K.M. Chandy and R. Sherman. "The Conditional Event Approach to Distributed Simulation." In Distributed Simulation Conference, Miami, 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....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 Protocol (CEP) [12], and a new protocol, which is a combination of the two [23] As discussed in the next section, MPI Sim exploits the determinism present in the communication pattern of the application to reduce, and in many cases, completely eliminate synchronization overheads. Although simulation protocol ....

K. M. Chandy and R. Sherman. The conditional event approach to distributed simulation. In Proceedings of Distributed Simulation Conference, 1989.

....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 Protocol (CEP) [12], and a new protocol, which is a combination of the two [23] As discussed in the next section, MPI Sim exploits the determinism present in the communication pattern of the application to reduce, and in many cases, completely eliminate synchronization overheads. Although simulation protocol ....

K. M. Chandy and R. Sherman. The conditional event approach to distributed simulation. In Proceedings of Distributed Simulation Conference, 1989.

....phase is usually a barrier synchronization such that all LPs complete the current computation phase before proceeding to the next one. Thus, these algorithms are accurate, nonaggressive, riskfree and synchronous. Examples include lookahead based windows [18] 19] Conditional Events [20] and YAWNS [21] The main limitation of synchronous algorithms is the determination of the size of the logical time window. The optimal size depends on the application and user expertise. Performance can be bad if the window is either too small or large. Furthermore, these algorithms su#er some ....

K. Chandy and R. Sherman, "The conditional event approach to distributed simulation,

....sequential and parallel optimistic execution of Maisie models have been described previously[2] In this paper, we evaluate 1 the performance of a set of conservative algorithms that have been implemented in Maisie. The set of algorithms include the null message algorithm[6] the conditional event[7] algorithm, and a new conservative algorithm called the Accelerated Null Message (ANM) algorithm that combines the preceding two approaches. Unlike previous performance studies which use speedup or throughput as their metric of comparison, we use efficiency as our primary metric. We define the ....

....contains a cycle, the cycle must include at least one LP with positive lookahead (i.e. if the LP accepts a message with timestamp t, any message generated by the LP must have a timestamp that is strictly greater than t) 13] 2. 2 Conditional Event Algorithm In the conditional event algorithm [7], the LPs alternate between an EIT computation phase, and an event processing phase. For the sake of simplicity, we first consider a synchronous version which ensures that no messages are in transit when the LPs reach the EIT computation phase. In such a state, the value of EIT for an LP p is ....

[Article contains additional citation context not shown here]

K.M. Chandy and R. Sherman. The conditional event approach to distributed simulation. In Distributed Simulation Conference, Miami, 1989.

....which is based on time advancement noti cations at per logical process level. CMB is an asynchronous algorithm; null messages function as time advancement noti cations between logical processes. Several synchronous algorithms exist, including Chandy and Sherman s conditional event approach [23], Lubachevsky s bounded lag algorithm [71, 72, 73] and Nicol s YAWNS protocol [86, 87] Synchronous 9 approaches exploit the computational eciency of global parallel operations like barriers and min reductions. They are valued for their simplicity as well as provable scalability. The rst ....

....such a graph. In this section, we propose ways to reduce the synchronization overhead based on the topology of the model. 5. 1 Composite Synchronization The literature on conservative synchronization of parallel discrete event simulations describes synchronous approaches based on barriers (e.g. [23, 72, 87]) and asynchronous approaches which govern a sub model s time advance as a function of the advance of other submodels which may a ect it (e.g. 22, 129] Synchronous approaches exploit the computational eciency of global parallel operations like barriers and min reductions, and are valued for ....

K. M. Chandy and R. Sherman. The conditional event approach to distributed simulation. Proceedings of the SCS Multiconference on Distributed Simulation, pages 93-9, 1989.

....phase is usually a barrier synchronization such that all LP s complete the current computation phase before proceeding to the next one. Thus, these protocols are accurate, non aggressive, without risk and synchronous. Examples include lookahead based windows [Luba89, AyRa92] Conditional Events [ChSh89] and YAWNS [Nico93] 16 The main limitation of synchronous protocols is the determination of the size of the logical time window. The optimal size depends on the application and its identification requires user expertise. Performance can be very bad when the window is either too small or too ....

....attributes is embedded in the simulation Knowledge about the simulated system may be embedded in the simulation or the protocol. The latter approach is more flexible. This knowledge can be used during the simulation to determine the safety of events as in the Conditional Knowledge approach of [ChSh89] or the use of lookahead in non aggressive windowing schemes [Luba89, AyRa92] Use of embedded knowledge prevents a simulation protocol from being completely general. Knowledge dissemination: LP s initiate the transmission of knowledge to other LP s Knowledge dissemination occurs when an LP ....

Chandy, K.M. and Sherman, R., "The conditional event approach to distributed simulation", Proceedings of the SCS Multiconference on Distributed Simulation, 1989, 93-99.

....the parallel implementation, each processor has its own event list and a synchronization algorithm must be used to enforce the global ordering. For parallel simulation, the simulation kernel has the capability of using the traditional null message protocol [Mis86] the conditional event protocol [Cha89], or a combination of both [Jha93] 4.3 MPISIM The MPISIM layer provides simulation of all MPI communication. This includes all the various point to point communication calls as well collective communication. We assume that communication between LP s is FIFO (First In, First Out) The simulation ....

K.M. Chandy and R. Sherman. "The Conditional Event Approach to Distributed Simulation." In Distributed Simulation Conference, Miami, 1989.

....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 Protocol (CEP) [12], and a new protocol, which is a combination of the two [22] As discussed in the next section, MPI Sim exploits the determinism present in the communication pattern of the application to reduce, and in many cases, completely eliminate synchronization overheads. Although simulation protocol ....

K. M. Chandy and R. Sherman, "The Conditional Event Approach to Distributed Simulation," Proceedings of Distributed Simulation Conference, 1989.

.... [Mors90] environment systems [Ebli89] Pres90] Bagr90] Nico90] flexible manufacturing systems [Nevi90] automobile traffic modelling [Merr90] aerodynamic simulation [Norr87] telecommunication networks [Muft90] Robe92] Eick91] Turn92] queueing networks [Reed88a] Fuji89] Fuji89a] [Chan89] [Ayan89] Nico93] Nico93] and Petri nets [Kuma90] Thom91] Nico91] Bacc93] Fers93] The execution of these computationally intensive simulations requires either modelling at higher levels of abstraction in order to reduce the computational load 4 or more powerful machines 5 . The former ....

.... adding transparency to conservative algorithms is continuing [Jha93] With regard to performance of conservative techniques, reported speedups include 16 on 25 processors of a Sequent Balance and 1900 on a 16384 processor Connection Machine [Luba89] Luba89b] 7 on 12 and 9 on 24 iPSC processors [Chan89], 18 on 31 transputers [Merr90] 8 on 64 and 10 to 20 on 128 iPSC processors [Su89] and 16 on a 16 processor BBN Butterfly [Fuji89] 3.8 Optimistic Synchronization Protocols The objective of optimistic approaches is to detect and recover from causality errors rather than strictly avoid them. The ....

Chandy, K. M., Sherman, R., "The Conditional Event Approach to Distributed Simulation", Proceedings of the 1989 SCS Multiconference on Distributed Simulation, SCS Simulation Series, March 1989, pp. 93-99.

....event simulation. For this discussion, we define the LBTS to be the minimum timestamp on any message that can possibly be received in the future. The LBTS algorithm is a fundamental computation used by conservative synchronization protocols, such as YAWNS[5] and the conditional event protocol[2], as well as optimistic protocols when computing a Global Virtual Time (GVT) Our two algorithms assume differing network service models, and we show that the service model assumptions and choices can have a large impact on overall performance. The main contribution of this work is the ....

K. M. Chandy and R. Sherman. The conditional event approach to distributed simulation. In Proceedings of the SCS Multiconference on Distributed Simulation, 1989.

....describes a two pass algorithm for lower bound timestamp computation which is similar in spirit to our method. Mattern s algorithm establishes a consistent cut point in which all messages sent between processors have either been accounted for, or are known to not affect the computation. Chandy s[4] conditional event protocol determines a range of safe events by finding a minimum of all possible events from any peer. Lubachevsky[5] describes a Bounded Lag protocol for determining safe events, which takes into account the minimum simulation time delay between any two simulation objects. In ....

K. M. Chandy and R. Sherman, "The conditional event approach to distributed simulation, " in Proceedings of the SCS Multiconference on Distributed Simulation, 1989.

....Global synchronizations that are implicit in a data parallel program allow for even more comprehensive optimizations. We apply the application level optimization to a novel conservative synchronization protocol that uses variations of existing null message[CM81] and conditional event[CS89] protocols. We demonstrate the signi cant reduction in the frequency and strength of synchronization in the parallel simulator. We present the results of an experimental study in which we simulate a number of applications including programs from the NAS Parallel Benchmark Suite (NPB 2) Using ....

....because the lookahead of an LP is poor and might result in many rounds of null messages being sent before and LP is able to proceed. The conditional event protocol o ers a simpler solution to that problem. 2.2. 3 The Conditional Event Protocol The conditional event protocol (CEP) CS89] identi es the earliest event in a simulation model by using global information. It calculates a global lower bound on the receive timestamp of the next message that will be received by any LP. In order to be able to compute EIT it is sucient to calculate the minimum ECOT of all LPs, and the ....

K.M. Chandy and R. Sherman. The conditional event approach to distributed simulation. In Distributed Simulation Conference, Miami, 1989.

....the difference of their timestamps must be bounded from above by a constant. A problem common 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 will not be cancelled or disabled by other events. Conditional events, on the other hand, will be ....

K. M. Chandy and R. Sherman. The conditional event approach to distributed simulation. Proceedings of the SCS Multiconference on Distributed Simulation, 21(2):93--99, March 28--31, 1989.

....by a series of discrete events which can be specified and programmed in great detail. The term distributed simulation seems to indicate a set of goals which are somewhat in common with those for virtual reality, or more specifically, with WAVES. This is not the case as is discovered in [CS88] and [Nut90] which demonstrate the more important issue of distributed simulation, which is synchronization of events. The authors are mostly concerned with running certain tasks in parallel, while maintaining certain sequential constraints. In fact, load balancing concerns are entirely glossed ....

K. M. Chandy and R. Sherman. The Conditional-Event Approach to Distributed Simulation. University of Southern California - Information Sciences Institute Research Report, ISI/RR-88-226, June 1989.

....because the lookahead of an LP is poor and might result in many rounds of null messages being sent before and LP is able to proceed. The conditional event protocol offers a simpler solution to that problem. 2.2. 3 The Conditional Event Protocol The conditional event protocol (CEP) CS89] identifies the earliest event in a simulation model by using global information. It calculates a global lower bound on the receive timestamp of the next message that will be received by any LP. In order to be able to compute EIT it is sufficient to calculate the minimum ECOT of all LPs, and the ....

K.M. Chandy and R. Sherman. The conditional event approach to distributed simulation. In Distributed Simulation Conference, Miami, 1989.

....distribution for the timestamp increments, negligible checkpointingrollback overhead and negligible communication delays. Modifications to the classical Time Warp synchronization have been proposed to embed barrier synchronizations in order to allow optimism of event execution at a limited extent [1, 4, 7, 25, 31, 32, 35]. Among the outcoming protocols, a complete performance model has been presented in [6] for the case of Bounded Time Warp [35] In this protocol a global window is iteratively defined and, in any iteration, LPs are allowed to process only the messages with timestamp in that window. The analysis in ....

K.Chandy and R.Sherman,"Conditional Event Approach to Distributed Simulation", Proc. 1989 SCS Multiconference on Distributed Simulation, pp.93-99, 1989.

....or non aggressive, with or without risk, and loosely synchronous. Examples of iterative PDES synchronization protocols include the Bounded Lag protocol [LUBA88a] the Moving Time Window protocol [SOBW88] Chandy and Sherman s protocol for converting conditional events into unconditional events [CHSH89], an iterative algorithm based on the distance between objects [AYAN89] Nicol s iterative algorithm [NICO90] NICO91] and the Global Windowing Algorithm [DICK93] 2.1.5. Hardware Support for PDES The need for special purpose hardware to support PDES is well established. In a recent survey on the ....

Chandy, K. M. and Sherman, R., "The Conditional Event Approach to Distributed Simulation", Proceedings of the SCS Multiconference on Distributed Simulation, Tampa, Florida, pp. 93-99, (March 1989).

....algorithms which include a sequential global event list algorithm running on a uni processor machine, as well as a conservative and optimistic parallel algorithm. The conservative algorithm has three different variants including the null message protocol [23] the conditional event protocol [7], and an asynchronous combination of the null message and the conditional event protocols [17] The optimistic algorithm uses the space time protocol [8] with lazy cancelation. Besides these protocols, Maisie also provides an Ideal Simulation Protocol(ISP) 16] runtime which provides a reference ....

K.M. Chandy and R. Sherman. The conditional event approach to distributed simulation. In Distributed Simulation Conference, Miami, 1989.

....processor mappings to reduce communications and other overheads in parallel execution. 5. 1 Synchronization protocols The performance of GloMoSim library is evaluated as a function of three different conservative synchronization algorithms: null message protocol [15] conditional event protocol [6], and Accelerated Null Message protocol (ANP) which is a combination of preceding two schemes [10] The PARSEC visual front end (PAVE) allows choice of a specific conservative runtime to be selected simply as an option in the execution command As the null message protocol is well known, we omit ....

K.M. Chandy and R. Sherman. "The Conditional Event Approach to Distributed Simulation." Distributed Simulation Conference, Miami, 1989.

....object (also called an LP) cannot process a message until the system can guarantee that it will not subsequently receive a message with an earlier timestamp. This constraint may introduce deadlocks, which are typically handled by incorporating deadlock detection[Mis86] or deadlock avoidance[Mis86, Bry77, Cha89a] mechanism into the simulation algorithm. Optimistic algorithms[Jef85, Cha89b] allow an LP to process messages out of order; causality errors are corrected by using rollbacks and re computations. Implementations of optimistic algorithms are usually more difficult because they require complex ....

....algorithms and is designed to reduce the synchronization overheads with both conservative and optimistic parallel algorithms. The simulation algorithms currently supported by Maisie include a sequential algorithm, parallel conservative algorithms based on null messages[Mis86] and conditional events[Cha89a], a new conservative protocol that combines null messages with conditional events[Jha93] and a new protocol called ISP which allows an analyst to estimate the potential for performance improvement from parallel execution of a model using a specific partitioning strategy and parallel architecture. ....

K.M. Chandy and R. Sherman. "The Conditional Event Approach to Distributed Simulation." In Distributed Simulation Conference, Miami, 1989.

No context found.

K.M. Chandy and R. Sherman. "The Conditional Event Approach to Distributed Simulation." In Distributed Simulation Conference, Miami, 1989.

No context found.

K. M. Chandy and R. Sherman. The conditional event approach to distributed simulation. In Proceedings of the 1989 SCS Multiconference on Distributed Simulation, January 1989.

No context found.

K.M. Chandy and R. Sherman. The Conditional Event Approach to Distributed Simulation. In Distributed Simulation Conference, Miami, 1989.

No context found.

Chandy, K.M. and R. Sherman, The Conditional Event Approach to Distributed Simulation, in Proceedings of the SCS Multiconference on Distributed Simulation, B. Unger and R.M. Fujimoto, Editors. 1989, Society for Computer Simulation. p. 93-99.

No context found.

Chandy, K. M. and R. Sherman. 1989a. The conditional event approach to distributed simulation. In Proceedings of the SCS Simulation Multiconference on Distributed Simulation, 93-99, March.

First 50 documents  Next 50

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