Shared Memory Architectures
System Bus
Communication pathway that allows various components within a computer system to communicate with each other.
- Serves as a central conduit through which data, instructions, and control signals are transmitted between the CPU (Central Processing Unit), memory, input/output devices, and other subsystems
- Multiple processors and cores of a single processor communicate through the system bus
- CS301: Topologies | Saylor Academy
Multistage Omega Network
-
Most common multistage interconnects
-
Stages lie in the middle while processors and memory modules are connected on the sides
-
Calculations
- Stages =
- Nodes =
- Costs =
Distributed Memory Architectures
Linear Array
- Each node has two neighbours, connected to left and right.
- Movement can be done from left to right or right to left
K-d Meshes
- Each node has four neighbours
- Movement can be done to north, south, east or west
Hypercube
- Based on the mathematical concept of an n-dimensional hypercube
- Each node has neighbours
- The distance between any two neighbours is at most
- d-dimensional hypercubes can be constructed by connecting corresponding nodes of multiple (d-1) dimensional hypercubes
Tree Based Network
- One path between any pair of nodes
- Linear Arrays are a special case of a tree based network
Fat Tree
- A node can communicate to the other node through the root node
- There exists higher traffic at the root node
Evaluation
Cost
-
Number of links for communication
-
Lower values are favourable
-
Example
- Cost of a Linear Array is
Diameter
- Shortest distance between the two most farthest node in the network
- Lower values are favourable
Bisection Width
- Minimum number of wires required to cut / remove in order to divide the network into two equal parts
- Higher values are favourable
Arc Connectivity
- The minimum number of arcs / links that should be cut down in order to separate a single node from the network