johnz@basser.oz (John Zic) (03/05/87)
I'm looking for people's definitions of the degree of coupling between
processors. Here are some that I think aren't unreasonable:
* a measure of the interprocess channel bandwidth utilisation
* closely coupled processors tend to be spatially close together, for
example, on a bus
* a measure of the amount of buffering present in the network between
processing elements
* the degree of overhead in sending data from one processor to another
I personally feel that the last two are the "best" definitions. However,
any comments on these definitions, or others, are certainly welcome.
--
John Zic
ACSnet: johnz@basser.oz UUCP: seismo!munnari!basser.oz!johnzdarrell@sdcsvax.UUCP (03/07/87)
I've also heard the common refresh cycle for memory as tight coupling, common address space (Cm* as a tightly coupled machine), and others. Personally, I don't know. There seem some relativistic Heisenbergs out there if we make things too precise. --eugene
darrell@sdcsvax.UUCP (03/09/87)
> I'm looking for people's definitions of the degree of coupling between > processors. Here are some that I think aren't unreasonable: > > * the degree of overhead in sending data from one processor to another > perhaps more significant than the bandwidth of an interprocessor connection is the latency -- that is, the minimum time for end-to-end communication. Many 10Mbit/sec links require several thousand instruction times to fire up a packet.