The authors propose a parallel machine, called HOBONET, in which the PUs are connected with a nearest neighbor mesh (NNM) connection structure. HOBONET consists of a set of switching modules (SwMs) for the purpose of interconnecting the PUs. With the detection of faulty PUs, connections of SwMs are changed so that faulty PUs are disconnected, but the NNM connection structure is maintained logically. The scheme is called lattice defect recovery.
|ホスト出版物のタイトル||Proceedings of the International Conference on Parallel Processing|
|編集者||Robert M. Keller|
|出版ステータス||Published - 1984 12 1|
|名前||Proceedings of the International Conference on Parallel Processing|
ASJC Scopus subject areas