×

You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Title
Proceedings of the Symposium on Progress in Data Processing and Analysis

292
Beside this dedicated processors the problem with multiple
processors on bus coupled systems is, that adding more
and more processors on the same bus, does not linear
increase the performance. From a number of 4 to 6
processors on the same bus the performance decreases.
This can be avoided by using bus-less processors like
transputers. The currently available INMOS T800 chips (see
Fig. 3) give a performance of 1.5 Mflops/10 Mips each.
Figure 3: The INMOS T800 Transputer Chip
They can be configured in any number and topology being
appropriate for the application. Communication is done via
so called links, giving a net point to point transfer rate of
1.8 Mbyte/sec for each link. Arrays of transputers forming
so called computing clusters can be build in topologies by
using software controllable network configuration units.
They may be used either as an add-on's on standard
graphic workstations or as stand alone systems with 64
up to 256 transputers. A configuration of 64 T800 super
computing cluster has a performance of 640 Mips and 96
Mflops at a price of approximately 250.000 $. Parellel
compilers of nearly all high level computer languages, like
C, FORTRAN, PASCAL and PROLOG are available as a
variety of software development tools. Today programming
of such arrays seems to be rather a tedious job in
comparision to programming standard processors. In addition
the latest processor developments like CISC (MC 68040)
show a performance that comes close to the requirements
in this application area.