2024-03-19T11:00:10Zhttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_oaipmhoai:ipsj.ixsq.nii.ac.jp:000599532017-03-31T05:36:57Z05471:05525:05529
Procedure Level Data Flow Processing on Dynamic Structure MultimicroprocessorsProcedure Level Data Flow Processing on Dynamic Structure Multimicroprocessorsenghttp://id.nii.ac.jp/1001/00059953/Articlehttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_action_common_download&item_id=59953&item_no=1&attribute_id=1&file_no=1Copyright (c) 1982 by the Information Processing Society of JapanFaculty of Engineering University of TokyoFaculty of Engineering University of TokyoFaculty of Engineering University of TokyoFaculty of Engineering University of TokyoTatsuo, SuzukiKen, KuriharaHidehiko, TanakaTohruMoto-OkaTOPSTAR a highly parallel multi-microprocessor system is designed and constructed. It executes Procedure level data flow processing (one data flow node is one procedure). TOPSTAR is composed of two kinds of modules (CM's and PM's). Each module is a conventional micro computer system. The connective structure of TOPSTAR is a locally connected bipartite graph. The control right belongs to each CM (which keeps unenabled data) and distributed control mechanism is attained. The load balancing problem is solved by the free competition rule. Five kinds of basic nodes are prepared for programming in the data flow graph language. Most Petri nets involved in the Free-Choice Petri net subclass can be realized by using these basic nodes. System software is implemented. The overflow in data buffers is avoided by introducing inhibitors (of the Petri net). Plural data buffers are attached to each node and tokens can get ahead of others to provide more pipeline parallelism. Several program structures (such as conditional branch loop and recursion) are available. TOPSTAR-II consisting of 24 microprocessors is constructed and working with some applications. They are parallel mergesort simulation of logical circuits some arithmetic calculation etc. Actually measured result in the case of parallel merge-sort are also reported.TOPSTAR, a highly parallel multi-microprocessor system is designed and constructed. It executes Procedure level data flow processing (one data flow node is one procedure). TOPSTAR is composed of two kinds of modules (CM's and PM's). Each module is a conventional micro computer system. The connective structure of TOPSTAR is a locally connected bipartite graph. The control right belongs to each CM (which keeps unenabled data), and distributed control mechanism is attained. The load balancing problem is solved by the free competition rule. Five kinds of basic nodes are prepared for programming in the data flow graph language. Most Petri nets, involved in the Free-Choice Petri net subclass, can be realized by using these basic nodes. System software is implemented. The overflow in data buffers is avoided by introducing inhibitors (of the Petri net). Plural data buffers are attached to each node, and tokens can get ahead of others to provide more pipeline parallelism. Several program structures (such as conditional branch, loop and recursion) are available. TOPSTAR-II consisting of 24 microprocessors is constructed and working with some applications. They are parallel mergesort, simulation of logical circuits, some arithmetic calculation, etc. Actually measured result in the case of parallel merge-sort are also reported.AA00700121Journal of Information Processing 5111161982-03-311882-66522009-06-30