Ethernet/IEEE 802.3

Ethernet/IEEE 802.3
An Introduction

Prof. David Bernstein
James Madison University

Computer Science Department

bernstdh@jmu.edu

History

Xerox PARC (1970s):
- Metcalfe and Boggs developed the first system
- The network had 256 stations spread over 1 kilometer (running at 3Mbps)
IEEE 802 Committee (1980s):
- Standard 802.3 describing CSMA/CD was published in 1985

Data/Signal

Baseband Transmission:
- The signal is not shifted into another frequency band
Manchester Encoding:
- Each bit contains a transition
- A 1 is a transition from up to down and a 0 is a transition from down to up

Different Standards

10BASE-5: Bus network (with taps), coaxial cable, 10Mbps, 500 meter maximum segment length
10BASE-2: Bus network (with taps), thin coaxial cable, 10Mbps, 200 meter maximum segment length
10BASE-T: Star network, twisted pair cable, 10Mbps
100BASE-TX: Star network, CAT-5 twisted pair cable (two pairs per connection), 100Mbps, 100 meter maximum segment length
100BASE-FX: Star network, multimode fiber-optic cable (one strand in each direction), 100Mbps, 412 meter maximum segment length
1000BASE-SX (1000BASE-LX): Star network, fiber-optic cable and short (long) wavelength, 1Gbps, 500 (5000) meter maximum segment length
1000BASE-T (1000BASE-LX): Star network, CAT-5 twisted pair cable, 1Gbps, 100 meter maximum segment length

Frames (in IEEE 802.3)

Preamble:
- Seven octets of 10101010 used to synchronize clocks (using the transitions in the bits)
- One octet of 10101011 used as an alert
Destination Media Access Control (MAC) Address:
- Six octets
Source MAC Address:
- Six octets
Length:
- Two octets indicating the length of the data field
- To remain compatible with the original standard, if the value is greater than 1500, these octets indicate the network-layer protocol being used (e.g., IP, Novell IPX, AppleTalk, DECnet)
Data:
- 46-1500 octets
Frame Check Sequence (FCS):
- Four octets used for a cyclic redundancy check (CRC)

Properties

Connectionless:
- No handshaking
Unreliable:
- CRC can be used to find errors but the receiver just discards the frame

Collisions

Why They Might Arise:
- It takes time for a signal to travel through the medium
- Devices at different locations might, as a result, transmit at the same time
How Detection is Ensured:
- A device must continue to transmit for the "round trip time" (the maximum time required for a round trip)

Collisions (cont.)

When They Can Arise:
- Bus networks
- Star networks using a hub (a physical layer device that simply retransmits individual bits to all edges making the star behave like a "virtual"/"logical" bus)
- Half-Duplex Connections on Star Networks
When They Do Not Arise:
- Full-duplex (a pair/fiber for each direction) connections on star networks using a "Fast Ethernet Switch" (a physical layer device that filters and forwards frames)

Collisions (cont.) - The Process

Carrier Sense (CS): Listen to the network to determine that no other devices are transmitting.
Start transmitting. [Note: More than one device can be transmitting at the same time, even if they don't start at exactly the same time. This is called Multiple Access (MA).]
Collision Detection (CD): If two devices are transmitting, the signals interfere and this is detected by the devices (by comparing the transmitted/expected signal with the actual signal).
A device that detects a collision sends out a jamming signal to inform other devices not to transmit.
Each device that caused the collision "backs off" for a random amount of time and re-transmits.

Collisions (cont.)

An Observation:
- One frame can be involved in multiple collisions (over time)
Exponential Backoff at Collision \(n\):
- Let \(m = \min(n, 10)\)
- Choose a value \(K\) "at random" from the set \({0, 1, 2, \ldots, (2^m-1)}\)
- Wait \(K \cdot 512\) "bit times"
Rationale for Exponential Backoff:
- When \(n\) is large (often because there are a large number of devices), the set of possible backoff times should be larger and more dispersed

Collisions (cont.)

Efficiency/Utilization:
- The long-run fraction of time during which frames are being transmitted without collisions (when there are a large number of devices each with a large number of frames to send)
An Approximation (without derivation):
- \(U = \frac{1}{1 + 5 d_{\text{prop}}/d_{\text{max}}}\)
- where \(d_{\text{prop}}\) denotes the propogation delay and \(d_{\text{max}}\) is the transmission delay for the maximum size frame (about 1.2 msecs for a 10Mbps Ethernet)

Persistence (or, informally, Tenacity)

The Issue:
- What should a device do when it determines another device is transmitting?
Possible Algorithms:
- Nonpersistent: wait a random amount of time and try again
- \(1\)-Persistent: Listen until the medium isn't busy

Persistence (cont.)

A Compromise:
- A \(p\)-persistent system tries to reduce both collisions and idle time
The Algorithm:
- 1. If idle, transmit with probability \(p\) and delay one time unit with probability \(1-p\)
- 2. If busy, continue to listen until idle and then go to 1.
- 3. If delayed go to 1.

Problems that Arise