J.S. Steinman. Breathing time warp. In Proceedings of the seventh workshop on Parallel and distributed simulation, pages 109--118. ACM Press, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....sorts of relative network delays one might see with medium grained workloads running on a workstation cluster or fine grained workloads on a large scale parallel architecture. For the simulations discussed in this paper, we used the optimistic Breathing Time Warp (BTW) synchronization protocol [4] that combines the Time Warp and Breathing Time Buckets protocols. At the beginning of each global virtual time (GVT) cycle, messages are sent aggressively using Time Warp. Later in the cycle, all messages are sent risk free using Breathing Time Buckets. Two run time parameters in SPEEDES ....

J. Steinman, "Breathing Time Warp", Proceedings of the 1993.

....yet achieved little success for general realworld simulations. Second, it is a perfect local rollback protocol. The well known distinction between aggressiveness and risk predicted that there should be a new class of protocols that allow aggressiveness but not risk [15] Two such protocols [5, 17] had been already developed, but the lookback based protocol is the first asynchronous local rollback protocol which controls the aggressiveness by maintaining an adequate amount of lookback. The understanding and application of lookback are still in early stages. Although lookback is always ....

J. Steinman. Breathing time warp. In Proceedings of the 7th Workshop on Parallel and Distributed Simulation, pages 109--118, 1993.

.... of local rollback mechanisms [25] Anti messages are strictly avoided but local rollbacks are still necessary, suggesting that lookback based pro tocols allow for aggressiveness, but not risk [57] Two such protocols previously known, SRADS with local rollback [25] and Breathing Time Buckets [63], require components to work collaboratively, by exchanging event information, to avoid send ing any erroneous messages. In lookback based protocols, however, each component can determine alone whether a message can be sent out (by allowing or disallowing execution of the corresponding event) ....

J. Steinman. Breathing time warp. In Proceedings of the 7th Workshop on Parallel and Distributed Simulation, pages 109-118, 1993.

....the universal lookback procedure fall into the class of synchronization protocols known as local rollback [7] the idea of which is to avoid anti messages but to allow events to be aggressively processed. Two such protocols previously known, SRADS with local rollback [7] and Breathing Time Window [8] require logical processes to work collaboratively, by exchanging event information, to avoid sending any erroneous message. In lookback based protocols, however, each logical process alone can determine whether a message can be sent out (by allowing or disallowing execution of the corresponding ....

Steinman, J. Breathing Time Warp, in Proc. of the 7th Workshop on Parallel and Distributed Simulation. 1993. San Diego. p. 109--118.

....per time bucket. Depending on the underlying simulation model, this is not always possible. This is no problem for the original TW paradigm, which does not use buckets and can process events with unlimited optimism. Breathing Time Warp (BTW) aims to combine the best of TW and BTB. Steinman [37] notes that the main problem su ered by TW is the rapidly growing possibility of cascading antimessage explosions when runaway LPs get more and more out ahead of the rest. Events with timestamp far away from the current GVT tend to have a larger chance of being rolled back than events close to ....

....events are processed optimistically, while new generated messages are postponed until the global EH (event horizon) has been determined, using the local EH of each simulation node. New messages for the current node need not be postponed and can be optimistically processed. The BTW algorithm [37] can be described as follows: before the next GVT is determined, the rst N risk (a user selected run time parameter) events will be processed locally on each LP using the TW mechanism. After that, events are processed using BTB. When N gvt events have been processed, or when the event horizon has ....

Je S. Steinman. Breathing Time Warp. In Proceedings of the 1993.

.... incremental state saving which saves only a subset of each LP s state each time the LP is activated[6] Two techniques to reduce the number of states that are saved are: to use a checkpoint interval that is greater than one[ and to constrain optimism in some manner such as Breathing Time Warp[7]. Algorithms that attempt to recover from a memory stall. The previous algorithms seek to delay the onset of a memory stall. Although this may be su#cient in some cases, there will always be those cases that will eventually stall due to memory starvation. Algorithms which attempt to recover from ....

J. S. Steinman. Breathing Time Warp. In R. Bagrodia and D. Je#erson, editors, Proc. 1993.

....to control the cascading of rollbacks. Several aggressive windowing algorithms have been proposed: Moving Time Window [24] and Window based throttling [25] Unified Distributed Simulation System [26] 18 Bounded Time Warp [27] Aggressive Global Windowing Algorithm [28] and Breathing Time Warp [29]. These algorithms apply the same basic principle described above. The major di#erences lie in how the time window is determined. Notable among these are UDSS and Breathing Time Warp. The latter uses a fixed size window based on the number of events rather than logical time to control the risk of ....

J. Steinman, "Breathing Time Warp," in Proceedings of the 7th Workshop on Parallel and Distributed Simulation, 1993, pp. 109--118.

....with a list of assumptions that were made during their generation. This can enable receiving processes to lter their input queue for messages that will de nitely be cancelled at a later stage. Another approach which operates on the output queue of a process is known as breathing time warp [37]. Here the processes of the system move in a roughly synchronous fashion through multiple phases of operation. One phase is normal Time Warp operation. A process may decide to hold its breath, or withold the transmission of its output messages, if it has advanced far beyond GVT. This is done on ....

Steinman, J. S. Breathing Time Warp. In Proceedings of the 7 th Workshop on Parallel and Distributed Simulation (PADS '93) (July 1993), pp. 109-118.

....Warp.html. SPEEDES SPEEDES stands for Synchronous Parallel Environment for Emulation and Discrete Event Simulation. SPEEDES began development in 1990, also at the Jet Propulsion Laboratory, as a system for experimenting with a particular synchronisation scheme [36]. It later grew to become a testbed for PDES synchronisation schemes, including Time Warp . It has the attractive feature of supporting real time monitoring of a simulation s performance by a user. A user employs the system by writing a con guration le for the system s main program. The con ....

Steinman, J. S. Breathing Time Warp. In Proceedings of the SCS Western Simulation Multi-conference (1991), vol. 23, pp. 109-118.

....they can be most of the time. The potential gains found in lifting the causality restrictions of conservative simulation is the conceptual basis for optimistic synchronization algorithms. 2.2. 2 Optimistic Synchronization Space Time simulation [16] Time Warp [32, 39] and Breathing Time Warp [82] 1 are a few of the more popular optimistic discrete event synchronization methods in use today. All of these algorithms employ asynchronous communication, have no global clock, and parallel processes are free to advance as there is no synchronization between them. If the processes are not ....

Steinman, J. S. Breathing Time Warp. In Proc. of the 7th Workshop on Parallel and Distributed Simulation (PADS 93) (July 1993), Society for Computer Simulation, pp. 109--118.

....execute the actions that they were supposed to do at this time and then report their next event times. The central clock should look at these times and decide the next time to be broadcast and when this broadcast should be done. The clock uses a variation of the breathing time bucket al..gorithm [4] [7]. Some of the important considerations for the clock are: the next time broadcast should be the minimum of the times reported by the components, the time should not be advanced when there is some message in transit between two components. So the components keep track of the messages sent and ....

J.S. Steinman, "Breathing Time Warp," Proceedings of the

....may also provide opportunities for improving performance, but they may not be easily predicted. Consider the vector for interaction between algorithms that is the communication patterns between processes. Process oriented algorithms tend to alter these patterns. The breathing Time Warp [12] and adaptive periodic checkpointing schemes [2] 7] are examples of this. As a result any algorithm that is operating concurrently and depends upon communication patterns for its operations will be a ected. For example one of the dynamic repartitioning schemes proposed in [13] tries to bring ....

Steinman, J. S. Breathing Time Warp. In Proceedings of the 7 th Workshop on Parallel and Distributed Simulation (PADS '93) (July 1993), pp. 109-118.

....a potential source of system errors. At the other end of the spectrum, there are modular implementations which break down the functionality of the scheduler into small pieces using an object oriented design approach. SPEEDES is the most widely used Time Warp system implemented in this framework [28, 29, 30]. Implemented in C , SPEEDES exports a plug and play interface which allows developers to easily experiment with new time management, data distribution and priority queue algorithms. All of this functionality and flexibility comes at a performance price. In a recent study conducted on the ....

J. S. Steinman. "Breathing Time Warp". In Proceedings of the 7 th Workshop on Parallel and Distributed Simulation (PADS '93), pages 109--118, May 1993.

....a lot of rollback and memory space overhead. In order to obtain better performance, many techniques are proposed by researchers who believe that the TW algorithm with low overhead will be the best: adaptive process scheduling [79] bounding window [19, 91, 103] synchronization granularities [98], memory management [2, 24, 39, 44, 66, 67] and partitioning and load balancing [19, 57, 60, 69, 96] In the Time Warp algorithm, logical processes may advance too far ahead and produce a significant rollback cost. Limiting optimism may be a solution to reduce the rollback cost. The first method ....

....gate can be selected dynamically based on the instantaneous rollback ratio [18] Because of the dynamic nature of the TW, it is very difficult to optimize the window size with this approach. By using frequent global synchronizations, optimism was limited in the Breathing Time Warp of the SPEEDES [98] and the Bounded Time Warp [103] As a result, rollback overhead is reduced in those schemes. However, their performance is only slightly better than the TW due to the overhead of frequent global synchronizations. A major problem of parallel logic simulation techniques is that they do not produce ....

[Article contains additional citation context not shown here]

J. Steinman, "Breathing Time Warp," Proc. of the 7th Workshop on Parallel and Distributed Simulation, 1993. 89

.... aggressive windowing algorithms have been proposed: Moving Time Windows [SoBW88] Filtered Rollback [LuWS89] Window based throttling [ReJe89] Unified Distributed Simulation System [McAf90] Bounded Time Warp [TuXu92] Aggressive 18 Global Windowing Algorithm [Dick93] and Breathing Time Warp [Stei93]. Most of these algorithms follow the same basic principle described above. The differences are in the methods used to determine the time window. Notable among these are UDS and Breathing Time Warp. The former uses independent windows for each LP. The latter uses a fixed size window based on the ....

....former is more likely to be rolled back. The situation is further deteriorated if the second LP is about to send a message to the first. This reasoning suggests that logical time should be included in the mapping. We note that this mapping is equivalent to the Breathing Time Buckets algorithm [Stei93]. ii) EP i = MT i minimum MT among all LP s, where we define MT i (minimum time) as the smaller of (i) the logical clock of LP i and (ii) the minimum timestamp of all messages sent by LP i which are either in transit or have not yet been processed by the receivers. Note, the minimum of the ....

[Article contains additional citation context not shown here]

Steinman, J.S., "Breathing Time Warp", Proceedings of the 7th Workshop on Parallel and Distributed Simulation, May 1993, 109-118.

....also be created. These events are then inserted into the event queue to be executed later. In a distributed simulation, there are multiple such event queues in the system. The key to success is to let the simulation proceed as efficiently as possible without violating the causality constraints [28]. Each of the nodes network have a separate event queue. We now have to decide on how to let the simulation clock proceed. One possibility (and which we have followed 32 in our simulator) is to have a central clock that collects the next event time from all the event queues, computes the next ....

....and network have their own local version of virtual time and run independently of others except when there is a need for interaction. They must satisfy causality constraints [17] to ensure correctness of the simulation. The RAPIDS simulator uses a variation of the Breathing Time Buckets technique [28]. Each process in the simulator has a notion of the Local Event Horizon, which is the time it can proceed to, without violating causality. The central clock maintains the Global Event Horizon which is always the minimum of the Local Event Horizons, and broadcasts this as the current time of the ....

J.S. Steinman, "Breathing Time Warp," Proceedings of the 1993 Workshop on Parallel and Distributed Simulation, 1993.

....and the processes representing these nodes exchange messages to check for virtual message passing. If a process observes that it does not have any message to be received from the neighbors, it can safely advance by an amount of time equal to the minimum message passing time among the neighbors [Fuji90, Stei93]. 3.3 Previous Work in Real Time Simulation 3.3.1 Stress Simulator: This is a continuous time uni process simulator developed at University of York [Auds94] Language extensions have been used to describe a system and the associated task set that runs on the simulator. In this scheme, the clock ....

.... N1 N2 VP1 VP2 VP3 VP4 Clock Workstations PVM interface Unix Processes PVM tasks Simulator Functions Figure 2: Overview of the Simulation Mechanism have a local virtual time and they are synchronized by a variation of the breathing time buckets methods of pessimistic synchronization [Stei93]. The implementation of this scheme is discussed in chapter 4. 22 The schematic layout of the design of the distributed simulator is shown in Figure 2. The PVM software is run on top of a set of one or more workstations. A set of PVM tasks can be run on top of the PVM interface and these tasks ....

[Article contains additional citation context not shown here]

J. S. Steinman, "Breathing Time Warp", Proceedings of the 1993 Workshop on Parallel and Distributed Simulation, 1993. 120

....virtual system behavior indirectly through a change of parameters. The console executes the user control commands, such as start up or termination of the simulation. 7.4. 2 Clock Mechanism The synchronization for the simulator is provided by using a variation of Breathing Time Buckets technique [66]. The Global Virtual Time(GVT) for the simulator is controlled by a central clock server. This is a PVM process (task) that broadcasts the current virtual time that is agreed over the entire system (including the network and processor LPs) The central clock maintains a table entry for each of the ....

Steinman, J. S. "Breathing Time Warp", In Proceedings of the 1993 Workshop on Parallel and Distributed Simulation, pp. 109-118, 1993

....communication bandwidth, etc. In the worst case, the rollback may be cascading and recursive, resulting in undesired rollback thrashing. Many methods have been proposed to reduce rollback thrashing and its overhead. Most of the solutions adopt the one sided approach to constrain the optimism [2, 7, 8, 10, 32, 33]. In addition, some form of synchronization, such as barrier or polling, is often required to bound the computations. As a result, the performance of these schemes largely depends on the events available for processing within the bounds. If the number and granularity of events available is ....

....Simulation is complete when the lower bound of time window passes the duration of simulation. 10 Constrained Window Adaptive Control Direct Control Indirect Control Process Migration Time Warp Control Schemes Moving Time Window [29] Breathing Time Bucket [32] Breathing Time Warp [33] Adaptive Time Warp [2] Probabilistic Cost Expected Function [10] Degree of Optimism [8] Near Perfect State Information [31] Virtual Time Window [6] Discard Message [21] Message Sendback [19] Artificial Rollback [20] Adaptive Memory Management [7] Effective ....

[Article contains additional citation context not shown here]

J. S. Steinman, "Breathing Time Warp," Proc. of the 7th Workshop on Parallel and Distributed Simulation, pp. 109-118, 1993.

....exchange timestamped event messages. TWOS was designed to run on bare hardware, with a bottom layer responsible for low level tasks (context management, message communication, interrupt handling, etc. 2 and a top layer responsible for time warp mechanisms (rollback, anti messages, etc. SPEEDES [40, 41, 39] is a direct descendant of TWOS; it is written in C , with an object oriented view. Like TWOS, SPEEDES is event based. It supports multiple synchronization protocols and a delta exchange mechanism [41] for incremental state saving [39] The Georgia Tech Time Warp (GTW) is a library supporting ....

....top layer responsible for time warp mechanisms (rollback, anti messages, etc. SPEEDES [40, 41, 39] is a direct descendant of TWOS; it is written in C , with an object oriented view. Like TWOS, SPEEDES is event based. It supports multiple synchronization protocols and a delta exchange mechanism [41] for incremental state saving [39] The Georgia Tech Time Warp (GTW) is a library supporting the construction of event based parallel discrete event simulations [18, 33] GTW is available for various shared memory parallel machines, and a new version of GTW which runs over distributed memory ....

[Article contains additional citation context not shown here]

J. S. Steinman. Breathing Time Warp. In Proceedings of the 7th Workshop on Parallel and Distributed Simulation, pages 109--118, 1993.

....authors [11, 12] as a mechanism for synchronization which circumvents the drawbacks described in the preceding section. The method is based primarily on observations of real manufacturing systems. Figure 5. Connection of subsystems 6 SIMULATION MARCH 1999 Steinman s Breathing Time Warp approach [13,14] shares common features with the Time Bucket method presented in this study. Event management in the Breathing Time Warp algorithm, however, is too strict to attain effective simulation for a virtual factory. The bucket size should be defined such that it is the minimum time interval between all ....

Steinman, J.S. "Breathing Time Warp." Proceedings of the 1993 Workshop on Parallel and Distributed Simulation, pp 109-117, 1993.

....generate events for each other so that the cluster would be simulated all the way until the end of the simulation time without running out of events. This is not desirable, since we don t want to create almost certain causality errors. Therefore we need to explore techniques for limiting optimism [6, 60, 65, 66, 67, 73]. These techniques involve setting 25 some limit on how far ahead an entity may simulate. The bound could be based on the amount of simulation time which elapses, the amount of memory required for state saving, or some information which the algorithm has about the simulation. We may be able to ....

Je# S. Steinman. Breathing time warp. In Proceedings of the 7th Workshop on Parallel and Distributed Simulation, pages 109--118, May 1993.

....rollbacks carried out, etc. This adds additional overheads to the standard Time Warp approach. In Time Warp, if there are excessive numbers of rollbacks, explosion of anti messages can result, which in turn leads to unstable performance of the simulation. Breathing Time Buckets (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 ....

....can result, which in turn leads to unstable performance of the simulation. Breathing Time Buckets (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 certain applications. 10 5 Mathematical Performance Models Mathematical performance modeling of ....

Jeff S. Steinman. Breathing time warp. In Proceedings of 7th Workshop on Parallel and Distributed Simulation (PADS'93), pages 109--118, San Diego, California, May 1993.

....a (common or independent) window are executed aggressively. Similarly only those messages within a (possibly different) window are sent out. This ensures that all of the LP s remain close to each other in logical time. Uncontrolled echoing and cascading rollbacks cannot occur. Examples are [SoBW88, LuWS89, ReJe89, McAf90, TuXu92, Dick93, Stei93]. Space based: Here, the boundaries for limiting optmism are spatial rather than temporal. In general, the processors are divided into clusters which use Time Warp internally. Interaction among clusters follows some non aggressive protocol. Examples are [Gima89, RaAT93] An interesting ....

....well known that good load balancing requires good estimation of tunable parameters such as thresholds and maximum number of transfer hops. Not surprisingly, we observe that all existing adaptive protocols admit tunable parameters of some sort (the time window bound in [TuXu92] N 1 and N 2 in [Stei93], cluster size in [RaAT93] blocking window size in [BaHo90] the scaling factor c i in [HaTr94] and M f and M p in [DaFu94] Clearly, there is a need to invent a scheme by which adaptive protocols may tune the values of such parameters dynamically. The Adaptive Time Warp algorithm [BaHo90] ....

Steinman, J.S., "Breathing Time Warp", Proceedings of the 7th Workshop on Parallel and Distributed Simulation, May 1993, 109-118.

....for dynamically recalculating the value of a parameter which defines the percentage of states to be recorded. The exploitation of the event history joins this sparse state saving approach and several techniques introduced with the aim of limiting the optimism of Time Warp (for example those in [3, 4, 21]) with the difference that the reduction of the rollback overhead is achieved by reducing the rollback cost instead of the rollback frequency. An experimental study of synthetic workloads is presented for a performance comparison with the adaptive methods in [19] and in [5] The results show ....

.... saving scheme we propose is based on the belief that the likelihood of a rollback point being contained in the virtual time interval I , the delimiting points of which are timestamps of successive events, increases with the distance between those points (a similar observation has been made in [21], with the difference that the starting point of the interval is the GVT of the simulation) So, let us suppose that LV T is the current local virtual time of an LP (the current state of the LP is denoted as SLV T ) and the timestamp of the next event e next to be scheduled is T . If T ....

J. Steinman, "Breathing Time Warp", Proc. 7-th Workshop on Parallel and Distributed Simulation, pp.109-118, May 1993.

....however, creates at least one significant problem: the efficient communication of target state between physical nodes of the host machine. The Wisconsin Wind Tunnel (WWT) 11] is one such parallel simulator, which runs on a Thinking Machines CM 5. WWT uses conservative, discrete event simulation [5, 8, 10, 15] to accurately calculate the logical execution time of the target application. Superior performance is obtained through direct execution [3] of identical target and host instructions on the native CM 5 hardware. WWT s solution to the problem of communicating simulation state is to guarantee ....

Jeff S. Steinman. Breathing Time Warp. In Proceedings of Parallel and Distributed Simulation, pages 109--118, 1993.

....of adaptive protocols we have described in [SrRe94] which we call NPSI adaptive protocols, also satisfy the AAWP assumptions. Windowing algorithms in which the windows are computed individually for different LP s (such as Unified Distributed Simulation system [McAf90] and Breathing Time Warp [Stei93]) are AAWP s as well since the LP s wait when they reach the ceilings of their independent windows. Note, global windowing algorithms do not fit the AAWP model since the global window forces all LP s to synchronize before any of them can proceed. Protocols in which optimism is limited based on ....

Steinman, J.S., "Breathing Time Warp", Proceedings of the 7th Workshop on Parallel and Distributed Simulation, May 1993, 109-118.

....directly execute all load and store instructions (excluding instruction fetches) rather than just computation instructions. Second, WWT exploits parallelism by simulating each target processor node on a separate host node. WWT uses a conservative synchronous discrete event simulation algorithm [15, 16, 20] to insure causal event ordering between nodes [12] This algorithm allows each node to directly execute instructions within a fixed length quantum (also called a fixed time window or time bucket [20] At the end of a quantum, the host nodes must synchronize to ensure that all target messages ....

....host node. WWT uses a conservative synchronous discrete event simulation algorithm [15, 16, 20] to insure causal event ordering between nodes [12] This algorithm allows each node to directly execute instructions within a fixed length quantum (also called a fixed time window or time bucket [20]) At the end of a quantum, the host nodes must synchronize to ensure that all target messages have arrived, so that events that must be processed in the next quantum are properly ordered. This is illustrated in Figure 1: messages sent during quantum Q i are guaranteed to arrive before the barrier ....

Jeff S. Steinman. Breathing Time Warp. In Proceedings of Parallel and Distributed Simulation, pages 109--118, 1993.

....that limiting optimism has little or no positive effect, and can have very negative effects if applied too strenuously. A number of other adaptive synchronization algorithms have been proposed that limit the optimism with the use of time windows beyond which events are not allowed for processing [2, 11, 26, 58]. Most of the these algorithms fix the time windows based on event histories and statistical methods. The biggest challenge to all these algorithms is in selecting the size of time windows so that neither they limit the optimism too much nor they allow too many erroneous computations. So far, ....

Steinman, J. S. Breathing time warp. In Proc. of the 7th Workshop on Parallel and Distributed Simulation (PADS 93) (July 1993), Society for Computer Simulation, pp. 109--118.

....that the processor is not utilized when all the processes are penalized by the system, thereby wasting CPU cycles. Performance results from such techniques has discouraged further use of this technique. Similar approaches to stall LP event processing are reported in [8, 16] Breathing Time Warp [45] is a combination of Breathing Time Buckets [44] and Time Warp. This technique works on the principle that events closer to GVT have a lower probability of being rolled back. Thus N1 events closer to GVT are executed optimistically, and N2 events after that point are executed using breathing time ....

Steinman, J. S. Breathing time warp. In 7 th Workshop on Parallel and Distributed Simulation (May 1993), A. H. Rutan, Ed., IEEE Computer Society Press, pp. 109--118.

....Time Ceiling (Matsumoto and Taki, 1993) is based on a similar concept, although the window sizes are chosen from a set of discrete values. The controlling parameter is very simple and does not contain global information such as antimessage count, positive message count, etc. Breathing Time Warp (Steinman, 1993) is a combination of breathing time buckets and time warp. This technique is based on the principle that events closer to GVT have a lesser probability of being rolled back. Thus N 1 events close to GVT are executed optimistically, and N 2 events after that point are executed using breathing time ....

Steinman, J. S. (1993). Breathing time warp. In 7th Workshop on Parallel and Distributed Simulation, pages 109-118. SCS.

....will emerge in practice. The first are those applications that contain a substantial amount of computation per event. For instance, models involving projectiles in free flight (e.g. missiles) fall into this category because a substantial amount of computation is required to compute trajectories [16]. A second class of applications involves simulation of large numbers, thousands or tens of thousands) of light weight objects with relatively little computation in each event. Each object will typically contain only a small amount of state. Large scale simulations of telecommunication ....

J. S. Steinman. Breathing time warp. In 7 th Workshop on Parallel and Distributed Simulation, volume 23, pages 109-- 118. SCS Simulation Series, May 1993.

....regard are described in [7, 18] Broadly, there are two classes of optimism control schemes: non adaptive and adaptive. System parameters, e.g. window sizes remain static in non adaptive schemes, where as they are dynamic in the adaptive schemes. A number of non adaptive schemes have been proposed [22, 24, 25, 19, 20, 2, 21, 17, 16]. These schemes typically leave the specification of parameters to the user, or rely on heuristics to statically set their parameters at the beginning of the simulation. The optimal parameter settings are in general dependent on the application. Adaptive schemes for optimism control have a better ....

J. S. Steinman. Breathing time warp. In 7 th Workshop on Parallel and Distributed Simulation, volume 23, pages 109--118. SCS Simulation Series, May 1993.

.... using memory management protocols like cancelback (Das and Fujimoto 1997b; 2 GVT is the minimum timestamp of all the unprocessed events in Time Warp and defines the commitment horizon (Jefferson 1985) Das and Fujimoto 1997a) v) limiting the number of events each LP may execute beyond GVT (Steinman 1993). All these mechanisms have been shown to perform better than the purely optimistic Time Warp for certain simulation models. In general, it is believed that an appropriate throttling of Time Warp execution (i.e. blocking one or more LPs even if they have unprocessed events in their future event ....

Steinman, J. S. 1993. Breathing Time Warp. In Proceedings of the 7th Workshop on Parallel and Distributed Simulation, pp. 109--118.

No context found.

J.S. Steinman. Breathing time warp. In Proceedings of the seventh workshop on Parallel and distributed simulation, pages 109--118. ACM Press, 1993.

No context found.

Steinman J. S., Bagrodia R. and Jefferson D., "Breathing Time Warp", in Proc. of the 1993.

No context found.

Steinman J. S., Bagrodia R. and Jefferson D., "Breathing Time Warp", in Proc. of the 1993.

No context found.

J.S. Steinman. "Breathing Time Warp". In 7th Workshop on Parallel and Distributed Simulation (PADS'93), pages 109--118, May 1993.

No context found.

J.S. Steinman. "Breathing Time Warp". In 7th Workshop on Parallel and Distributed Simulation (PADS'93), pages 109--118, May 1993.

No context found.

Steinman, J. S., R. Bagrodia, and D. Je#erson: 1993, `Breathing Time Warp'. In: Proc. of the 1993 Workshop on Parallel and Distributed Simulation. pp. 109--118.

No context found.

J. S. Steinman. Breathing Time Warp. In Proceedings of the 7th Workshop on Parallel and Distributed Simulation, pages 109--118, May 1993.

No context found.

J. S. Steinman, "Breathing Time Warp," Proc. of the 7th workshop on Parallel and Distributed Simulation, pp. 109-118, 1993.

No context found.

J.S. Steinman. 1993. "Breathing Time Warp." In Proceedings of the 7th Workshop on Parallel and Distributed Simulation. IEEE, 170-178.

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