Multi-CMP module system based on a look-ahead configured global network

Authors:
Eryk Laskowski;Łukasz Maśko;Marek Tudruj
Affiliations:
Institute of Computer Science, PAS, Warsaw, Poland;Institute of Computer Science, PAS, Warsaw, Poland;Institute of Computer Science, PAS, Warsaw, Poland and Polish-Japanese Institute of Information Technology, Warsaw, Poland
Venue:
PPAM'09 Proceedings of the 8th international conference on Parallel processing and applied mathematics: Part I
Year:
2009

Citing 5
Cited 0

Towards Massively Parallel Numerical Computations Based on Dynamic SMP Clusters with Communication on the Fly

ISPDC '05 Proceedings of the The 4th International Symposium on Parallel and Distributed Computing
Efficient Parallel Embedded Computing through Look-Ahead Configured Dynamic Inter-Processor Connections

PARELEC '06 Proceedings of the international symposium on Parallel Computing in Electrical Engineering
Research Challenges for On-Chip Interconnection Networks

IEEE Micro
A view of the parallel computing landscape

Communications of the ACM - A View of Parallel Computing
Scheduling moldable tasks for dynamic SMP clusters in soc technology

PPAM'05 Proceedings of the 6th international conference on Parallel Processing and Applied Mathematics

Quantified Score

Hi-index	0.00

Visualization

Abstract

The impressive progress of Systems on Chip (SoC) design enables a revival of efficient massively parallel systems based on many Chip Multiprocessor (CMP) modules interconnected by global networks. The paper presents methods for the optimized program execution control for such modular CMP systems. At the CMP module level, communication through shared memory is applied, improved by a novel efficient group communication mechanism (reads on the fly). The inter-module global communication is implemented as a message passing between module memories placed in a shared address space. A two-phase structuring algorithm is described for programs represented as macro data-flow graphs. In the first phase, program tasks inside the CMP modules are scheduled, using an algorithm based on the notion of moldable tasks. In the next phase, the moldable task graph is structured for optimized communication execution in the global interconnection network according to the look-ahead link connection setting paradigm. Simulation experiments evaluate the efficiency and other properties of the proposed architectural solutions.