S. Kuo and D. Moldovan. Implementation of multiple rule firing production system on hypercube. J. Parallel and Distr. Computing, 13(4):383-- 394, December 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....the execution as well as for the decomposition of the system. Our main interest is in the later, since it makes the divide and conquer approach feasible to lower the complexity of the analysis and synthesis of the system. In this thesis we refer to an approach introduced by Kuo and Moldovan [KM91]. They use the notion of context, which groups the rules that interfere with each other. Several contexts can be executed in parallel if they are mutually independent. Cheng proposes similar parallel execution model [Che93] providing also the means of determining the response time bound for a ....

....first identified using the rule transition graph and then further investigated using our approach. This would additionally decrease the complexity of the synthesis, since the algorithm would only focus on the limited set of rules and variables. ffl The parallel execution of the rule based program [Che93, KM91] is a relatively new trend and can radically decrease the response time. The methods that determine if two rules can be fired in parallel consider the variables in the enabling condition and assignment part of the rule. Our synthesis method changes the enabling condition and in the future we may ....

S. Kuo and D. Moldovan. Implementation of multiple rule firing production system on hypercube. J. Parallel and Distr. Computing, 13(4):383-- 394, December 1991.

....ffl At the rule set level, individual rule is not decomposed. Intead, parallelism is achieved by dividing the rule set into multiple partitions, each of which consists of multiple rules. This approach is adopted in most of the research on multiple rule firing for parallel production systems [11,13,15,17,20,21,24,26]. 3 ffl At rule level, copy and constraint technique [25] creates multiple copies of the same rule, where each copy is constrained to match a subset of the data. In deductive databases, an equivalent idea using discriminating predicates is also independently reported in [8,30] ffl At relation ....

Steve Kuo & Dan Moldovan, "Implementation of Multiple Rule Firing Production Systems on Hypercube," Journal of Parallel and Distributed Computing, Vol. 13, December 1991, 383--394.

....utilizes a new parallel conflict 13 resolution scheme to maximize performance by firing as many rules as possible. The new parallel conflict resolution scheme is based upon a dependency analysis amongst the rules so that rules can be fired simultaneously without any undesirable side effects [24]. To optically implement this parallel conflict resolution scheme, an algorithm creates a conflict resolution control matrix (CRCM) for any given rulebase. To increase readability, this algorithm is shown in Appendix A. The CRCM performs parallel rule selection by taking a 1 D SRV as an input, ....

....Control Matrix (CRCM) is a novel parallel optical conflict resolution method that can resolve conflicts amongst the selected rules in a single step. Here, we describe how a CRCM is created for any given rulebase. To create a CRCM, the dependency between rules must be identified as follows [24]: ffl Inhibit Dependency : When the firing of rule R j modifies facts such that rule R i cannot be a selected rule, R i is said to be inhibit dependent on R j . ffl Enable Dependency : When the firing of rule R j modifies facts such that rule R i becomes a selected rule, R i is said to be enable ....

[Article contains additional citation context not shown here]

S. Kuo and D. Moldovan, "Implementation of Multiple rule firing production systems on hypercube," Journal of Parallel and Distributed Computing, 13, 383-394, (1991).

.... Head: Design and Control of PRFPS Special Characters: ffl f : Left bracket ffl g : Right bracket ffl : Caret Special Fonts: Bold face Typewriter Font Italic Font 1 Introduction The parallel firing of rules has been explored as a method of increasing the performance of rule based systems [6, 7, 11, 12, 16, 13, 17, 21, 23, 29, 34] . This paper summarizes the results of the research performed at the University of Massachusetts using a Lisp based OPS5 that exploits parallelism at the rule instantiation level, action level, and match level on a shared memory multicomputer. Our major concern in this paper is maximizing the ....

....of instantiations, UMPOPS incorporates a meta level notation for annotating certain rules as modechangers. The scheduler then ensures that all domain rules have executed before allowing a mode changing rule to fire. Similar mechanisms have been independently derived by a number of researchers [8, 12, 18, 23] . In addition, we have found that new correctness criteria must be developed to support asynchronous execution of rules. 2.1.1 Experiments with Rule Firing Policies In this section, we illustrate the advantages of asynchronous rule firing using the results of a simple circuit simulator ....

[Article contains additional citation context not shown here]

Steve Kuo and Dan Moldovan. Implementation of multiple rule firing production systems on hypercube. In Proceedings of the Ninth National Conference on Artificial Intelligence, pages 305--309, 1991.

....large scale and scalable parallelism in production systems, we analyzed several benchmark problems and the corresponding rule based programs implemented in OPS5. These programs have been widely used in previous studies to evaluate the effectiveness of language extensions and compilation techniques [8, 11, 9, 13]. While the amount of parallelism found in previous research has typically been modest and not necessarily scalable, we found, contrary to previous expectations, that most of these programs had the potential for massive and scalable parallelism. In this section, we give the results of simulated ....

Steve Kuo and Dan Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal of Parallel and Distributed Computing, 13(4):383--394, December 1991.

....is selected for execution even though multiple rules can be executed (or fired) without violating rule priorities. In the field of Artificial Intelligence, a significant amount of research has been focused on executing rules in parallel and at the same time maintaining their control structures [38, 25, 40, 27]. Similar to our system, RUBIC [27] and PARULEL [40] provide several control constructs to control the parallel execution of rules. Nevertheless, they do not guarantee the serializable execution of rules. Rule systems presented in [38, 25] guarantee serializability during the parallel execution of ....

....rules can be executed (or fired) without violating rule priorities. In the field of Artificial Intelligence, a significant amount of research has been focused on executing rules in parallel and at the same time maintaining their control structures [38, 25, 40, 27] Similar to our system, RUBIC [27] and PARULEL [40] provide several control constructs to control the parallel execution of rules. Nevertheless, they do not guarantee the serializable execution of rules. Rule systems presented in [38, 25] guarantee serializability during the parallel execution of rules by a static analysis of ....

S. Kuo and D. Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal on Parallel and Distributed Computing, 13(4):383--394, December 1991.

....[7] abacab 107 8.0 171 1028 [7] Table 1: Static measures for some existing benchmarks. The benchmarks included in Tables 1 and 2 are found in the works of Acharya et al. 2, 1] Amaral [3, 4] Bouaud [7] Brant et al. 9] Dixit and Moldovan [8] Gupta [10] Kuo et al. 14] Kuo and Moldovan [15], Miranker and Lofaso [17, 18] Neiman [19] Oflazer [20] Schmolze [21] and Ishida [13] We now briefly state what each benchmark program does. Bench. # Prod Ant. prod # WMEs Pub. MAB 13 2.6 11 [17, 18] Mud 884 2.4 241 [17] Waltz 33 3.9 42 [17] Mesgen 155 2.9 34 [17, 18] Mapper 237 3.3 ....

....[17] Waltz 33 3.9 42 [17] Mesgen 155 2.9 34 [17, 18] Mapper 237 3. 3 1153 [17, 18] Jig25 6 50 [18] Tourney 17 123 [18] Robot 75 410 [18] Rubik 70 287 [18] weaver 637 152 [2, 9, 18] waltz 33 42 [9, 18] manners 8 [9] ARP 118 [9] Cafeteria 94 [15] Tournam. 26 [15] Toru Waltz 48 [15, 14] Hotel 723 [15] Snap 574 [15] CKT Des 107 400 [13] Table 2: Static measures for some existing benchmarks. In Tables 1 and 2, MAB, M B, mab are different implementations of the classic monkey and bananas problem, ....

[Article contains additional citation context not shown here]

S. Kuo and D. Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal of Parallel and Distributed Computing, 13:383--394, December 1991.

....of Rete appear in [3, 30, 39, 52, 80, 81, 73] 2.3 Current Research and Trends Parallel firing systems is a current research topic that The ability to fire productions in parallel introduces some new issues such as the identification of dependencies among productions. Kuo and Moldovan [50, 51] Schmolze [71] Kuo et al. 49] and Xu and Hwang [92] have worked in this area and suggested solutions to ensure the correctness of a parallel firing engine operation. The partitioning of production among processors also becomes an important issue in parallel firing systems. Oflazer [64] and Xu ....

S. Kuo and D. Moldovan, Implementation of multiple rule firing production systems on hypercube, Journal of Parallel and Distributed Computing, 13 (1991), pp. 383--394.

....implements Conway s LIFE. WALTZ 33 A constraint satisfaction problem using Waltz s algorithm for scene labeling [158] MANNERS 8 A combinatorial search problem for seat assignment. Table 1.1: Benchmark programs. to evaluate the effectiveness of language extensions and compilation techniques [80, 84, 103, 123]. While the amount of parallelism found in previous work has typically been modest and not necessarily scalable, we found, contrary to previous expectations, that several of these programs had the potential for massive and scalable parallelism. In this section, we give the results of preliminary ....

....by a simple protocol similar to the two phase locking protocol [11, 40] 2.1.2. 5 Kuo and Moldovan s Work Kuo, Moldovan, and their colleague have developed a parallel inference environment under the RUBIC project at USC for the analysis, simulation, and execution of parallel production programs [82, 81, 83, 84, 105, 106]. Two problems that must be solved by a multiple rule firing system are identified: 19 ffl Compatibility Problem To avoid interference between concurrently executing rule instantiations, a system must determine which rule instantiations are compatible, i.e. they do not interfere with each ....

[Article contains additional citation context not shown here]

Steve Kuo and Dan Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal of Parallel and Distributed Computing, 13(4):383--394, December 1991.

....This number gives an idea for the size of the database manipulated by each benchmark. The benchmarks included in Tables 1 and 2 are found in the works of Acharya et al. 2, 1] Amaral [3, 5] Bouaud [7] Brant et al. 9] Dixit an Moldovan [8] Gupta [10] Kuo et al. 11] Kuo and Moldovan [12], Miranker and Lofaso [13, 14] Neiman [15] Oflazer [16] Schmolze [17] Bench. # Prod Ant. prod Cons. prod # WME types # WMEs Pub. life 40 6.1 1.3 5 104 [3, 11] hotel 80 4.1 2.0 62 484 [3] patents 86 5.2 1.2 4 136 [3] waltz2 10 2.7 8.0 7 60 [3] R1 1932 5.6 2.9 31 [10] XSEL 1443 3.8 ....

....[13] Waltz 33 3.9 42 [13] Mesgen 155 2.9 34 [13, 14] Mapper 237 3. 3 1153 [13, 14] Jig25 6 50 [14] Tourney 17 123 [14] Robot 75 410 [14] Rubik 70 287 [14] weaver 637 152 [2, 9, 14] waltz 33 42 [9, 14] manners 8 [9] ARP 118 [9] Cafeteria 94 [12] Tournam. 26 [12] Toru Waltz 48 [12, 11] Hotel 723 [12] Snap 574 [12] Table 2: Static measures for benchmarks used. that such benchmark facilities use a standard production system language to facilitate its use by many different research institutions. In section ....

[Article contains additional citation context not shown here]

S. Kuo and D. Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal of Parallel and Distributed Computing, 13:383--394, December 1991.

.... and experimented with (Hwang et al., 1987; Kuo and Moldovan, 1992) A majority of these are based on a tightly coupled shared memory (SM) multiprocessor architecture, but a few results have also been reported for distributed memory multicomputers that use message passing (MP) Acharya et al., 1992; Kuo and Moldovan, 1991). This chapter reports an experimental implementation of a rule based image interpretation system on a parallel machine with distributed memory architecture and discusses how IU AI can be supported with such architectures in general. Data access locality and task granularity are the two major ....

....has been the dominant software strategy for implementing production rule systems. A significant portion of the research effort in this area has been devoted to parallelizing or modifying Rete, rather than considering radically different systems such as those which allow multiple rule firings (Kuo and Moldovan, 1991), or those that are written in a parallel language rather than in OPS5 (Kuo, Miranker, and Browne, 1991) The concept of Rete is based on the observation that the firing of a rule makes only incremental changes in the state of a production system, as described by the various (partial) matches of ....

Kuo, S., and Moldovan, D., 1991, "Implementation of Multiple Rule Firing Production systems on Hypercube," Journal of Parallel and Distributed Computing, vol. 13, No. 4, pp. 383-394, December 1991.

....of Naval Research under a University Research Initiative grant, number N00014 86 K 0764, NSF contract CDA 8922572, and DARPA contract N00014 89 J 1877. 1 Introduction In recent years, the parallel firing of rules has been explored as a method of increasing the performance of rule based systems [9, 10, 11, 12, 16, 17, 20, 22, 24, 27]. Much of this work has concentrated on automatically identifying sets of concurrently executable rules based on a syntactic analysis of rule sets. Our own research in this area has indicated that the automatic extraction of parallelism and run time detection of concurrently executable rules ....

Steve Kuo and Dan Moldovan. Implementation of multiple rule firing production systems on hypercube. In Proceedings of the Ninth National Conference on Artificial Intelligence, pages 305--309, 1991.

....In these systems, a single rule is selected in the conflict resolution phase, although multiple rules can be executed (or fired) without violating priorities. In the field of Artificial Intelligence, a significant amount of research has focused on parallel rule execution and rule control [30, 19, 32, 21]. Similar to our system, RUBIC [21] and PARULEL [32] provide several control constructs to control the parallel execution of rules, nevertheless, they do not guarantee the serializable execution of rules. Rule systems presented in [30, 19] guarantee serializability during the parallel execution of ....

....in the conflict resolution phase, although multiple rules can be executed (or fired) without violating priorities. In the field of Artificial Intelligence, a significant amount of research has focused on parallel rule execution and rule control [30, 19, 32, 21] Similar to our system, RUBIC [21] and PARULEL [32] provide several control constructs to control the parallel execution of rules, nevertheless, they do not guarantee the serializable execution of rules. Rule systems presented in [30, 19] guarantee serializability during the parallel execution of rules by the static analysis of ....

S. Kuo and D. Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal on Parallel and Distributed Computing, 13(4):383--394, December 1991.

....primary memory, and secondary memory, the best way of exploiting parallelism is by decomposing a task into several independent subtasks and processing them in parallel. In active systems, since multiple rules can trigger simultaneously, they can be executed in parallel on different processors [30]. Several AI researchers have worked on the execution semantics of such parallel firing systems [41, 45, 28, 30] Two important problems have been identified in such systems [40] i) controlling the execution order of rules, ii) guaranteeing the serializable execution of rules. In database ....

....a task into several independent subtasks and processing them in parallel. In active systems, since multiple rules can trigger simultaneously, they can be executed in parallel on different processors [30] Several AI researchers have worked on the execution semantics of such parallel firing systems [41, 45, 28, 30]. Two important problems have been identified in such systems [40] i) controlling the execution order of rules, ii) guaranteeing the serializable execution of rules. In database systems, the second problem can be easily solved by viewing rules as database tasks, for which the serializability is ....

[Article contains additional citation context not shown here]

Steve Kuo and Dan Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal on Parallel and Distributed Computing, 13(4):383--394, December 1991.

....scalable and broadly available parallelism. 2 Related Work Early research on parallel production systems focused almost exclusively on parallel matching [1, 5] Multiple rule firing systems parallelize not only the match phase, but also the act phase by firing multiple rules in parallel [7, 9, 12]. Some systems even fire rules asynchronously [8] Compile time syntactic analysis of data dependency graph [7] is used to detect possible interference between rules. Instantiations of compatible rules [9] can be fired in parallel. For dependencies that can not be resolved at compile time, ....

....not only the match phase, but also the act phase by firing multiple rules in parallel [7, 9, 12] Some systems even fire rules asynchronously [8] Compile time syntactic analysis of data dependency graph [7] is used to detect possible interference between rules. Instantiations of compatible rules [9] can be fired in parallel. For dependencies that can not be resolved at compile time, run time analysis is applied to increase the parallelism. All techniques above are domain insensitive since parallelism specific to the application domains is not exploited. The benefit of firing multiple rules ....

Steve Kuo and Dan Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal of Parallel and Distributed Computing, 13(4):383--394, December 1991.

....set. The rule set is observably deterministic if the execution order does not make any difference in the order of appearance of the observable actions. A different approach to response time optimization is to speed up the rule based system by parallel rule execution. For example, Kuo and Moldovan [15] propose a parallel OPS5 rule firing model. They use a notion of context, which groups the rules that interfere with each other. Several contexts can be executed in parallel if they are mutually independent. Cheng [7] proposes a similar rule firing model for EQL programs and further investigates ....

S. Kuo and D. Moldovan, "Implementation of multiple rule firing production system on hypercube," J. Parallel and Distr. Computing, vol. 13(4), pp. 383--394, Dec. 1991.

....on parallel matching [8, 32, 9, 22, 10, 14] These systems parallelized only the match phase. The speedup is therefore limited by the sequential execution of rules. Multiple rule firing systems parallelize not only the match phase, but also the act phase by firing multiple rules in parallel [13, 12, 17, 18, 28]. Some systems even fire rules asynchronously [29, 16] Compile time syntactic analysis of data dependency graph [13] is used to detect possible interference between rules. Instantiations of compatible rules [18] can be fired in parallel. For dependencies that can not be resolved at compile time, ....

....phase, but also the act phase by firing multiple rules in parallel [13, 12, 17, 18, 28] Some systems even fire rules asynchronously [29, 16] Compile time syntactic analysis of data dependency graph [13] is used to detect possible interference between rules. Instantiations of compatible rules [18] can be fired in parallel. For dependencies that can not be resolved at compile time, run time analysis is applied to increase the parallelism. The copy and constraint (C C) technique proposed by Pasik [27] proved to be quite effective in reducing the variance in rule processing time and improving ....

[Article contains additional citation context not shown here]

Steve Kuo and Dan Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal of Parallel and Distributed Computing, 13(4):383--394, December 1991.

....for the PARULEL version of Toru Waltz varied between 6.75 to 12.7 for the figures tested. As in Waltz, this average as well as the average number of RHS actions per cycle is more dependent on the shape of the input figure than on the size of the input. In another recent work, Kuo and Moldovan [3] reported that their system could speedup (in terms of cycles) ToruWaltz by a factor of 3.18. Averaging the number of cycles reported in tables 4 and 5 we found that our implementation of Toru Waltz averaged a speed up of 11.2 over the serial implementation. Finally, the EIGs were almost the same ....

S. Kuo and D. Moldovan. Implementation of Multiple Rule Firing Production Systems on Hypercube. Journal of Parallel and Distributed Computing, 13(4):383--394, 1991.

....almost exclusively on parallel matching [10, 4] These systems parallelized only the match phase. The speedup is therefore limited by the sequential execution of rules. Multiple rule firing systems parallelize not only the match phase, but also the act phase by firing multiple rules in parallel [6, 8, 13]. Some systems even fire rules asynchronously [12, 7] Compiletime syntactic analysis of data dependency graph [6] is used to detect possible interference between rules. Instantiations of compatible rules [8] can be fired in parallel. For dependencies that can not be resolved at compile time, ....

....only the match phase, but also the act phase by firing multiple rules in parallel [6, 8, 13] Some systems even fire rules asynchronously [12, 7] Compiletime syntactic analysis of data dependency graph [6] is used to detect possible interference between rules. Instantiations of compatible rules [8] can be fired in parallel. For dependencies that can not be resolved at compile time, run time analysis is applied to increase the parallelism. All techniques above are domain insensitive since parallelism specific to the application domains is not exploited. The benefit of firing multiple rules ....

[Article contains additional citation context not shown here]

S. Kuo and D. Moldovan. Implementation of multiple rule firing production systems on hypercube. Journal of Parallel and Distributed Computing, 13(4):383--394, Dec. 1991.

No context found.

Kuo S. and Moldovan D. "Implementation of Multiple Rule Firing Production Systems on Hypercube". Proc. in Ninth National Conference on AI. AAAI-91 pp. 310-316. 1991.

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