Scribd
Upload
34 views

Structure Type Subtype Features Pros Cons Bisection Width Links / Switches

The document describes several common interconnection network types used to connect processors and memory in parallel computing systems. It provides a table comparing the structure, type, features, pros, cons, bisection width, and number of links or switches for different networks including bus, crossbar switch, ring, toroidal mesh, hypercube, crossbar switch, and omega network. The table allows for easy comparison of the characteristics and performance trade-offs between the different interconnection network topologies.

Uploaded by

Roshini
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
34 views

Structure Type Subtype Features Pros Cons Bisection Width Links / Switches

The document describes several common interconnection network types used to connect processors and memory in parallel computing systems. It provides a table comparing the structure, type, features, pros, cons, bisection width, and number of links or switches for different networks including bus, crossbar switch, ring, toroidal mesh, hypercube, crossbar switch, and omega network. The table allows for easy comparison of the characteristics and performance trade-offs between the different interconnection network topologies.

Uploaded by

Roshini
Copyright
© © All Rights Reserved
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 1

INTERCONNECTION NETWORK TYPES

Structure Type Subtype Features Pros Cons Bisection Width Links / Switches
- Low cost
- Collection of parallel - Too many devices
- Flexible
Shared communication wires lead to contention
Bus - - Multiple device - p+1 links
Memory - Communication wires shared - Performance
connection with less
by devices degradation
additional cost
- Collection of switches
arranged in a matrix - Faster than bus
configuration - Waits during
Shared - Multiple input and output lines contention - Cost of switch and
Crossbar Switch - p mXn switches
Memory - At each crosspoint is a - Bidirectional links is high
switch; when closed, it communication is
connects one of the inputs to possible
one of the outputs
- Nodes are connected with
Distributed each other to form a ring - Simplest ways of
Ring Direct - Increased latency 2 (p->even) 2p links
Memory - Sends the messages to its connecting
neighbors.
- Higher speed, lower
- Mesh interconnect with nodes
latency
Distributed in rectilinear array of N = 2, 3 - Complexity of wiring
Toroidal Mesh Direct - Better fairness 2q (q->even) 3p links
Memory - Topologically homogeneous - Cost
- Lower energy
to an N-dimensional torus
consumption
- Nodes form the vertices of
- Higher node
squares - Increased
Distributed complexity
Hypercube Direct - Multidimensional mesh performance p/2 (or) 2^D-1 1+log2(p) links
Memory - Costlier the toroidal
network with two nodes per - Reliability
mesh
dimension
- Every node is connected to
Distributed - Low latency - Not scalable
Crossbar Switch Indirect every other node p mXn switches
Memory - High bandwidth - Cost
- Non blocking configuration
- Consists of multiple stages of - Constant node
2X2 switching elements degree
- Blocking network
- Each input has a dedicated - Less links and
Distributed - Fixed distance p/2 switching elements in
Omega Network Indirect connection to an output switching p/2
Memory between any two each stage
- Relies on the perfect shuffle elements than
nodes
interconnection algorithm" hypercube
- log(p) number of stages - Hierarchical network
( p - no. of processors ; p = qXq in Toroidal Mesh ; D - dimension ; m - matrix rows ; n - matrix columns)

You might also like