Multiprocessing System Architectures - Analog Devices ADSP-2106x SHARC User Manual

7 Multiprocessing
7.2

MULTIPROCESSING SYSTEM ARCHITECTURES

Multiprocessor systems typically use one of two schemes to
communicate between processor nodes. One scheme uses dedicated
point-to-point communication channels. In the other, nodes
communicate through a single shared global memory via a parallel
bus.
The ADSP-2106x SHARC supports the implementation of point-to-
point communication through its six link ports. It also supports an
enhanced version of shared parallel bus communication called cluster
multiprocessing. Cluster multiprocessing features of the ADSP-2106x
are described in this chapter, while point-to-point connections are
described in the Link Ports chapter of this manual.
Multiprocessing systems must overcome two problems: interprocessor
communication overhead and data bandwidth bottlenecks. The ADSP-
2106x SHARC architecture addresses these concerns in several ways,
as illustrated in the following discussion of three basic multiprocessing
topologies.
7.2.1 Data Flow Multiprocessing
Data flow multiprocessing is best suited for applications requiring
high computational bandwidth but only limited flexibility.
Programmers partition their algorithm sequentially across multiple
processors and pass data linearly down an "assembly line" of
processors, as shown in Figure 7.2.
ADSP-2106x
Data Link
Port
Figure 7.2 Data Flow Multiprocessing
7 – 4
www.BDTIC.com/ADI
ADSP-2106x
Link Link Link
Port Port Port
ADSP-2106x
Link Link
Port Port