Richer Networks

Tom Kelliher, CS 315

Apr. 19, 1999

Administrivia

Announcements

Assignment

Read openings of 3.1, 3.2, 3.3.

From Last Time

Multiplication, convolution on linear arrays.

Outline

1. Hypercubes.

2. Butterflies.

Coming Up

Shuffle-Exchanges.

The Hypercube

Definition of r-dimensional hypercube:

1. Has nodes and edges.

2. Each node is labeled with an r-bit number.

3. Two nodes are connected only if they differ in one bit.

4. Edges partition nicely along dimensions. An edge is called a dimension k edge only if it connects two nodes which differ in the kth bit position.

5. Recursive structure.

Examples: 0- through 4-dimensional hypercubes.

Properties

1. How can we find a path from node to node ?

2. What is the diameter of the graph?

3. What is the bisection width of the graph? ( --- remove any dimension's edges.)

Containment of Arrays

1. Mapping a linear array.
   1. Hamiltonian cycles.

   2. Proof by induction. Basis: N=4.

      Step: Partition the N-node hypercube into 2 -node hypercubes. Assume that there is an edge (x010, x110) and, symmetrically, (x011, x111) in each sub-hypercube, respectively.

      To construct the linear array for the hypercube, remove those edges and add edges (x010, x011) and (x110, x111).

2. Mapping a 2-D array. Example: construct a 4x4 array from a 16 node hypercube.

The Butterfly

Definition of r-dimensional butterfly:

1. Has nodes and edges.

2. Each node is labeled , where r (binary) is the row and l (decimal) is the level.

3. Two types of edges:
   1. Straight edges connect and .

   2. Cross edges connect and , where r and differ only in the lth bit position.

   3. Recursive structure.

Example: 3-dimensional butterfly.

Observation: Collapsing horizontally, we get the hypercube.

Properties

1. How can we find a path from node to node (r',l)?

   Extend to nodes on arbitrary levels.

2. What is the diameter of the graph?

3. What is the bisection width of the graph? Is r correct?