A. Gafni. Rollback mechanism for optimistic distributed simulation systems. In Proceedings of the SCS Multiconference on Distributed Simulation, pages 61--67, 1988.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....a Local Virtual Time (LVT) clock. A causality error arises if an LP receives a message with a time stamp earlier than its LVT value (a straggler message) Canceling events in the input queue of other LPs is performed by cancellation strategies such as Aggressive, Lazy, or Dynamic Cancellation [8, 15, 17]. However, these cancellation strategies do nothing to prevent cascading rollbacks. Several strategies have been suggested to stop the incorrect computations [3, 12, 20] Deelman et al. propose a Breadth First rollback mechanism to stop the propagation of erroneous computations in adjacent e1 e2 ....

....useful computation performed is inversely proportional to the number of events rolled back and hence frequent rollbacks reduce efficiency and degrade performance. Optimizations in Time Warp concentrate on reducing the number of rollbacks or reducing the effect of rollbacks to improve performance [8, 17, 15]. In Time Warp, the anti messages sent to other LPs could result in new anti messages to the LP that sent the initial anti messages. Such cascaded rollbacks degrade efficiency and run time performance of the simulation. In addition to cascading rollbacks, a positive straggler event in a causally ....

A. Gafni. Rollback mechanisms for optimistic distributed simulation systems. In Distributed Simulation, pages 46-- 53. Society for Computer Simulation, Jan. 1988.

....and therefore an anti message is unnecessary. The only difference between the absolute and dynamic impact times is that the latter must be reset as each straggler is processed. As we discuss later, often the value set after one straggler is equal to the absolute impact time. Lazy cancellation [7] is a well known technique of avoiding sending unnecessary anti messages when the previously delivered messages are not affected by a straggler. This technique first compares the new messages generated by the reprocessing of a rolled back event and the old messages generated in the first execution ....

A. Gafni. Rollback mechanisms for optimistic distributed simulation. In Proceedings of the SCS Multiconference on Distributed Simulation, pages 61--67, 1988.

....anti message. The only difference between the absolute impact time and dynamic impact time is that the dynamic impact time must be reset as each straggler is processed. As we discussed earlier, often the reset value after one straggler is equal to absolute impact time. Lazy cancellation [30] is the technique of avoiding unnecessary anti messages if the previously delivered messages have not been affected by a straggler, by com paring the new messages generated by the reprocessing of the rolled back event and 92 the old messages generated during the first execution of the same event. ....

A. Gafni. Rollback mechanisms for optimistic distributed simulation. In Proceedings of the SCS Multiconference on Distributed Simulation, pages 61-67, 1988.

....of how much lookback is available, the notion of impact time has another practical use. In optimistic simulation, if a straggler does not affect the message produced by an event later than the straggler, it is always beneficial not to deliver a corresponding antimessage. Lazy cancellation [9] is one of the well known techniques that can achieve this goal. Upon receiving a straggler, it does not immediately cancel out messages affected by the straggler. Instead it rolls back affected events, processes the straggler, and re executes affected events. In the process, it compares the old ....

Gafni, A. Rollback Mechanisms for Optimistic Distributed Simulation, in Proc. of the SCS Multiconference on Distributed Simulation. 1988.

....performance of a DDES running on a multiprocessor. The traces in Figure 4 represent the behavior of the optimistic protocol with the lazy cancellation message annihilation strategy (i.e. postponing annihilation until it turns out that messages are not reproduced in the repeated simulation [11]) There is a rollback indicator column RB instead of a blocking column; an entry in FC without denotes a firing of a transition that could be rolled back eventually, whereas one with denotes a confirmed firing: We see that in contrast to SE LP 1 can start immediately simulating internal ....

A. Gafni. Rollback mechanisms for optimistic distributed simulation systems. In Proc. of the SCS Multiconference on Distributed Simulation 19, pages 61--67, 1988.

....a rollback chain that eventually terminates. Reducing the size of the rollback chain is attempted by filtering messages with preempted effects, by postponing erroneous message annihilation until it turns out that they are not reproduced in the repeated simulation (lazy cancellation, lc) [5], or by maintaining a causality record for each event to support direct cancellation of messages that will definitely not be reproduced. The performance of rollback mechanisms is investigated in [6] and [7] 1.3 Paper Organization The balance of the rest of this paper is the following. Section ....

A. Gafni. Rollback mechanisms for optimistic distributed simulation systems. In Proc. Conference on Distributed Simulation 1988, pages 61--67, California, 1988. Society for Computer Simulation.

....messages) are considered for returning. An indirect effect of the sendback could also be storage release in remote LPs due to annihilation of messages triggered by the original sender s rollback procedure. Gafni s Protocol In a message traffic study of aggressive and lazy cancellation, Gafni [37] notes that past (RT(m) GVT) and present messages (ST(m) GVT RT(m) and events accumulate in IQ, OQ, SS for the two annihilation mechanisms at the same rate, pointing out also the interweaving of messages and events in memory consumption. Past messages and events can be fossil collected as ....

A. Gafni. Rollback Mechanisms for Optimistic Distributed Simulation Systems. In B. Unger and D. Jefferson, editors, Proceedings of the SCS Multiconference on Distributed Simulation, 19 (3), pages 61--67. SCS, February 1988.

....does not send an antimessage (m ) for m immediately upon receipt of a straggler. Instead, it delays its propagation until the resimulation after rollback has progressed to LVT = ts(m ) producing 6= m . If the resimulation produced m = m , no antimessage has to be sent at all [36]. Lazy cancellation thus avoids unnecessary cancelling of correct messages, but has the liability of additional memory and bookkeeping overhead (potential antimessages must be maintained in a rollback queue) and delaying the annihilation of actually wrong simulations. Lazy cancellation can also ....

A. Gafni. Rollback Mechanisms for Optimistic Distributed Simulation Systems. In Proc. of the SCS Multiconference on Distributed Simulation 19, pages 61--67, 1988.

....rollback chain that eventually terminates. Reducing the size of the rollback chain is attempted by filtering messages with preempted effects [Prak 91] by postponing erroneous message annihilation until it turns out that they are not reproduced in the repeated simulation (lazy cancellation, lc) Gafn 88] or by maintaining a causality record for each event to support direct cancellation of messages that will definitely not be reproduced [Kona 91] The performance of rollback mechanisms is investigated in [Lin 91] and [Luba 91] 9 Comparison of DDES is usually based on critical path analysis [Lin ....

A. Gafni. "Rollback Mechanisms for Optimistic Distributed Simulation Systems ". In: Proc. of the SCS Multiconference on Distributed Simulation 19, pp. 61--67, 1988.

....an e#cient memory management protocol which guarantees that the memory consumption of parallel simulation is of the same order as sequential simulation. Such an algorithm is referred to as an optimal memory management algorithm; a formal definition will be given in Definition 1. Previous work [6, 7, 8] has been devoted to reducing the space complexity of Time Warp simulation. The first optimal memory management protocol (called cancelback) was proposed by Je#erson [8] Although cancelback is considered as a complete solution for the storage management problem in Time Warp, some e#ciency issues ....

....3 is that Time Warp always contains a sequential snapshot in order to support rollback. Thus, more storage must be used in Time Warp. Lemma 3 holds for Time Warp with memory management protocols such as the optimal checkpoint interval approach [11, 25] fossil collection, cancelback protocols [6, 8], or the artificial rollback protocol to be described later. The following lemma holds for Time Warp if fossil collection is the only means for memory management (the proof is omitted) Lemma 4 For all positive integer functions f(x) x, there exists a Time Warp simulation with fossil ....

[Article contains additional citation context not shown here]

Gafni, A. Rollback Mechanisms for Optimistic Distributed Simulation. Proc. 1988.

....which the error occurred. Then any incorrect simulation, including communication with other LPs, is cancelled and resimulated. The first optimistic PDES algorithms, including Time Warp, worked this way. Later, optimizations to the rollback procedure such as Lazy Cancellation and Lazy Rollback [70] were introduced. Using these techniques allows the simulation to check whether part or all of the previous simulation must be rolled back and resimulated from the point in logical time at which the causality error occurred. In [1] Reynolds investigates the variables for the design of a PDES ....

A. Gafni, "Rollback mechanisms for optimistic distributed simulation systems," in Proceedings of the SCS Multiconference on Distributed Simulation, 1988, pp. 61--67.

....acknowledgement within a system employing it, yet the communications interface visible to a Time Warp process is assumed to be unacknowledged. The issue of message acknowledgement in GVT algorithms is a dicult one, and discussion can be found in Section 2.4. Gafni has presented a generalisation [16] of Je erson s ow control that gives a more complete solution to managing future objects. In Gafni s scheme any future object can be reclaimed. If a process has exhausted its available memory then it selects and reclaims the future object with the highest times22 tamp. If this object is an input ....

Gafni, A. Rollback mechanisms for optimistic distributed simulation systems. In Proceedings of the SCS Multiconference on Distributed Simulation (July 1988), vol. 19(3), pp. 61-67.

....history item was erroneously reclaimed. In Time Warp, a lazy cancellation strategy has been proposed to undo events during causality errors only when the computation results during such errors are different from the computations that preserve causality. This results in few number of antimessages [24]. This cancellation strategy can also give super critical speedup. We have seen that there are many Time Warp specific optimizations that aim at reducing the number of messages generated, thus reducing the amount of time spent in communication. The optimizations that have been proposed are ....

Gafni, A. Rollback mechanisms for optimistic distributed simulation systems. In Distributed Simulation (January 1988), Society for Computer Simulation, pp. 61-- 67.

....was just received) to its original sender. This sender must then rollback 3 to the state it was in when it sent the message, and resend the message as it executes forward again. The natural choice [14] of the message to be sent back is the one with the largest send time. Gafni s protocol [11] generalizes message sendback by removing a stored object (input message, state vector, or output message) from a process P that runs out of memory. If the discarded object is an input message, it is returned to the sender, as in message sendback. If it is an output message, it is transmitted to ....

A. Gafni, "Rollback Mechanisms for Optimistic Distributed Simulation Systems," Proc. SCS Multiconf. Distributed Simulation, vol. 19, no. 3, pp. 61-67, July 1988.

....must still be processed as usual. However, memoryless processes (those without state) do not require a state queue and can be managed such that, upon receipt of a straggler message, minimal re computation is necessary. These memoryless processes allow for the advantages of rollback reevaluation [22, 49] without its cost. The notion of rollback relaxation arises with the classi cation of logical processes into two categories, namely: state full processes and memoryless processes. More precisely, a set of processes in a simulation system can be partitioned into two non intersecting subsets: ....

A. Gafni. Rollback mechanisms for optimistic distributed simulation systems. In Distributed Simulation, pages 46-53. Society for Computer Simulation, January 1988.

....An exception is the list of saved states. Here, at least one state older than or at GVT must be saved to allow the LP to restore its state properly, in case of a rollback to GVT. 165 Cancellation strategies Several approaches have been suggested to perform cancellation of unintended messages [Gafn88, RFBJ90]. In aggressive cancellation, whenever an LP rolls back, it sends out antimessages for all erroneous positive messages before restarting forward processing (this is the scheme described above) In lazy cancellation, as the name suggests, the LP does not send an antimessage until it can determine ....

Gafni, A., "Rollback mechanisms for optimistic distributed simulation systems", Proceedings of the SCS Multiconference on Distributed Simulations, Vol. 19, No. 3, July 1988, 61-67.

....a returned message is received, such an overhead is less costly as the side effect of the potential straggler is undone earlier. Gafni s protocols is an enhancement of message sendback where an anti message can also be sent to annihilate its positive copy whenever memory storage is insufficient [14]. The optimism in a fast LP can therefore be constrained. A cancelback scheme, also proposed by Jefferson, allows an LP not necessarily running out of memory to reclaim spaces used for storing positive and negative messages and system states [19] In this protocol the selection of LP for memory ....

A. Gafni, "Rollback Mechanisms for Optimistic Distributed Simulation Systems", Proc. of the SCS Multiconference on Distributed Simulation, Vol. 19, No. 3, pp. 61-67, 1988.

.... however, if rollbacks are necessary, aggressive cancellation will enable them to occur sooner, thus preventing erroneous computation from spreading further into the model; it has been shown that in most cases lazy cancellation tends to perform better that its aggressive counterpart [Gafn85] [Gafn88] [Lomo88] Righ89] 3.8.1.1 Global Virtual Time In order to have the ability to cancel past actions, each process maintains a record of its past history which is periodically updated; typically, this record includes past state vectors, processed input events, and previously sent output messages. ....

....attempt to manage memory so that the simulation does not run out of space; a number of techniques have also been developed CHAPTER 3. MODELLING AND SIMULATION 75 to address the problem of freeing memory when this situation does indeed occur. These techniques include message sendback [Jeff85a] [Gafn88], cancellback [Jeff90] and artificial rollback [Lin92] the basic idea behind these mechanisms is to rollback overly optimistic computations and reclaim the memory they use. 3.8.1.3 Characteristics of Optimistic Protocols Optimistic synchronization approaches can accommodate dynamic creation of ....

Gafni, A., "Rollback Mechanisms for Optimistic Distributed Simulation Systems", Proceedings of the 1988 SCS Multiconference on Distributed Simulation, SCS Simulation Series, July 1988, pp. 61-67.

.... To support rollback, data structures such as the state queue and the output queue are required (to be elaborated) Several strategies for cancelling incorrect computation were surveyed by Fujimoto [45] Two popular cancellation strategies called aggressive cancellation [44] and lazy cancellation [46] are described in this section. A. Cancellation Strategies Consider the example in Figure 10. Figure 10 about here. For simplicity, assume that cell C has one radio channel (i.e. LPC .channelNo=1 in PDES) In this example, LIN AND FISHWICK: ASYNCHRONOUS PARALLEL DISCRETE EVENT SIMULATION 9 ....

Gafni, A., "Rollback Mechanisms for Optimistic Distributed Simulation", Proc. 1988 SCS Multiconference on Distributed Simulation, pp. 61--67, February 1988.

No context found.

A. Gafni. Rollback mechanism for optimistic distributed simulation systems. In Proceedings of the SCS Multiconference on Distributed Simulation, pages 61--67, 1988.

No context found.

A. Gafni. Rollback mechanisms for optimistic distributed simulation systems. In In Proceedings of the SCS Multiconference on Distributed Simulation, volume 3, pages 61--67, July 1988. 8

No context found.

A. Gafni, "Rollback Mechanisms for Optimistic Distributed Simulation Systems", Proc. of the SCS Multi-conference on Distributed Simulation, 19, 3, 1990, pp. 61-67.

No context found.

A. Gafni. "Rollback mechanisms for optimistic distributed simulation systems". In SCS Multiconference on Distributed Simulation, V.19 N.3, pages 61--67, 1988.

No context found.

A. Gafni. Rollback mechanisms for optimistic distributed simulation systems. In B. W. Unger and R. M. Fujimoto, editors, SCS Multiconf. on Distributed Simulation, volume 19, pages 61--67, Tampa, FL, February 1988. Society for Computer Simulation.

No context found.

A. Gafni, "Rollback Mechanisms for Optimistic Distributed Simulation Systems," Proc. of the SCS Multiconference on Distributed Simulation, Vol. 19(3), pp. 61-67, 1988.

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