RT-cube: An interconnection network architecture for multiprocessor systems
We are at the threshold of designing fast and parallel computers, due to advancements in hardware and software technologies. Recently the processor and I/O technology has spurred many new avenues in designing very fast multiprocessor systems. The effect of these advances has put even more burden on communication networks to rout faster. In order to build efficient parallel computer architecture, we must have all its components working at coherent rates. It turns out that communication architecture has the most burden of catching up with the rest of the machine. This is due to many factors, such as: (1) large diameter of the network, (2) large inter node distances, (3) slow communication channels, (4) small communication bandwidth, (5) large channel latencies, (6) sharing of wires and channels by many resources resulting in contentions, (7) slow routing algorithm, (8) inadequate message buffering space, (9) utilizing processor's local memory for message buffering, (10) making the processor do the routing tasks along with its computing responsibilities, (11) inefficient transmission protocols, (12) and inefficient handling of the load. All these and many other factors work against the communication architecture. In order to design an efficient, fast, and reliable interconnection network for a multiprocessor system, we must eliminate all these bottlenecks from the system. The topology is independent of the technology used to implement the communication architecture in any multiprocessor computer system. The most important characteristics of a topology are: (1) diameter and latencies, i.e., how fast it allow to communicate between any two nodes; (2) scalable, i.e., how large the system can grow up to and at what cost; (3) routing algorithm, a simple algorithm is desired as it will be easy to implement. Hypercube has been extensively used in many multiprocessor systems due to their moderate diameter and scalability.
- Research Organization:
- State Univ. of New York, Binghamton, NY (United States)
- OSTI ID:
- 7260681
- Resource Relation:
- Other Information: Thesis (Ph.D.)
- Country of Publication:
- United States
- Language:
- English
Similar Records
Communications-processor architecture for point-to-point multiprocessor networks
An efficient communication scheme for solving the S{sub n} equations on message-passing multiprocessors