Blue Gene/Q: by co-design

摘要

Blue Gene/Q represents the third generation in the IBM Blue Gene line of supercomputer systems. Architecturally, BG/Q installations can scale to over 100 PetaFLOPS. Blue Gene/Q follows the same philosophy as the earlier Blue Gene/L and Blue Gene/P systems, namely to build a massively parallel and highly reliable High Perfor-mance Computing (HPC) system out of power-efficient processor chips. Such power-efficient chips, in turn, allow dense packaging, which consequently results in power efficient, cost efficient, and floor-space efficient systems. A focus on reliability in the design of each component contributes to the ability to run a single application across even the largest configurations. Because of the tight interaction between the hardware, system software, and application teams, the machine was truly co-designed. The intended characteristics of architecture were presented early to the system software and application teams, and those teams requested specific features that were then incorporated into the hardware. The contributions of this paper are a description of the overall system and packaging of Blue Gene/Q, a description of the core system software, and a detailed description of the software's needs that drove the co-design process, in turn affecting the hardware.