A. J. Pasik, "A source-to-source transformation for increasing rule-based system parallelism," IEEE Trans. of Knowledge and Data Engineering, vol. 4, no 4, August, pp. 336-343, 1992.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of rule programs is not new. In fact, the problem comes up almost immediately in any large scale effort [4] Following is a taxonomy of solutions. Structural Protocols: Within a flat monolithic rule language, such as OPS5, specific protocols concerning the structure of rules have been suggested [4,75]. Rule Groups: In a number of systems rules can be grouped together into rule groups . The flow of control among the groups is performed by explicit procedural invocation [38,45,53] In [7] a method is proposed in which rule developers must establish metrics to determine the stratum ....

A. J. Pasik, "A source-to-source transformation for increasing rule-based system parallelism," IEEE Trans. of Knowledge and Data Engineering, vol. 4, no 4, August, pp. 336-343, 1992.

....only with a conflict resolution strategy which has no constraint on the firing order of instantiations, but not with the sequential conflict resolution strategies such as LEX and MEA for the OPS5 [3] language and CLIPS [12] language in which the instantiations have to be fired in some order. In [9] and [10] more production system parallelism is extracted by creating constrained copies of culprit rules and distributing them to different processors for parallel processing. Because the condition elements of culprit rules require comparisons with many more working memory elements than those of ....

A. Pasik, "A Source-to-Source Transformation for Increasing Rule-Based System Parallelism," IEEE Trans. on Knowledge and Data Eng., 1992, Vol. 4, No. 4, pp. 336--342.

.... tree of objects of a specified depth (path length) and breadth (branching factor) was created; a path length of n means that n 1 objects are 13 RETE is the most commonly used incremental match algorithm in production systems [22] Although other match algorithms have appeared in the literature [23, 18, 22], we could not find any public domain versions of OPS5 that used these algorithms. Thus, we have been forced to limit our comparisons to RETE. 24 path branching # of R rate R rate C5 rate length factor leaves for sets for lists 0 NA 1 120000 120000 2083 1 1 1 25575 10591 1428 1 2 2 23831 ....

....relative to pattern matching rules and to C on a wellknown line labeling application program is shown in Table 2. This benchmark program labels line drawings using Waltz s constraint propagation algorithm [24] and has previously been used to evaluate matching algorithms for OPS5 style systems [23, 18, 25, 22]. The C5 version of the benchmark is a slightly modified version of a simple OPS5 program obtained from Daniel P. Miranker. 14 Because OPS5 and R have different rule languages, the benchmark had to be recoded from OPS5 to R . The R version was derived from the OPS5 program by first ....

Alexander J. Pasik. A source-to-source transformation for increasing rule-based system parallelism. IEEE Transactions on Knowledge and Data Engineering, 4(4):336--343, 1992.

....and C , by coding two well known examples (a benchmark expert system program and a standard active database example) in all three languages. 13 RETE is the most commonly used incremental match algorithm in production systems [23] Although other match algorithms have appeared in the literature [24, 19, 23], we could not find any public domain versions of OPS5 that used these algorithms. Thus, we have been forced to limit our comparisons to RETE. 16 path branching # of R rate R rate C5 rate length factor leaves for sets for lists 0 NA 1 120000 120000 2083 1 1 1 25575 10591 1428 1 2 2 23831 ....

....relative to pattern matching rules and to C on a well known line labeling application program is shown in Table 2. This benchmark program labels line drawings using Waltz s constraint propagation algorithm [25] and has previously been used to evaluate matching algorithms for OPS5 style systems [24, 19, 26, 23]. input C5 R C size Standard Sets Hash Sets Standard Sets Hash Sets secs MB secs MB secs MB secs MB secs MB 12 166.0 1.0 5.0 1.3 1.0 0.7 3.4 0.7 0.6 0.5 25 737.0 1.7 15.0 2.1 2.5 1.1 13.1 0.9 1.6 0.7 37 1839.0 2.3 30.0 2.8 4.2 1.4 27.4 1.2 3.1 0.8 50 3752.0 2.9 53.0 3.6 6.4 1.6 49.1 1.5 ....

Alexander J. Pasik, "A source-to-source transformation for increasing rule-based system parallelism," IEEE Transactions on Knowledge and Data Engineering, vol. 4, no. 4, pp. 336--343, 1992. 35

....programs and writing parallel programs from scratch. Another direction of our future research is to formally compare the programmability, complexity, and effectiveness of different approaches including static transformation [101] metarule programming [143] source to source transformation [119], rule rewriting [46] and decomposition abstraction. Chapter 7 Performance Assessment With decomposition abstraction, programmers can specify the parallelism inherent in the problem domain, thereby increasing the concurrency that can be exploited by the language system. However, just like any ....

....and run time interference detection. ffl Based on working technologies and existing sequential implementation. This includes the use of Lazy Match algorithm [99, 100] and the Venus implementation. ffl Static work load distribution and prescheduling by copy and constrain (C C) [118, 119, 141]. ffl The nondeterministic safe assumption, i.e. firing rules nondeterministically should not affect the correctness of the program execution. In other words, the input source program should not depend on rule priority or any implicit conflict resolution strategy for its correctness. 1 By using ....

Alexander J. Pasik. A source-to-source transformation for increasing rule-based system parallelism. IEEE Trans. on Knowledge and Data Engineering, 4(4):336--343, August 1992.

....variables within the scope of the present module. Modifications in the RHS are automatically recognized by the Venus parser, and thus, no special modification keyword is needed. Operations on containers are made using the container interface which includes insertion to and deletion from containers [20]. 3.2 Modularity and Semantics A Venus module fires rule instantiations until a fixed point is reached[5] when no rules are satisfied, or the firing of a rule does not produce a state change. The entire RHS action is treated as a single atomic action. Rules have an optional user specified ....

A. J. Pasik. A Source-to-Source Transformation for Increasing Rule-Based System Parallelism, IEEE Trans. of Knowledge and Data Engineering, 4(4), August 1992

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