|
|
Company > Research
Optimal Non-blocking Switches
New cost-efficient designs and routing algorithms -
the theoretical results provide the first major breakthrough in this
area since the pioneering works of C. Clos (1953) and V. Benes
(1962) of Bell Labs. The proposed designs are two-sided multistage
interconnection network (MIN) built of switching elements of desirable
sizes. They are optimally compact and scalable to virtually any size
N×M (i.e. having N input and M output ports).
Each design is supplied with a highly efficient routing algorithm. The
hardware saving is approximately 30-60% compared to best-known
conventional designs of the same functionality. The saving is especially
great for odd, asymmetric and large switch sizes. It is achieved by
utilizing less number of switching elements, possibly reducing their
size and, consequently, reducing the total cost of switching elements
and the number of inter-stage links. The great hardware saving would
also affect the physical size of the switch, its reliability, and
provide more possibilities for higher security and fault tolerance
features. Moreover, the proposed designs minimize the number of stages,
which is critical for some application with power loss issues. In
addition to that the switches allow modular construction and can have
various additional non-blocking properties such as multi-casting and
loop-backs. The new designs can find variety of applications in ever
expanding telephony and computer networks, DSL, photonic switching,
parallel processing, etc.
Functionally, the proposed 2-sided non-blocking switching networks can
be divided into the following 4 categories:
-
Conventional Rearrangeable (CR)
-
Non-interruptive rearrangeable (NIR)
-
Wide-sense non-blocking (WNB)
-
Virtually non-blocking (VNB)
The major common feature of a 2-sided
non-blocking network is that it can simultaneously connect all its
source input ports (inputs) to its destination ports (outputs)
in a one-to-one manner for all possible permutations of input-to-output
pairings. For scheduled traffic there is no essential difference between
the above categories. However, they differ in network ability to handle
dynamic traffic. CR networks are also called rearrangeably
non-blocking networks, and NIR networks were initially introduced as
hitlessly rearrangeable networks. In what follows we
overview the main properties and application of each category of the
proposed networks.
|