The IBM Blue Gene/P system (BG/P) is the successor of the Blue Gene/L system. In this new version, the Blue Gene uses quad-processor symmetric multiprocessing (SMP) nodes for both computing and input/output (I/O) activities. The BG/P also uses a three-dimensional (3D) torus network enhanced with direct memory access (DMA) to connect the computing nodes. Each processor is a PowerPC 450 that supports two simultaneous floating-point operations. There are a lot of other enhancements over the previous BG.
This paper relates the main components of the BG/P system, briefly discussing architecture and memory organization, energy consumption, the operating system (OS), applications, programming models, scaling, and performance. It is a very interesting work, promoting the IBM Blue Gene/P system.
However, the authors should also illuminate the weaknesses of the BG/P. They should show more of the real benefits of this platform, beyond just speedup. The use of 262,144 quad-processor nodes cannot be justified as the only comparison between new and old systems.