Interconnection Networks
- PDF / 3,781,333 Bytes
- 94 Pages / 565.087 x 755.008 pts Page_size
- 64 Downloads / 202 Views
Synonyms Blue Gene/L; Blue Gene/P; Blue Gene/Q
Definition The IBM Blue Gene Supercomputer is a massively parallel system based on the PowerPC processor. A Blue Gene/L system at Lawrence Livermore National Laboratory held the number position in the TOP list of fastest computers in the world from November to November .
Discussion Introduction The IBM Blue Gene Supercomputer is a massively parallel system based on the PowerPC processor. It derives its computing power from scalability and energy efficiency. Each computing node of Blue Gene is optimized to achieve high computational rate per unit of power and to operate with other nodes in parallel. This approach results in a system that can scale to very large sizes and deliver substantial aggregate performance. Most large parallel systems in the – time frame followed a model of using off-the-shelf processors (typically from Intel, AMD, or IBM) and interconnecting them with either an industry standard network (e.g., Ethernet or Infiniband) or a proprietary network (e.g., as used by Cray or IBM). Blue Gene took a different approach by designing a dedicated system-on-achip (SoC) that included not only processors optimized for floating-point computing, but also the networking infrastructure to interconnect these building blocks into
a large system. This customized approach led to the scalability and power efficiency characteristics that differentiated Blue Gene from other machines that existed at the time of its commercial introduction in . As of , IBM has produced two commercial versions of Blue Gene, Blue Gene/L and Blue Gene/P, which were first delivered to customers in and , respectively. A third version, Blue Gene/Q, was under development. Both delivered versions follow the same design principles, the same system architecture and the same software architecture. They differ on the specifics of the basic SoC that serves as the building block for Blue Gene. The November TOP list includes four Blue Gene/L system and ten Blue Gene/P systems (and one prototype Blue Gene/Q system). In this article we cover mostly the common aspects of both versions of the Blue Gene supercomputing and discuss details specific to each version as appropriate.
System Architecture A Blue Gene system consists of a compute section, a file server section, and a host section (Fig. ). The compute and I/O nodes in the compute section form the computational core of Blue Gene. User jobs run in the compute nodes, while the I/O nodes connect the compute section to the file servers and front-end nodes through an Ethernet network. The file server section consists of a set of file servers. The host section consists of a service node and one or more front-end nodes. The service node controls the compute section through an Ethernet control network. The front-end nodes provide job compilation, job submisson, and job debugging services.
Compute Section The compute section of Blue Gene is what is usually called a Blue Gene machine. It consists of a threedimension
Data Loading...
