D. Bruce. The treatment of state in optimistic systems. In Proc. of the 9th Workshop on Parallel and Distributed Simulation, pages 40--49. ACM/SCS, June 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....into their original state locations until the state to be restored is obtained. A main problem related to the incremental approach is the need for identifying the state variables that are updated by a task execution. Interesting efforts to achieve transparency through persistent objects [4] and overloaded C operators [21] have been proposed. A technique in which saving calls of before images are automatically inserted by directly editing the application executable is presented in [26] Another checkpointing recovery scheme has been recently proposed, which relies on a ....

....benchmark for two main reasons: i) its parameters (e.g. task processing time, dimension of the state, etc. can be easily modified, ii) in the context of optimistic parallel discrete event simulation it is one of the most used benchmarks for testing performance of checkpointing recovery schemes [4, 18, 20, 22, 26]. Furthermore, the PHOLD model shows a rollback behavior similar to many other synthetic benchmarks and also to several real world models, therefore a wide class of simulation models would lead to performance data quite similar to those obtained with it. We have considered two different pairs of ....

D. Bruce, "The Treatment of State in Optimistic Systems", Proc. 9-th Workshop on Parallel and Distributed Simulation (PADS'95), pp.40-49, June 1995.

....which reuses an existing optimistic simulator. Generally speaking, conservative protocols are only suitable for a small class of simulations [Fujimoto 93] therefore we choose to focus on optimistic synchronisation protocols in this thesis. State saving is handled using persistent datastructures [Bruce 95] 10 1.2 Thesis statement A simulation oriented language can enable optimisations to be performed that are usually difficult or cumbersome, such as granularity control. Granularity control is the process of ensuring that the event granularity is large compared with the per event overheads, ....

....of APOSTLE described in this thesis is based on a compiler written in C , which uses lex and yacc [Levine 92] and a run time system, also written in C , which is based on an existing Breathing Time Buckets [Steinman 92] simulator. State saving is handled using persistent datastructures [Bruce 95] The compiler and state saving mechanisms were developed by David Bruce. The underlying simulator was developed by Chris Booth, Mike Kirton and Roy Milner, Defence Research Agency, Malvern. APOSTLE is implemented on the distributed memory Meiko CS 2 and the shared memory Sun SPARCserver 1000, ....

[Article contains additional citation context not shown here]

D. Bruce, "The treatment of state in optimistic systems", Proc. 9th Workshop on Parallel and Distributed Simulation, Lake Placid, New York, 14-16 June 1995, pp. 40-49.

....very expensive when the state size is large. This is particularly acute when each event has a small computational grain and only a small part of the state is modified. On the other hand, ISS mechanisms only log information related to changes to the state i.e. incrementally log state modifications [6, 16, 27, 26, 77, 82, 84]. ISS schemes are particularly useful in applications where only a small percentage of a large LP s state is modified per event. Generally, ISS is robust in its performance. Its cost tends to be a constant fraction of the event execution time, so that it seldom causes unexpected costs. ISS ....

....Similar behavior (rollback distances of 1 2 events) was noted in certain queueing network simulations [36] In order for PSS to be efficiently used in such applications, more frequent checkpoints would need to be made i.e. small checkpoint intervals. 2.2. 3 Incremental State Saving (ISS) In ISS [6, 16, 27, 26, 77, 82, 84], only modifications to state are logged in a backtrail. State modifications in every event need to be incrementally logged. Prior to a word being modified, its old value and address are logged. During state restoration, the backtrail is traversed in the reverse order, from the most recent event ....

[Article contains additional citation context not shown here]

D. Bruce. Treatment of state in optimistic systems. In Proceedings of the 9th Workshop on Parallel and Distributed Simulation (PADS95), pages 40--49, Lake Placid, New York, June 1995. 186

....small rollback extent. A main problem related to this solution is the need for identifying the parts of the state that are updated by event execution. Due to cost issues the use of special hardware [8] is usually infeasible. Interesting efforts to achieve transparency through persistent objects [3] and overloaded C operators [18] have been proposed. A technique in which saving calls of before images are automatically inserted by directly editing the application executable is presented in [22] Finally, multiplexed [5] and hybrid [15] state saving are schemes that mix features of both ....

D. Bruce, "The Treatment of State in Optimistic Systems", Proc. 9-th Workshop on Parallel and Distributed Simulation, pp.40-49, June 1995.

....information during forward execution. The second problem is to solve the first in a way which is transparent to the programmer, i.e. that doesn t substantially complicate model development. State saving mechanisms can be categorized as copy state saving (CSS) or incremental state saving (ISS) [1, 2, 3, 13]. In CSS mechanisms, a checkpoint of the entire object s state is taken on each event. CSS mechanisms are processor intensive and consume a large amount of memory. An optimization of the CSS strategy is to reduce the frequency of checkpointing using periodic state saving (PSS) 5, 11, 12] ....

.... My link; int My index, Id; sk time Cell transm delay; RPT STYLE Rpt; char Report 1, Report 2; void report formatted(sk time) void report(sk time) void reset stats(sk time) void report wsv(sk time) nm buffer Buffer; int Cells received; int Link state; int Cells sent[AFR N CLASS][2]; afr tally State ctd[AFR N CLASS] recover: public: int id(void) afr link link(void) void terminate(void) void initialize( void process(const sk event ) nm output( Figure 1: Recoverable state declaration example 5 Conclusions Two significant barriers to the ....

D. Bruce. Treatment of state in optimistic systems. In Proceedings of the 9th Workshop on Parallel and Distributed Simulation (PADS95), pages 40--49, Lake Placid, New York, June 1995.

....a fraction of the state is updated in each event execution. In such applications it may be inefficient or even infeasible to save copies of the complete state which can be in the order of hundreds of kilobytes [6] In such applications it is often appropriate to use incremental state saving (ISS) [2, 4, 24, 25, 26], in which only the updated parts of the state are saved. Thus the state saving mechanism builds a chain of state changes. In case of rollback the state is restored by undoing these changes. Few systems, if any, implement transparent ISS primarily due to problems associated with identifying which ....

....The SPEEDES environment features several interesting and efficient techniques to implement ISS [24, 25] Good performance results with ISS in the context of VLSI simulations [2] and simulations of large telecom networks [6] have been reported. ISS has also been implemented using persistent objects [4] in an interesting effort to achieve a transparent implementation of ISS. However, these implementations of ISS have the common denominator that they put the responsibility on the user of writing either special purpose code for the ISS or calling special functions when updating state variables or ....

D. Bruce, "The Treatment of State in Optimistic Systems", Proceedings of the 9th Workshop on Parallel and Distributed Simulation (PADS95), pages 40-49, June 1995.

....The SPEEDES environment features several interesting and efficient techniques to implement ISS [15, 16] Good performance results with ISS in the context of VLSI simulations [1] and simulations of large telecom networks [4] have been reported. ISS has also been implemented using persistent objects [3] in an interesting effort to achieve a transparent implementation of ISS. However, all these systems require that the user supplies special code for the ISS methods, either in the form of explicit calls to state saving functions or code which copies user defined objects. From a user point of view ....

D. Bruce, "The Treatment of State in Optimistic Systems", Proceedings of the 9th Workshop on Parallel and Distributed Simulation (PADS95), pages 40-49, June 1995.

....In practice yet more context is needed for a complete definition. The most important omission here is that of the computer s memory, which will contain objects whose state must be saved in case of rollback. In the full semantics for APOSTLE, we model the computer s memory using a persistent store [5], which supports a rollback operation. Since the semantics of APOSTLE tell us what APOSTLE programs mean, we can implement an APOSTLE interpreter simply by writing C classes that represent the continuations found in the semantics. A compiler for APOSTLE is equally simple: it just emits code for ....

D. Bruce. The treatment of state in optimistic systems. In Ninth Workshop on Parallel and Distributed Simulation, pages 40--49, Lake Placid, New York, 14--16 June 1995.

.... to save and restore objects states, whilst others restrict the form of the state (for example by prohibiting the use of dynamic memory data structures) Fortunately, several techniques for automatic support for (incremental) state saving have been developed in recent years; see, for example, Bruce 95, West Panesar 96] Despite recent work addressing the issues of ease of use, we believe that the future of PDES lies in simulation oriented languages that can completely hide the PDES protocol specific complexities from simulation practitioners, thereby allowing them to concentrate their efforts ....

....If its scope does not suffer any context switches, then there can be no opportunity to roll its thread back and it therefore need not be subject to state saving. Depending on the scheduling policy, it may be safe to extend this to variables whose scopes that do not elapse time; i.e. events. Bruce 95] discusses these and other state saving optimizations in more detail. As for language definitions, the use of effects for detecting optimizations leads to increased generality. Here the general case is provided as the default, but the common special case is easily identified and implemented ....

David Bruce, "The treatment of state in optimistic systems", pp. 40--49 in Proc. 9 th Workshop on Parallel and Distributed Simulation, Lake Placid, New York, 14--16 June 1995; proceedings published as Simulation Digest, Vol. 25, No. 1, July 1995.

No context found.

D. Bruce. The treatment of state in optimistic systems. In Proc. of the 9th Workshop on Parallel and Distributed Simulation, pages 40--49. ACM/SCS, June 1995.

No context found.

D. Bruce, "The Treatment of State in Optimistic Systems", Proc. 9th Workshop on Parallel and Distributed Simulation (PADS'95), pp.40-49, 1995.

No context found.

D. Bruce, "The Treatment of State in Optimistic Systems", Proc. 9th Workshop on Parallel and Distributed Simulation (PADS'95), pp.40-49, June 1995.

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