Abstract: "KL1, a committed choice language based on Flat GHC, was implemented on a distributed memory multi-processor, the Multi-PSI, which has up to 64 processing elements (PEs) connected by a message passing network. The key issues for a distributed implementation of KL1 are: (1) how to redu...

Full description

Bibliographic Details

Main Author:	Nakajima, Katsuto (Author)
Format:	Book
Language:	English
Published:	Tokyo, Japan 1991
Series:	Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report 715
Subjects:	KL1 (Computer program language) Multiprocessors
Summary:	Abstract: "KL1, a committed choice language based on Flat GHC, was implemented on a distributed memory multi-processor, the Multi-PSI, which has up to 64 processing elements (PEs) connected by a message passing network. The key issues for a distributed implementation of KL1 are: (1) how to reduce the amount of inter-PE communication, (2) how to achieve efficient intra-PE and inter-PE garbage collection, and (3) how to avoid making redundant copies of data objects over many processors. The well- defined semantics of KL1 allows incremental intra-PE garbage collection by the Multiple Reference Bit (MRB) technique and incremental inter-PE garbage collection by the Weighted Export Counting (WEC) technique A global structure management mechanism is also introduced to avoid making duplicate copies of large data objects such as program codes. The communication required for inter-PE process control is minimized by the Weighted Throw Counting (WTC) scheme. The implementation has been completed and the performance of inter-PE communication was evaluated. Although the cost for inter-PE message handling is high, the effective overhead in the benchmark programs remains within 20% on 64 PEs.
Physical Description:	24 S.