## Parallel Discrete Event Simulation: A Survey (1999)

Citations: | 5 - 0 self |

### BibTeX

@TECHREPORT{Vee99paralleldiscrete,

author = {Voon-yee Vee and Wen-jing Hsu},

title = {Parallel Discrete Event Simulation: A Survey},

institution = {},

year = {1999}

}

### OpenURL

### Abstract

In the past decade, parallel processing has gained very significant advances in all fronts of the theory, systems, and applications. However, despite years of research and its apparent significance, parallel simulation remains a major outstanding challenge. In particular, there has been no simulation system which facilitates an early prediction of the program performance. In this report, we document a survey of the major existing approaches for parallel simulation as well as a comparative study of two leading computational models, namely, Valiant's BSP and Leiserson's Cilk, which are useful formal models for performance prediction of simulation programs. 1 Introduction Simulation has been heavily relied upon by computer scientists, physicists, circuit designers, mathematicians, military force, and even video game designers [LK91, Fis95, Chi92]. For decades, simulationists have been devising simulation models for large and complex systems to facilitate performance analysis, stud...

### Citations

8531 | Introduction to Algorithms - Cormen, Leiserson, et al. - 2009 |

1247 | Simulation Modelling and Analysis - Law, Kelton - 2000 |

1130 |
A bridging model for parallel computation
- Valiant
- 1990
(Show Context)
Citation Context ... scalable performance, as long as the parallel program adheres to the Cilk programming paradigm [BJK + 95]. 8.2 The BSP Model The Bulk-Synchronous Parallel (BSP) model was proposed in 1990 by Valiant =-=[Val90] as a-=- bridging model of parallel computation. The notion and the role of a bridging model was aptly captured and introduced by Valiant in [Val90]: ". . . the von Neumann model is the connecting bridge... |

859 | Virtual Time - Jefferson - 1985 |

691 |
Parallel discrete event simulation
- Fujimoto
- 1990
(Show Context)
Citation Context ...ill required to develop efficient simulation programs. There are no "golden rules" that a programmer can follow to guarantee an efficient program. Generally speaking, parallel simulation is =-=very hard [Fuj90a]-=-. Nevertheless, the field of parallel simulation does deserve future research and cannot be simply ignored. Presently, because of its importance, there is much active research to further develop the p... |

644 | The Art of Computer Systems Performance Analysis – Techniques for Experimental Design, Measurement, Simulation, and Modeling - Jain - 1991 |

396 | Scheduling multithreaded computations by work stealing - Blumofe, Leiserson - 1999 |

322 | The implementation of the Cilk-5 multithreaded language - Frigo, Leiserson, et al. - 1998 |

242 | Distributed Discrete-Event Simulation
- Misra
- 1986
(Show Context)
Citation Context ...lt in anomalous behavior and consequently incorrect simulation. It is the responsibility of the synchronization mechanism to ensure proper and correct interactions among the LPs. It has been shown in =-=[Mis86]-=- that a parallel simulation obeys the causality constraint if and only if every LP processes events in nondecreasing timestamp order---a condition formulated as the local causality constraint in [Fuj9... |

165 | Direct bulk-synchronous parallel algorithms
- Gerbessiotis, Valiant
- 1994
(Show Context)
Citation Context ...s well as the latency or the startup cost of routing. For a superstep with a h-relation, i.e., a routing request where each processor sends and receives at most h messages, its cost model is given by =-=[GV94]-=-: maxfL;x + hgg where x is the maximum number of local computations executed by any processor in that superstep. There are also alternative charges, such as maxfx; L+gh s ; L+gh r g [GV94], or maxfL; ... |

136 | Testing random number generators - L’Ecuyer - 1992 |

130 | Efficient algorithms for distributed snapshots and global virtual time approximation - Mattern - 1993 |

127 | The gradient model load balancing method - Lin, Keller |

126 | Simulation of packet communication architecture computer systems - BRYANT - 1977 |

111 | Fast concurrent simulation using the Time Warp mechanism - Jefferson, Sowizral |

109 | Performance of time warp under synthetic workloads - Fujimoto, January |

103 | Parallel and distributed simulation of discrete event systems
- Ferscha, Tripathi
- 1994
(Show Context)
Citation Context ...urring event. It is well-known that the time-driven approach is not a good choice for general-purpose simulation; in particular, the approach is inefficient for events irregularly dispersed over time =-=[Fer95]-=-. It is often difficult to fix a proper 4: to maintain the accuracy of the simulation generally requires a smaller 4; to run the simulation efficiently requires a larger value of 4. The event-driven a... |

103 |
Time Warp on a shared memory multiprocessor
- Fujimoto
- 1989
(Show Context)
Citation Context ...timistic protocols. There are also a number of hybrid approaches combining certain aspects of conservative approaches and optimistic approaches. Some of the important ones are outlined here. Fujimoto =-=[Fuj89]-=- proposes direct cancellation which is optimized for multiprocessors with shared memory architecture. Let \Gamma E be the set of events scheduled when an event E is processed. In this strategy, the si... |

91 | Efficient Distributed Event-Driven Simulations of Multiple-Loop Networks - Lubachevsky - 1989 |

81 |
Breathing time warp
- Steinman
- 1993
(Show Context)
Citation Context ...ets (BTB) approach [Ste93, Ste92], proposed by Steinman, is a window-based approach which does not require anti-messages and therefore does not suffer from the this problem. Breathing Time Warp (BTW) =-=[Ste93]-=- is later proposed also by Steinman aiming to combine the best aspects of Time Warp and BTB while eliminating their shortcomings. Preliminary results show that BTW outperforms BTB and Time Warp for ce... |

78 | Parallel Simulation Today
- Nicol, Fujimoto
- 1995
(Show Context)
Citation Context ...n the time for a conservative approach to execute a simulation. For a discussion of more complex models and a more comprehensive survey on mathematical performance models, the readers are directed to =-=[NF94]-=-, a survey conducted by Nicol and Fujimoto. 6 Other Important Issues Besides simulation strategies, which define the operational principle of a simulation framework, there are other components which c... |

77 | Computer systems performance evaluation - Ferrari - 1978 |

77 |
Rollback mechanisms for optimistic distributed simulation systems
- Gafni
(Show Context)
Citation Context ...s out anti-messages for any previously sent messages with timestamp greater than T . To reduce the anti-message traffic and the frequency of roll backs, a lazy cancellation strategy has been proposed =-=[Gaf88]-=-. In lazy cancellation, instead of immediately sending anti-messages upon receiving a straggler message, an LP examines if the rerun of the simulation generates the same positive messages. An anti-mes... |

77 | General purpose parallel computing - McColl - 1993 |

70 | Executing multithreaded programs efficiently
- Blumofe
- 1995
(Show Context)
Citation Context ... and is asymptotically optimal, since T 1 =P and T1 are both lower bounds [BL94, BJK + 95]. The equation also resembles Brent's theorem [CLR90, p. 709] which yields the upper bound of T psT 1 =P + T1 =-=[Blu95]-=-. It has been verified empirically that the constant factor hidden by the order notation is small, so that T psT 1 =P + T1 is a good approximation for a wide range of applications [Blu95,BJK + 95, FLR... |

63 | Global virtual time algorithms - Bellenot - 1990 |

60 |
Conditional event approach to distributed simulation
- CHANDY, SHERMAN
- 1989
(Show Context)
Citation Context ...n to the window-based approaches is the size of the time window. To tackle this problem, unfortunately, requires information which is application-specific. 4.1.5 Conditional Events Chandy and Sherman =-=[CS89a]-=- suggest a conditional events approach which classifies events into definite events and conditional events. Definite events, which generally have smaller timestamps, will definitely be processed and w... |

60 | Cilk: Efficient Multithreaded Computing - Randall - 1998 |

59 |
Parallel Discrete Event Simulation: Will the Field Survive
- Fujimoto
- 1993
(Show Context)
Citation Context ... to simulating those "intractable" systems. Unfortunately, despite over two decades of research, the technology of parallel simulation has not significantly impressed the general simulation =-=community [Fuj93b]. Consider-=-able efforts and expertise are still required to develop efficient simulation programs. There are no "golden rules" that a programmer can follow to guarantee an efficient program. Generally ... |

59 |
Mtw: A strategy for scheduling discrete simulation events for concurrent execution
- SOKOL, BRISCOE, et al.
- 1988
(Show Context)
Citation Context ...ndows Similar to conservative simulation, window-based approaches have also been proposed for optimistic protocols with the goal of reducing the number of causality errors. Sokol, Briscoe and Wieland =-=[SBW88]-=- have proposed 9 the Moving Time Window (MTW) approach that uses a time window with a fixed size W . The LPs can only process the events within the interval [GV T ; GV T +W ]. Similar to the time wind... |

58 | A Distributed Drafting Algorithm for Load Balancing - Ni, Xu, et al. - 1985 |

58 | A spectrum of options for parallel simulation - Reynolds - 1988 |

56 |
Parallel discrete-event simulation of FCFS stochastic queueing networks
- Nicol
- 1988
(Show Context)
Citation Context ...nformation also helps to reduce the possibility of deadlocks and thus speed up simulation. Many papers related to the derivation and the use of lookahead information have appeared. For example, Nicol =-=[Nic88]-=- proposes a scheme to improve the lookahead capability by precomputing the service times for some future events. Cai and Turner [Cai90, CT90] introduce carrier null messages, which play a similar role... |

54 |
Performance Measurements of Distributed Simulation Strategies", Transactions of the Society for Computer Simulation 6(2
- Fujimoto
- 1989
(Show Context)
Citation Context ...as that of the null messages, but carry additional information on lookahead and the route taken. This approach aims to further improve the quality of lookahead to reduce the message traffic. Fujimoto =-=[Fuj88]-=- demonstrates the importance of lookahead with experimental results, which show that simulators with poor 6 lookahead properties can easily become overburdened by excessive overhead. Cota and Sargent ... |

52 | Models of parallel computation: a survey and synthesis
- Maggs, Matheson, et al.
- 1995
(Show Context)
Citation Context ...enomena is an old art. The purposes of modeling are to capture the salient characteristics of phenomena with clarity, and to provide the right degree of accuracy to facilitate analysis and prediction =-=[MMT95]-=-. In sequential computing, the Random Access Machine (RAM) has been a very successful model of computation that promoted consistency and coordination among algorithm developers, computer architects an... |

49 | Efficient detection of determinacy races in cilk programs
- Feng, Leiserson
- 1997
(Show Context)
Citation Context ...wn, a sync statement, or a return from a Cilk procedure. With the Cilk call/return semantics (spawn and sync) for parallelism, the computation of a Cilk program can be viewed as a series-parallel dag =-=[FL97]-=- (called a Cilk dag henceforth) that unfolds dynamically as the computation progresses. Figure 4 shows an example of the computation of a Cilk program. The Cilk procedure C 1 , shown as the shaded box... |

44 | SPEEDES: A multiplesynchronization environment for parallel discreteevent simulation", pp 251-286 - Steinman - 1992 |

42 | The Cilk system for Parallel Multithreaded Computing - Joerg - 1996 |

40 | Concurrent Operations on Priority Queues - Jones - 1989 |

39 | Design and evaluation of the rollback chip: special purpose hardware for time warp - Fujimoto, Tsai, et al. - 1992 |

36 | Optimal dynamic remapping of data parallel computations - Nicol, Reynolds - 1990 |

34 | An Algorithm for Distributed Discrete-Event Simulation -- the “Carrier Null Message” Approach - Cai, Turner - 1990 |

33 | Optimal memory management for time warp parallel simulation - Lin, Preiss - 1991 |

33 | Limitation of optimism in the Time Warp operating system - Reiher, Jefferson - 1989 |

32 | Distributed simulation of discrete event systems - RIGHTER, WALRAND - 1989 |

31 | and Vijaya Ramachandran, “Parallel algorithms for shared-memory machines - Karp - 1990 |

29 | Rollback sometimes works ... if filtered - Lubachevsky, Shwartz, et al. - 1989 |

25 |
Time warp vs. chandy-misra: A worst-case comparison
- Lipton, Mizell
- 1999
(Show Context)
Citation Context ...d Lazowska prove that, assuming (unrealistically) zero-cost rollbacks and infinite memory space, Time Warp approach outperforms the CMB approach in feedforward network 1 simulation. Lipton and Mizell =-=[LM90]-=- show that there exists a simulation model for which Time Warp outperforms the CMB approach by a factor of p, p being the number of processes---but the reverse is not true; the CMB approach can only o... |

24 | BSP programming - McColl - 1994 |

24 |
Cancellation strategies in optimistic execution systems
- Reiher, Fujimoto, et al.
- 1990
(Show Context)
Citation Context ... also entails the overhead of matching an anti-message with an earlier positive message. A comprehensive comparison of the aggressive cancellation scheme with the lazy cancellation scheme is given in =-=[RFBJ90]-=-. Lazy cancellation does reduce the traffic overhead, but it may also allow erroneous computations to propagate further. This is because an LP requires additional time overhead to check whether same m... |