Next | Prev | Up | Top | Contents | Index
Parallel Hardware and Programming Models
Silicon Graphics makes a variety of multiprocessor systems, including
- The CHALLENGE/Onyx systems (and their POWER versions) are symmetric multiprocessor (SMP) computers. In these systems at least 2, and as many as 36, identical microprocessors access a single, common memory and a common set of peripherals through a high-speed bus.
- The POWER CHALLENGEarray(TM) comprises 2 or more POWER CHALLENGE(TM) systems connected by a high-speed local HIPPI network. Each node in the array is an SMP with 2 to 36 CPUs. Nodes do not share a common memory; communication between programs in different nodes passes through sockets. However, the entire array can be administered and programmed as a single entity.
Most programs that run on these systems execute as if they were in a uniprocessor, employing the facilities of a single CPU. The IRIX operating system applies CPUs to different programs in order to maximize system throughput.
However, it is possible to write a program so that it makes use of more than one CPU at a time. The software interface you use for this is the parallel programming model. Each model is designed around a different set of assumptions about the hardware, and in particular about the memory system available.
- Parallel Programs on Uniprocessors
-
- Memory Systems
-
- Types of Parallel Models
-
Next | Prev | Up | Top | Contents | Index