AND Alan E. Klietz* and Stephen Saroff* Minnesota Supercomputer Center, 1200 Washington Avenue South, Minneapolis, Minnesota 55415
The Network Architecture of the Connection Machine CM-5
โ Scribed by Charles E. Leiserson; Zahi S. Abuhamdeh; David C. Douglas; Carl R. Feynman; Mahesh N. Ganmukhi; Jeffrey V. Hill; W.Daniel Hillis; Bradley C. Kuszmaul; Margaret A. St. Pierre; David S. Wells; Monica C. Wong-Chan; Shaw-Wen Yang; Robert Zak
- Publisher
- Elsevier Science
- Year
- 1996
- Tongue
- English
- Weight
- 341 KB
- Volume
- 33
- Category
- Article
- ISSN
- 0743-7315
No coin nor oath required. For personal study only.
โฆ Synopsis
combines the best aspects of SIMD (single instruction stream, multiple data stream) and MIMD (multiple instruction stream, multiple data stream) machines . Each processor in the CM-5 executes its own instructions, providing the flexibility of a typical MIMD machine. And, like many MIMD machines, the CM-5 is a distributed memory machine (as opposed to shared memory machine ) in which processors communicate among themselves by sending messages through the data network of the machine. A deficiency of typical MIMD machines, especially as compared with their SIMD cousins, however, is that they provide little or no support for coordinating and synchronizing sets of processors. To offset this deficiency, the CM-5 also contains a control network, which makes synchronization and multiparty communication primitives competitive with comparable functions on SIMD machines. These primitives include the fast broadcasting of data, barrier synchronization , and parallel prefix (scan) operations .
Figure shows a diagram of the CM-5 organization. The machine contains between 32 and 16,384 processing nodes, each of which contains a 32-MHz SPARC processor, 32 megabytes of memory, and a 128-megaflops vector-processing unit capable of processing 64-bit floating-point and integer numbers. System administration tasks and serial user tasks are executed by a collection of control processors, which are Sun Microsystems workstation computers. There are from one to several tens of control processors in a CM-5, each configured with memory and disk according to the customer's preference. Input and output is provided via high-bandwidth I/O interfaces to graphics devices, mass secondary storage, and high-performance networks. Additional low-speed I/O is provided by Ethernet connections to the control processors. The largest machine, configured with up to 16,384 processing nodes, occupies a space of approximately 30 by 30 meters and is capable of over a teraflops (10 12 floating-point operations per second).
The processing nodes, control processors, and I/O interfaces are interconnected by three networks: a data network, a control network, and a diagnostic network. The data network provides high-performance point-to-point data communications between system components. The
๐ SIMILAR VOLUMES
Massively parallel architectures such as the Connection Machine can process large knowledge bases efficiently. An appropriate knowledge representation scheme for parallel processing is a semantic network. We have defined and implemented software data structures, marker propagation rules, and an inst