Conference paper
Sp2 high-performance switch architecture
Abstract
•Scalable unified data & service interconnection network - SP2 offered for 4 - 128 nodes - Up to 512 nodes available upon request • Maintains synchronized global clocks • Buffered wormhole routing • Communication co-processor offloads protmol, increases node bandwidth • Source-based oblivious routing • Highly fault tolerant.
Related
Conference paper
MPI-F: An efficient implementation of MPI on IBM-SP1
Conference paper
MPI on IBM SP1/SP2: Current status and future directions
Conference paper