Processor Configurations

Processors can be interconnected in a number of ways, see Figures 31 and 32:

  
Figure 31: A 2D Processor Grid

  
Figure 32: A 1D Processor Chain

Data is distributed (or `overlaid') onto the processor grids.

A more complex configuration is a hypercube. This is similar to a 2x2x2 cube except it is constructed in much more than 3 dimensions.

There are two main types of parallel computers

Distributed Memory Machines

-- Each processor has its own local memory, no other processor can access this storage. Examples include the Cray T3E and Intel Paragon.

Shared Memory Machines

-- All processors use the same (shared) memory. These machines are also known as SMPs (symmetric multi-processors), examples are the Silicon Graphics Power Challenge or Sun SPARCcenter 2000.

The former is more scalable than the latter meaning that DMMs can support more processing elements (processors).

There are also Distributed Shared Memory machines such as the IBM SP-2 and NUMA (Non-Uniform Memory Architecture) such as the KSR-1 which can loosely be described as hybrids of the main two types.

Return to corresponding overview page