Benes Networks

Permutation Networks
and
Rearrangeability

Kosta Tachtevrenidis
 Introduction
What is this paper about?
This paper came out of a time when telephone companies
were expanding like mad and were seeking ways to connect
all nodes of a telephone network without using n2
connections.

Do I need to understand ALL this math to fully grasp the

material?
I certainly hope not because I don't understand it either.
Yeah, I know it is my paper but if somebody can do a better
job please stand up.
 Preliminary Definitions

stage (of switching)

a series of switches that define a routing step
link (pattern)
the pattern of connections that links two switching stages
square switch
a switch with the same amount of inputs as outputs
non-blocking networks
A set of one-to-one connection requests on an N N
network can be defined by a permutation. A network is non-
blocking, if any permutation can be realized by edge-disjoint
paths in the network. Depending on whether the permutation
is realized statically or dynamically we have the following
types of non-blocking networks.
 Definitions continued...
Rearrangeable non-blocking (benes network)
A network is rearrangeable non-blocking if any permutation can be realized by
edge-disjoint paths when the entire permutation is known. In other words,
any permutation can be statically realized. The word rearrangeable refers to
that if the connection requests in a permutation arrive dynamically, the
permutation can be realized with possible rearranging active connections. This
is equivalent to realizing a permutation statically.

Wide-sense non-blocking
the permutation can be realized by edge-disjoint paths w/o rearranging active
connections subject to the condition that a selected path is used for each new
connection request. In other words, any permutation can be dynamically
realized with the help of a wise algorithm.

Strict-sense non-blocking (n x n crossbar network)

the permutation can be realized by edge-disjoint paths without rearranging
active connections, any idle path can be used for each new connection
request. In other words, any permutation can be dynamically realized.
 Permutation Networks

Imagine a network with n inlets labeled [1...n]. A

permutation network is one in which the set of links used to
route a call from a given inlet, leads to an outlet which is a
way is a permutation of those inlets.
Example:
0

This is a butterfly network. The result of routing to the outlets

is a permutation of the inlets.
 Rearrangeable Networks
Rearrangeable non-blocking (benes network)
A network is rearrangeable non-blocking if any permutation can be
realized by edge-disjoint paths when the entire permutation is known.
In other words, any permutation can be statically realized. The word
rearrangeable refers to that if the connection requests in a permutation
arrive dynamically, the permutation can be realized with possible
rearranging active connections. This is equivalent to realizing a
permutation statically.

The above is from a previous slide. What that says is that given a
network with n inputs [1,2,...,n], if I know what the resulting
permutation should be, I can 'chart' a path for all inlets in a way that no
path is blocked and everything is routed correctly.
 Why Should I Care?

Benes networks provide provably congestion-free

communication from inputs to a permutation of the
outputs.

Any n-permutation can be routed (off-line) on an n-

input Benes network with node-disjoint paths.
Proof? The proof relies on the fact that benes networks
are composed of smaller benes networks. Induction is
the key to the proof.

An n-input Benes network can simulate any n-node,

degree-d network in O(d lg n) time.
 How do I construct one?
Short answer: recursively

Long answer: To construct one start with a small (2 input)

benes networks and build on from there. To build a 4 input
benes network, cascade two 2 input networks and add a stage
to the left and one to the right. Multiplex the links as shown in
the next slide. Confused yet?

Really long answer: read the paper.

2 input benes network

1
2
 Constructing the network...
4 input benes network

0
1

2
3

Two versions of the same

Benes network are depicted 0
here. The above is the shuffle 1
version and the one below is 2
3
the butterfly version. Can you
tell they route in the exact
same way and therefore they
are equivalent?
 Constructing the network...
8 input benes network (butterfly)

0
1
2
3
4
5
6
7
The benes
network
is composed of
the same sub-
network mirrored.
What kind of
network is this?
 Constructing the network...
8 input benes network (shuffle)
 Constructing the network...
16 input benes network (butterfly)
recursion
 What are the advantages?

The advantages of using Benes networks are profound.

Pure non-blocking and rearrangeable network switching power.

But let's see how it ranks on Pfeiffer's Network Ranking Test:

Good- latency how many hops to the destination?
The number of hops corresponds to the number of stages. It
turns out to be: O(log(n))
Fast bandwidth # of simultaneous transmissions?
Benes networks are non-blocking so all nodes can be
transmitting at the same time. O(n)
Cheap how many switches needed to implement?
The number of switches required to implement a benes
network is logarithmically proportional to the number of
nodes. O(n*log(n))
 Example Please!

What if 0 and 3 are already routed but I still want to realize

the permutation (0->1, 3->2, 2->0, 1->3)? Obviously it can't
be done with the existing state of things. But this is a
rearrangeable network so I can reroute 1 or 4 or both to create
a routing scheme that does not block!
 Example Continued!

Tada!
 Real World Applications
The switching network (of this Gigabit ATM switch)
uses a Benes network topology. The Benes network
extends to arbitrarily large configurations by way of a
recursive expansion. The figure in the next slide shows a
64 port Benes network. We can construct a 512 port
network by taking eight copies of the 64 port network
and adding a first and fifth stage on either side, with 64
switch elements apiece.

It is shown that when dynamic load distribution is

performed in the first k-1 stages of a Benes network,
that the load on the internal data paths of the switching
network cannot exceed the load on the external ports;
that is the network achieves ideal load balancing.
 ATM Gigabit Switch Schematic

DESIGN OF A GIGABIT ATM SWITCH

Tom Chaney, J. Andrew Fingerhut,
Margaret Flucke, Jonathan S. Turner
Washington University, St. Louis
 What's next? Is there something better?
You bet there is...

Conventional Benes networks have control complexity

n(2log(n) -1). This probably has to do with the cost of
combining the two butterfly networks times the amount
of nodes. But can we do better than that?

What if we could create a network that could create all

the same permutations in a non-blocking way with less
switches? Enter KR Benes Networks!

This paper presents a network topology built on benes

networks which has control complexity of O(log(n)).
How do they do it? They exploit the locality property of
permutations.
 KR-Benes Network with N=16

The KR Benes Network by Rajgopal Kannan

 Alternatives to Benes Networks

Butterfly Networks (blocking):

Benes networks have the advantage that any
permutation of inputs to outputs can be routed
without conflicts. However, computing the paths for a
given permutation takes a long time, and requires
global information about all of the messages in the
network, so typically can't be done at run-time.
Butterflies have some conflicts, but can be routed at
run-time. They also have lower latency because they
have only half as many stages as a Benes network for
the same number of nodes. 0

1
2

3
 Alternatives to Benes Networks

Cantor Networks (strictly non-blocking):

m copies of Benes network.
Network size : N log2 N

m=3
 References

Kannan, Rajgopal. The KR-Benes Network: A Control-

optimal Rearrangeable Permutation Network. Department
of Computer Science, Louisiana State University.

Boppana Rajendra, Raghavendra C.S. Designing Efficient

Benes and Banyan Based Onput-Buffered ATM Switches.
The AeroSpace Corporation University of Texas, San
Antonio.

Chaney, Fingerhut, Flucke, Turner. Design of a Gigabit

ATM Switch. Washington University, St. Louis, Missouri.