Difference between revisions of "Multi-processor"
| m (Avoid redir) |  (Clarify) | ||
| Line 1: | Line 1: | ||
| − | A '''multi-processor''' is a system with more than one [[Central Processing Unit|CPU]]. There are a tremendous range of designs, from 'tightly-coupled' (where they share access to some [[main memory]]) onward. Another axis for dividing them is among: | + | A '''multi-processor''' is a system with more than one [[Central Processing Unit|CPU]]. There are a tremendous range of designs, from 'tightly-coupled' (where they share access to some [[main memory]], which usually has to be [[multi-port memory]]), onward. | 
| + | |||
| + | Another axis for dividing them is among: | ||
| * MIMD - 'multiple instruction streams, multiple data sets' | * MIMD - 'multiple instruction streams, multiple data sets' | ||
| Line 6: | Line 8: | ||
| In the latter, each CPU is fed different data, but they all execute the same [[instruction]]s in the same order. | In the latter, each CPU is fed different data, but they all execute the same [[instruction]]s in the same order. | ||
| − | Among tightly-coupled systems, one can distinguish between 'symmetric' and 'asymmetric' systems | + | Among tightly-coupled systems, one can distinguish between 'symmetric' and 'asymmetric' systems. In the former, all the CPUs are 'equal', and no CPU has a special, distinguished role (so often any CPU can e.g. field [[interrupt]]s from any [[peripheral]]). The latter are often so-called 'master/slave' systems, with one CPU in the first class, and one or more in the second. | 
| − | With the advent of very-high-speed [[local area network|LANs]], it is now common to create multi-processors by simply linking  | + | With the advent of very-high-speed [[local area network|LANs]], it is now common to create multi-processors by simply linking together a number of ordinary systems by use of such a network. | 
| {{stub}} | {{stub}} | ||
Revision as of 19:19, 19 August 2018
A multi-processor is a system with more than one CPU. There are a tremendous range of designs, from 'tightly-coupled' (where they share access to some main memory, which usually has to be multi-port memory), onward.
Another axis for dividing them is among:
- MIMD - 'multiple instruction streams, multiple data sets'
- SIMD - 'single instruction streams, multiple data sets'
In the latter, each CPU is fed different data, but they all execute the same instructions in the same order.
Among tightly-coupled systems, one can distinguish between 'symmetric' and 'asymmetric' systems. In the former, all the CPUs are 'equal', and no CPU has a special, distinguished role (so often any CPU can e.g. field interrupts from any peripheral). The latter are often so-called 'master/slave' systems, with one CPU in the first class, and one or more in the second.
With the advent of very-high-speed LANs, it is now common to create multi-processors by simply linking together a number of ordinary systems by use of such a network.

