Ethernet: Distributed Packet Switching for Local Computer Networks

1
Ethernet Distributed Packet Switching for Local
Computer Networks

• Authors Robert M. Metcalfe and David R. Boggs
• Presentation Christopher Peery

2
Outline

• Quick Overview
• Design Choices
• Ethernets Behavior
• Limitations
• The Good Side of the Ether

3
Quick Overview

• So we probably all heard about it one time or
  another, so what is it?
• In brief, its a broadcast medium with no
  centralized control, that is used to interconnect
  computers in a localized area.
• Was greatly influenced by the Aloha Network used
  by the University of Hawaii. This was a satellite
  based communication system
• Exists as 10 Mbs, 100 Mbs, and 1 Gbps
• Uses 1-Persistent CSMA-CD (Carrier Sense Multiple
  Access Collision Detection)

4
Design Choices

• Requirements
• Was intended for allow localized communication
  among workstations
• Should be easily extensible
• Control should be completely decentralized to
  avoid creating a bottleneck
• Should be relatively inexpensive
• Ethernet will not make any absolute guarantees
  with respect to packet deliver. It is
  best-effort.
• Communication medium should be as simple as
  possible

5
Identification

• Each Ethernet packet contains a source and
  destination address.
• All packets reach every workstation on the
  segment
• Packets are filtered at the network interfaces
• There is a reserved address for broadcast

6
Ethernet Behaviour

• Ethernet can be examined in the following areas
• Topology
• Control
• Collision detection

7
Topology

• Ethernet is a broadcast medium
• Workstations single attach themselves to the
  shared medium
• Imagine a straight line with workstations
  attached to it on either side.
• Several Ethernet segments can be connect in
  sequence using a device known as a repeater or a
  bridge (a little smarter then a repeater).

8
Control

• Control in Ethernet is completely decentralized.
• The medium has no control over who uses it. Its
  only task to to move information from point A to
  B.
• Is this good?
• Every time a workstation needs to transmit, it
  tries to take control of its Ethernet segment.
• This is done by listening to the shared medium to
  determined if it can be used and then actually
  using it.
• How does this work?

9
When to transmit

• If a node want to transmit, it will first listen
  on the segment.
• If the node hears a transmission, it will wait
  for one transmission period for the sending node
  to finish. Then it will immediately attempt to
  send its data. This is known as deference.
• If the node hears no one sending, it will send
  immediately.
• The idea then once a node is transmitting, it
  effectively owns the segment is known as
  acquisition.
• This is 1-Persistent CSMA. Meaning when you
  detect a chance to transmit do so with
  probability 1.
• What about collisions?

10
Collisions Handling

• If a collision is detected, the two sending nodes
  stop transmitting.
• Each node will then chose a new random slot from
  2k slots, where k is the number of collisions
  already suffered.
• This is done 16 times then a failure is reported
• This call Binary Exponential Back-off
• Note for this to work everyone has to play
  nicely
• The duration of the retransmission interval is
  computed using collision history. This allows it
  to change with the network load.

11
Detecting a Collision

• When a node sends a packet, it also listens to
  the segment to ensure the information is not
  damaged. This is collision detection.
• When a collision is detected the sending node
  will jam the Ether.
• This is done to guarantee that all sending nodes
  are made aware of the collision.
• Ethernet also specifies that a checksum be
  computer for each packet to detect errors from
  noise this kind of an error does not result in a
  back-off.

12
Limitations

• Ethernet has been very successful but also has
  some limitations
• The Ethernet Standard specifies that up to most
  5 segments may be connected with repeaters. Each
  segment at most 500m.
• 256 workstations can be attached to a given
  segment (this is never done in practice why?)
• Most sys admins try to keep Ethernet segments at
  60 utilization. Else they become unusable.
• The minimum packet size is determined by the
  segment bandwidth and length (RTT). Why is this?

13
The Good side of the Ether

• Good example of the End-to-End argument
• Completely distributed control
• Simple
• Easy
• Cheap