usually lead to more realistic ....

P. M. Dickens, P. Heidelberger, and D. M. Nicol. Parallelized direct execution simulation of message-passing parallel programs. IEEE Trans. on Parallel and Distributed Systems, 7(10):1090--1105, 1996.

.... [9, 13, 14] However, even with the use of abstract models and direct execution, sequential program simulators tended to be slow with slowdown factors ranging from 2 to 35 for each process in the simulated program [9] Several recent efforts have been exploring the use of parallel execution [10, 17, 18, 24, 25, 28, 29] to reduce the model execution times, with varying degrees of success. In order to have multiple simulation processes and maintain accuracy, simulations use protocols to synchronize the processes. One of the widely used protocols is the Quantum protocol, which lets the processes compute for a ....

....synchronizing them. In general, synchronous simulators that use the quantum protocol must trade off simulation accuracy with speed (frequent synchronizations slowdown the simulation, but synchronizing less frequently introduces errors, by possibly executing statements out of order) Both LAPSE [17, 18] and Parallel Proteus use some form of program analysis to increase the simulation window beyond a fixed quantum. MPI Sim uses parallel discrete event simulation with the conservative protocol [25, 28] Supported protocols include the Null Message Protocol (NMP) 11] the Conditional Event ....

P. M. Dickens, P. Heidelberger, and D. M. Nicol. Parallelized direct execution simulation of message-passing parallel programs. IEEE Transactions on Parallel and Distributed Systems, 7:1090--1105, 1996.

....a global synchronization window. It is possible we have to resort to appointment based scheduling. Here appointment means that two logical processes agree to coordinate at a pre speci ed simulation time. There have been several kinds of appointment based protocols in the literature (e.g. 81] and [31]) The eciency of constructing such appointment schedule needs to be closely studied. Another essential aspect that makes the xed schedule possible is to constrain the length of the schedule. As the size of the global synchronization window grows, the length of the schedule could become a ....

P. M. Dickens, P. Heidelberger, and D. M. Nicol. Parallelized direct execution simulation of messagepassing parallel programs. IEEE Transactions on Parallel and Distributed Systems, 7(10):1090-105, 1996.

....accurate EOT. At some point in its execution, let T LC represent the execution time of a code block of an LP, then its EOT becomes T L T LC , and thus enables the blocked LPs to have a better estimate of the EIT. In previous work, researchers have estimated this execution time using pre simulation [15]. In this paper, we show how this can be computed using compiler analysis, and used to improve the efficiency of the resulting model. Example 1: MPI Recv( for (j=1; j N; j ) mdiag[j] mdiag[j 1] ndiag = ndiag mdiag[j] MPI Recv( a, for (k=0; k ndiag; k ) a[k] ....

P. M. Dickens, P. Heidelberger, and D. M. Nicol, "Parallelized direct execution simulation of message-passing parallel programs," IEEE TPDS, vol. 7, pp. 1090-1105, 1996.

....requiring communication or not) potentially affect other target processors, so it is not correct to allow application threads to run ahead until a communication operation is encountered. LAPSE is a conservative, direct execution based parallel simulator of the message passing Intel Paragon [10] [11] [12] It achieves good performance by exploiting two sources of lookahead. First, like the runtime analysis in predictive barrier scheduling, LAPSE lets some application code execute in advance of the simulation of its timing. However, unlike Parallel Proteus, LAPSE does not augment this ....

Philip M. Dickens, Philip Heidelberger, and David M. Nicol. Parallelized direct execution simulation of message-passing parallel programs. Technical Report 94-50, ICASE, June 1994.

.... [9, 13, 14] However, even with the use of abstract models and direct execution, sequential program simulators tended to be slow with slowdown factors ranging from 2 to 35 for each process in the simulated program [9] Several recent efforts have been exploring the use of parallel execution [10, 16, 17, 23, 24, 27, 28] to reduce the model execution times, with varying degrees of success. In order to have multiple simulation processes and maintain accuracy, simulations use protocols to synchronize the processes. One of the widely used protocols is the Quantum protocol, which lets the processes compute for a ....

....before synchronizing them. In general, synchronous simulators that use the quantum protocol must trade off simulation accuracy with speedfrequent synchronizations slowdown the simulation, but synchronizing less frequently introduces errors, by possibly executing statements out of order. Both LAPSE [16, 17] and Parallel Proteus use some form of program analysis to increase the simulation window beyond a fixed quantum. MPI Sim uses parallel discrete event simulation with the conservative protocol [24, 27] Supported protocols include the Null Message Protocol (NMP) 11] the Conditional Event ....

P. M. Dickens, P. Heidelberger, and D. M. Nicol, "Parallelized direct execution simulation of message-passing parallel programs," IEEE Trans. on Parallel and Distributed Systems, vol. 7, pp. 1090-105, 1996.

....between processors is deterministic, and where the execution path of each application process is completely insensitive to timings. This is equivalent to generate and treat the trace online (for instance EPG sim is able to use this mode) Another tool adapts its behavior dynamically : Lapse [7] : LAPSE is a parallel simulator for the Paragon. It assumes that the behavior of applications is insensitive to timings most of the time. It uses windows technics, that work well in this case. If the application becomes sensitive to timings, the windows become so small that this part of the ....

Phillip M. Dickens, Philip Heidelberger, and David M. Nicol. Parallelized direct execution simulation of message-passing parallel programs. Technical Report 94-50, Institute for Computer Application in Science and Engineering, NASA Langley Research Center Hampton, VA 23861-0001, June 1994.

....the send statement is executed) and its predicted receive time. We use a simple contention free model to predict the communication latency of a message. In this model, the latency of a 6 message is a function only of its size. This simple model yields good results for a variety of applications [PHN96] The results presented in this paper also support this assumption. 3. For a receive statement, lp i uses a simulation protocol to remove messages from its inqueue in their simulation timestamp order rather than the order in which messages are physically deposited in its inqueue. When lp i ....

P.M.Dickens, P. Heidelberger, and D.M. Nicol. Parallelized Direct Execution Simulation of Message-Passing Parallel Programs. IEEE Trans. on Parallel and Distributed Systems, 7(10):1090-1105, October 1996.

....when the send statement is executed) and its predicted receive time. We use a simple contention free model to predict the communication latency of a message. In this model, the latency of a message is a function only of its size. This simple model yields good results for a variety of applications [PHN96] The results presented in this paper also support this assumption. 3. For a receive statement, lp i uses a simulation protocol to remove messages from its inqueue in their simulation timestamp order rather than the order in which messages are physically deposited in its inqueue. When lp i ....

P.M.Dickens, P. Heidelberger, and D.M. Nicol. Parallelized Direct Execution Simulation of Message-Passing Parallel Programs. IEEE Trans. on Parallel and Distributed Systems, 7(10):1090--1105, October 1996.

....analysis [14] is an example of how performance measurement intrusion issues can be addressed during post mortem program trace analysis. Sometimes problems of performance information accuracy can be dealt with by deriving performance metrics in vivo of a system, as in direct execution simulation [6,18]. However, one of the most difficult problems involves non deterministic program behavior between two execution environments. The performance extrapolation methodology makes the implicit assumption that the measured performance information in one execution environment is useful to predict the ....

P. Dickens, P. Heidelberger and D. Nicol, Parallelized Direct Execution Simulation of Message Passing Parallel Programs, Technical Report ICASE/94/94-50, Institute for Computer Applications in Science and Engineering, NASA Langley, 1994.

....provides detailed information. Simulation is an important tool for designing all aspects of parallel computer systems. It has a broad spectrum of applications ranging from virtual prototyping of new parallel computers as in Wisconsin s Wind Tunnel [7] to studying parallel applications as in LAPSE [8]. Simulating multiprocessors on a uniprocessor can be done in several ways as Figure 1 shows. The simplest (and the slowest) way is to cycle through all the processors simulating one instruction at a time from each processor. This gives a great deal of accuracy but is very slow. The simulator ASIM ....

....CM 5. Their implementation is based on several kernel modifications and other researchers cannot use their simulator easily. ffl They choose to perform the simulation on a massively parallel machine (the CM 5) which has very high cost and more controlled (i.e less interactive) access. LAPSE [8] is much more general but has the restrictive goal of performing scalability and performance analysis only on Intel Paragon applications. It has a similar disadvantage of choosing a massively parallel machine (the Intel Paragon) to perform the simulation. But the results obtained from both ....

Phillip Dickens, Philip Heidelberger, and David Nicol. Parallelized direct execution simulation of messagepassing parallel programs. Technical Report ICASE/94/94-50, Institute for computer applications in science and engineering, NASA Langley, 1994. Available on WWW from ftp://ftp.icase.edu/pub/techreports/94/9450. ps.Z.

....simulators (Brewer et al. 1991, Davis et al. 1991, Covington et al. 1991) use direct execution to simulate the sequential blocks of code, and simulate only the communication and or I O events. As sequential execution of such models (Legedza and Weihl 1996, Reinhardt et al. 1993, Dickens et al. 1994, Dickens et al. 1996) are typically slow (slowdown factors of 2 to 15 per processor are not atypical) several researchers have used parallel execution of such models with varying degrees of success. The primary difficulty in obtaining better performance is the significant synchronization overhead in the parallel ....

Dickens, P. M., P. Heidelberger, and D.M. Nicol. Parallelized Direct Execution Simulation of MessagePassing Parallel Programs. IEEE Transactions on Parallel and Distributed Systems, 6(4):297-320, October 1996.

....other than simulation in the classical sense. For example, simulated executions of SIMD programs in asynchronous environments can accelerate their execution [58] and parallel simulations executing parallel programs with message passing communication have already been shown to be possible [19]. Other work has shown that an intrusion free monitoring and trace collection of distributed memory parallel program executions is possible by superimposing the execution with a distributed DES protocol [61] The difficult problem of debugging parallel programs finds a high chance to be tackled by ....

Ph. M. Dickens, Ph. Heidelberger, and D. M. Nicol. Parallelized Direct Execution Simulation of Message-Passing Parallel Programs. Technical report, ICASE, NASA Langley Research Center, Hampton, VA, 1994.

No context found.

P.M. Dickens, P. Heidelberger, and D.M. Nicol. Parallelized direct execution simulation of message-passing programs. Technical Report 94-50, ICASE, July 1994.

....a channel has lookahead x if a message sent over that channel at time s never affects the recipient before time s x. Parallel discrete event simulation has proven successes in several application areas, most notably in aviation control[23] Markov chain simulation [15] architectural simulation [20, 4, 5] and telecommunications [3] Nevertheless, every success involves some tuning of synchronization protocol to the model. This is one of several reasons why parallel discrete event simulation is viewed by many as a domain for experts only. This research is supported in part by DARPA Contract ....

....does not advance beyond that appointment time before its appointment partner reaches that point) A simulation s ability to construct an appointment schedule depends very much on the model being simulated. In [15] the appointments are derived from mathematical properties of Markov chains and in [5] they are derived from observation of directly executed computer application code. Another compositional approach is motivated by parallel machines comprised of networked clusters of shared memory multiprocessors (SMP) 9] The same sort of asynchronous algorithm as we ve just described 17 0.001 ....

P. Dickens, P. Heidelberger, and D. Nicol. Parallelized direct execution simulation of message passing programs. IEEE Trans. on Parallel and Distributed Systems, 7(10):1090--1105, October 1996.

....service. The problem is that such clocks do not account for the time a process is swapped out due to the effects of multiprogramming. The issue then is how to factor out the length of time an applet is swapped out due to the effects of multiprogramming. The interested reader is directed to [Dick96] for a detailed discussion of this problem. One solution to this problem would be to develop a model of the effects of multiprogramming, and to use this model to determine the amount of time an applet would have executed had it not been for the effects of multiprogramming. This certainly ....

P. Dickens, P. Heidelberger and D. Nicol. Parallelized Direct Execution Simulation of MessagePassing Parallel Programs. In IEEE Transactions on Parallel and Distributed Systems. Zvolume 7, Number 10, October 1996.

....how simulation time advances as a function of the application process execution and simulated virtual machine behavior. LAPSE has been implemented on the Intel Paragon, and provides excellent performance predictions for a wide range of application codes executing on top of a virtual Intel Paragon [4]. Also, we have ported LAPSE to a network of workstations ( 5] 6] to extend to this platform the performance prediction techniques developed for massively parallel machines. It turns out that modeling the performance of application codes executing on top of a network of workstations poses at ....

Dickens, P., Heidelberger, P. and D. Nicol. Parallelized Direct Execution Simulation of Message-Passing Parallel Programs. In IEEE Transactions on Parallel and Distributed Systems, Volume 7, Number 10, October 1996, pages 1090 - 1105.

....own sense of time, not the time on the Paragon being simulated. Thus nx lib cannot be used to provide accurate timing estimates of how long the code would have taken had it been run on the Paragon. We have ported the distributed memory LAPSE (Large Application Parallel Simulation Environment, Dickens et. al 1994) to the nx lib environment, creating a tool we call here LAPSE nx lib. The combined system augments nx lib s functionality with more accurate temporal behavior and information. This paper briefly describes LAPSE (which has been reported upon in more detail elsewhere (Dickens et al. 1994) and ....

....Dickens et. al 1994) to the nx lib environment, creating a tool we call here LAPSE nx lib. The combined system augments nx lib s functionality with more accurate temporal behavior and information. This paper briefly describes LAPSE (which has been reported upon in more detail elsewhere (Dickens et al. 1994) ) and our early experience with LAPSE nxlib on a small network of Sun Sparc 10 s. We primarily examine performance issues related to distribution of application and simulation processes, paying special attention to slowdowns and speedups. The performance obtained so far is very promising. 2 ....

[Article contains additional citation context not shown here]

Dickens, P., P. Heidelberger, and D. Nicol 1994. Parallelized direct execution simulation of messagepassing programs. Technical Report 94-50, ICASE, July 1994.

....advantage of message passing codes structure. This paper sketches the tool, discusses the addition of a new network simulator within the LAPSE framework, and then considers issues related to using the tool well. 2 LAPSE Usage LAPSE (with only the no delay network) is described more fully in [5]. Here we sketch its structure. To use LAPSE a programmer simply modifies the code s make file to call LAPSE scripts instead of compilers, and sets up a file specifying LAPSE parameters such as the number of simulated and the number of actual nodes. The LAPSE system transforms copies of the ....

P.M. Dickens, P. Heidelberger, and D.M. Nicol. Parallelized direct execution simulation of messagepassing programs. Technical Report 94-50, ICASE, July 1994.

No context found.

P.M. Dickens, P. Heidelberger, and D.M. Nicol. Parallelized direct execution simulation of message-passing programs. Technical Report 9450, ICASE, July 1994.

No context found.

P. M. Dickens, P. Heidelberger, and D. M. Nicol. Parallelized direct execution simulation of message-passing parallel programs. IEEE Trans. on Parallel and Distributed Systems, 7(10):1090-1105, 1996.

No context found.

Philip M. Dickens, Philip Heidelberger, and David M. Nicol. Parallelized direct execution simulation of message-passing parallel programs. Technical Report 94-50, ICASE, June 1994.

No context found.

P. Dickens, P. Heidelberger, and D. Nicol. "Parallelized Direct Execution Simulation of Message-Passing Parallel Programs, " IEEE Transactions on Parallel and Distributed Systems, vol.7, (no.10), Oct. 1996.

No context found.

P. M. Dickens, P. Heidelberger, and D. M. Nicol, "Parallelized direct execution simulation of message-passing parallel programs," IEEE Transactions on Parallel and Distributed Systems 7(10), pp. 1090--1105, 1996.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC