3. PREAMBLE TO THE DESCRIPTION
Ring Connected Binary Tree (RiCoBiT) is a structured Network on Chip (NoC)
topology that has nodes in the structure connected in the shape of a binary tree with
Ring connections at each level for ‘he adjacent nodes. Parallel processing in Ring
connected binary tree can be achieved by dividing the computation into sub
processing modules and distributing them to the other computation nodes. Ring
connected Binary tree structure is one of the new topologies in Network on chip
structures and would need a unique mechanism for distribution, A Ring Connected
Binary Tree would have one node as a Controller node and the remaining nodes
would be the computation nodes, where the parallel processing happens. The
controller node in the Ring Connected Binary Tree would define the processing
operands and its respective processing modules, that are used in the computation.
The novel design described here is the design of a module which distributes data with
low latency and optimal power consumption, in Ring Connected Binary Tree
structures. This work designs a new module that possess a method to reach the nodes
based on the hop count from the controller node to distribute the data for complex
processing to other nodes. The distribution module designed in the controller node of
the Ring Connected Binary Tree Architecture works with a multi-level quadrant based distribution mechanism, which was never present in Ring connected binary tree type of architecture.
4. DESCRIPTION:
We propose the design of a novel Quadrant based data distribution module within the
controller node of a Ring Connected Binary Tree architecture, where there are no
existing distribufion mechanisms defined for complex data parallel processing in
RjCoBiT architecture. The distribution module is a multi—level quadrant-based
distribution mechanism, where there are level locators and dynamic node span
detectors with quadrant mappers. All these modules would help in identifying the
nodes based on the hop count to cover for each cycle of distribution. The operation of
this distribution module is scalable and so it could be applied for any number of rings
in this structure. Since it is possible to identify the nodes having less hops from
controller node, the latency from controller node can be reduced during distribution
and also it in turn reduces the power consumption.
The Figure 1 represents the design of the distribution module of RiCoBiT Controller
Node where the parallel processing of complex data is achieved by dividing and
distributing the processing to all the computation nodes of the RiCoBiT structure.
The distribution would identify the levels with the help of the Level Locator, then
identifies the range of nodes using the quadrant mapping method. Once the range of
nodes are identified, the processing would then be distributed to those nodes with the
help of a dispatcher. The total number of nodes per hop count and the method on how
the nodes are distributed across different levels are determined, to identify the range
of nodes.
5. CLAIMS: -
WE CLAIM,
1. Design of a novel Quadrant based distribution module inside a Controller node
in RiCoBiT topology Network on Chip architecture to perform parallel
processing by dislIibution of complex processing to all other nodes in the
structure.
2. Design of a new Dynamic span detector which uses a quadrant mapper for
identifying the address range of nodes in each ring in a RiCoBiT structure that
helps in optimal power consumption and reduced latency by distributing the
data to nodes having less hop count from controller node.
3. Design of a new Level Locator that identifies the relevant levels in which the
nodes reachable for a specific hop count are located.