O. Berry and D. Jefferson. Critical Path Analysis of Distributed Simulation. In Proceedings of the 1985.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....by the concurrency of the simulation. By viewing a trace obtained from a conventional sequential (event list driven) simulation as a directed graph of sequential constraints, it is possible to compute a measure of the inherent concurrency of a simulation using critical path analysis techniques[24]. The measure we use is detected concurrency. Detected concurrency is equal to the maximum speedup possible under 18 certain ideal assumptions as defined below. In order to compute the detected concurrency of the benchmark simulations, we have instrumented our sequential (event list driven) ....

O. Berry and D. Jefferson, "Critical Path Analysis of Distributed Simulation," Proc. SCS Eastern Multiconf. --- Distributed Simulation, Society for Computer Simulation, Vol. 15, No. 2, pp. 57-60, January 1985.

....on lower bounds. An understanding of these issues is essential to the successful application of critical path analysis to PDES. This work appears in [SrRe95a] 3. 1 Overview Critical path analysis is a technique to derive lower bounds on the completion time of PDES s [Livn85] Berry and Jefferson [BeJe85] applied critical path analysis to PDES and argued that the critical path time is a lower bound on the completion time. This application of critical path analysis is particularly interesting because of the somewhat counterintuitive result that it is possible for certain simulations to complete in ....

Berry, O. and Jefferson, D, "Critical path analysis of distributed simulation", Proc. of the 1985 SCS Conference on Distributed Simulation, January 1985, 57-60.

....the results obtained from occarm with 12 The estimation of the maximum possible speedup of a distributed simulation is currently an active area of research. Techniques which have been suggested for this purpose include the employment of a critical path analysis of a trace from a given simulation [Berr85] [Som89] and the treatment of the process graph as a queueing network model [Wagn89] CHAPTER 8. VALIDATION OF THE OCCARM MODEL 208 those provided by the Asim model; the results have been obtained by executing the Dhrystone benchmark on the two models. The distribution of preemtions over time ....

Berry, O., Jefferson , D., "Critical Path Analysis of Distributed Simulation ", Proceedings of the SCS Distributed Simulation Conference, SCS Simulation Series, 1985, pp. 57-60.

....are either too general or too dedicated to certain problems to be widely used in parallel and distributed simulation. One general method used for performance prediction and analysis of parallel and distributed simulation systems is the critical path analysis proposed by Berry and Jefferson in [Ber85]. It is based on an event trace, which presents the events and their interactions. The critical time values represent the parallel properties of the application and the algorithm. In this paper the critical path analysis is used for predicting the execution time of the conservative simulation and ....

....and the communication network. When the directed graph has been constructed a critical path algorithm can be applied to find out the longest weighted path in the graph. This logical path from the first event to the last event represents the lower bound for the simulation in the given hardware [Ber85]. This lower bound is valid mainly in the conservative methods as certain variants of the optimistic Time Warp method are shown to be supercritical [Gun94, Jef91, Sri95] The creation of a directed graph for several processors is limited by two constraints, i.e. predecessor and creator ....

Berry O. and Jefferson D.: Critical Path Analysis of Distributed Simulation, Proceedings of the 1985 Conference on Distributed Simulation, 1985, pp. 57-60.

....computation and the cancellation of those messages can be quite long. By delaying cancellation (of erroneous messages) lazy cancellation may allow the erroneous computation to spread, causing more rollbacks in other processes, thereby prolonging the execution time of the simulation. Berry [3] has documented an intriguing and beneficial characteristic of lazy cancellation. She has shown that, under some circumstances, Time Warp using lazy cancellation is capable of Supercritical Speedup i.e. it can run faster than the critical path of the equivalent sequential simulation. The ....

....event message at simulation time 40, the simulation finally completes. 29 Chapter 5 Dynamic Cancellation Time Warp has received considerable attention as a PDES mechanism because its optimistic nature is well suited for exploiting the parallelism that is typically available in large simulations [10, 3]. Although the Time Warp mechanism is capable of speeding up large simulations, its potential to do so has not been fully exploited due to excessive rollbacks (resulting in instability and wastage of the lookahead computation done by the LPs) For optimum performance, Time Warp simulations should ....

[Article contains additional citation context not shown here]

Berry, O., and Jefferson, D. Critical path analysis of distributed simulation. In Distributed Simulation (1985), Society for Computer Simulation, pp. 57--60.

....simulation of a fixed experiment on a fixed model. From the above constraints, we can derive a partial order on this set, called causality . The representation of this order as a directed graph G = V; E) is called event precedence graph (EPG) introduced independently by Berry and Jefferson [4] and Livny [18] V is the set of events, e 1 ; e 2 ) is an edge iff. e 1 schedules e 2 or e 1 is the last event before e 2 on the same LP. The weight function : V R 0 assigns to each event the runtime to execute it 2 . We call an event e 2 dependent on e 1 iff. there exists a path in G ....

O. Berry and D. Jefferson. Critical path analysis of distributed simulation. In Proceedings 1985 SCS Multiconference on Distributed Simulation, pages 57--60, January 1985.

....are appropriate for evaluating the usefulness of parallel simulation for a particular simulation application, they tell us little about the efficiency of the parallel simulation protocol used. It is well known that the execution time for most simulation protocols is bound by the critical path time [3] of the simulation model (there exist protocols, called supercritical protocols[9] that are theoretically capable of beating the critical path time. For the rest of the paper, we limit our attention to non supercritical simulation protocols) While evaluating and comparing performances of such ....

O. Berry and D. Jefferson. Critical path analysis of distributed simulation. In Proceedings of 1985 SCS Multiconference on Distributed Simulation, pages 57--60, January 1985.

....for each event, these values are useful to tune up the computation since they give information about what parts of the computation have much synchronization delay. This feature is particularly important if the measure is used in certain applications, such as interactive monitoring or debugging [1, 11]. 4 Computation of measures 4.1 Computation of weight Vector time, independently introduced by Fidge and Mattern [7, 12] may be used to compute the number of events that have actually preceded event e. The system allocates a vector of N counters, V i , to each process. Let V i [j] denote the j ....

....Then, V e i [j] represents the number of events which occur at process P j and causally precede event e. The value of V i for the computation C 1 are given below in Figure 3. The value of V i is represented by ( Delta Delta Delta) For example, V e2;5 2 = 4; 5; 6) that is, V e2;5 2 [1] = 4, V e2;5 2 [2] 5, and V e2;5 2 [3] 6. The value wt (CON(e) the exact number of events that causally affected event e, is computed by: wt(CON(e) P N j=1 V e i [j] Gamma 1. The reason for subtracting 1 is to exclude event e from the count. The value wt (CYL (C) the exact ....

[Article contains additional citation context not shown here]

O. Berry and D. Jefferson. Critical path analysis of distributed simulation. In Proceedings of the Conference on Distributed Simulation, pages 57-- 60, 1985.

....a speed up of n and an efficiency value of 100 percent. In this section, three methods that deal with this issue are presented. Critical path analysis on an ideal multiprocessor. Berry and Jefferson describe a technique to calculate the achievable speed up of an ideal parallel simulation run [7]. The authors propose to measure the physical times needed for each event evaluation (including event queue operations) in a sequential simulator. Using these measurements, the speed up of an ideal parallel simulation compared to a sequential one is calculated. In the following, the term real time ....

Berry, O. and Jefferson, D. (1985) Critical Path Analysis of Distributed Simulation. Proc. of the Distributed Simulation Conference, San Diego, pp. 57 --

....not meaningful. ffl The code skeleton is specific to N MAP, and its portability depends on which platforms N MAP can run. There has been work on performance analyzer algorithms for specific protocols (e.g. 17] for a conservative, asynchronous protocol and [12] for an optimistic protocol. [1] and [9] describe how critical path analysis is used to measure the maximum parallelism inherent in a simulation model. Each publication s focus is mainly on one algorithm s description and proof of its correctness, but little has been reported on the actual implementation issues. There has also ....

....parallel simulation, and (b) deadlock avoidance. Approach (b) introduces null messages to prevent deadlocks from happening. T asyn is based on a performance analyzer algorithm for CMB, null message based protocol. The analyzer algorithm is described in [17] 3) T cp is a critical path analyzer [1, 9] which computes the ideal, maximum speedup which can be achieved by any simulation model, independent of the simulation protocol and assuming infinite number of processors. This gives us some realistic expectations of the possible speedups with more processors. Initialize performance prediction ....

O. Berry and D. Jefferson. Critical path analysis of distributed simulation. In Proceedings of 1985 SCS Multiconference on Distributed Simulation, pages 57--60, January 1985.

....of units, the LPs are kept busy for longer periods of time. 5.2.3 Briner In his Ph.D. thesis [BRI90a] Briner concentrated his work on Time Warp, a parallel asynchronous optimistic simulation scheme, originally described by Jefferson in [JEF85a] He also considers critical path analysis (CPA) [BEJ85a]. This method is a scheme to calculate an upper bound on acceleration which is possible with a distributed simulator and must not be confused with critical path calculations from VLSI design (see, e.g. LDC91a] The CPA uses a sequential simulator and calculates theoretical speed up factors ....

....Round Robin, Kernighan Lin [KEL70a] and SOCCER [RUN88a] A simulated annealing based algorithm will be integrated soon. In addition to the results gained from the usual experiments, the parallelism available from our benchmarks [BBK89a] is also examined using the critical path approach [BEJ85a] and oracle lag [SWF87a] to reflect both the ideal speed up and the effects of real communication. Furthermore, we are implementing several variants of Time Warp to be able to faithfully compare conservative and optimistic approaches. To ease the design of new synchronization strategies, all ....

Berry, O. and Jefferson, D. (1985) Critical Path Analysis of Distributed Simulation, Proc. Distributed Simulation Conference, San Diego, pp. 57 -- 60

....can be achieved with PDES cannot be immediately identified. Perhaps before launching into a long term project, one should first evaluate whether the application offers enough inherent parallelism to justify the effort. A technique called critical path analysis has been developed for this purpose [14, 36] and uses small fragments of code inserted into sequential simulations. The use of this technique not only provides insight as to the inherent parallelism in one application, but also produces data that allows evaluation of a parallel implementation. With the results of critical path analysis, it ....

Orna Berry and David Jefferson. Critical path analysis of distributed simulation. In Proceedings of the SCS Multiconference on Distributed Simulation, pages 57--60, 1985.

No context found.

O. Berry and D. Jefferson. Critical Path Analysis of Distributed Simulation. In Proceedings of the 1985.

No context found.

Berry, O. and Jefferson, D., "Critical path analysis of distributed simulation", Proceedings of the 1985 SCS Conference on Distributed Simulation, January 1985, 57-60.

No context found.

Berry, O. and Jefferson, D., "Critical path analysis of distributed simulation", Proceedings of the 1985 SCS Conference on Distributed Simulation, January 1985, 57-60.

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