Packets, Frames, and LANs

Tom Kelliher, CS 318

Feb. 7, 2000

Administrivia

Announcements

Assignment

Read Ch. 7, Chs. 9--12.

From Last Time

Cross site scripting

Outline

  1. Packets and frames.

  2. LANs.

Coming Up

Perl: minMax, parsing.

Packets and Frames

Packets

  1. Packet: a block of data.

  2. Why break data into packets?
    1. Improve error detection.

    2. Fairness: a network is a shared resource.

      Examples: A 5MB ftp over a 56K line, telnet over the College's T1.

  3. A LAN is a broadband medium, use TDM to share.

Frames

An RS-232 ``frame:''

If we have two adjacent frames, they'll be separated by EOT/SOH, which is a bit redundant. Why can't we just separate frames with a single SOH?

Byte/Bit Stuffing

Suppose the payload is binary data and contains either SOH or EOT?

  1. Is that a problem?

  2. What can we do?

  3. Byte/bit stuffing: an ``escape'' sequence:
    1. SOH becomes ESC x.

    2. EOT becomes ESC y.

    3. Are we done? No: ESC becomes ESC z.

    4. Are we done? What about ESC z in the source block?

Error Detection

  1. Why do we need to do this? It's overhead, after all.

  2. Common types of errors: bit, vertical, burst.

    Causes.

  3. How do we detect errors?
    1. Must add redundancy.

    2. Goals: maximize detection, minimize overhead.

Error Detection Methods

  1. Parity checking.
    1. What is it?

    2. Computed across bytes.

    3. How effective? (Only catches odd # of bit flips.)

    4. How costly? (12%)

  2. Checksums.
    1. What is it?

    2. Computed across 16- or 32-bit chunks of data.

    3. Vulnerable to vertical errors.

  3. CRC.
    1. Better error detection at no additional (bandwidth) cost.

    2. Needed hardware: XOR gates, shift registers.

      XOR truth table.

      Shift register operations: initialize/clear, shift, (hold).

    3. Example CRC circuit:

      Polynomial: .

    4. Complex interactions between bits in a packet. A single bit ``counts'' more than once.

LANs

Goals:

  1. Minimize cost.

  2. High reliability.

Basics:

  1. Point-To-Point networks.
    1. connections --- expensive.

    2. Does not scale well.

    3. Used for long-haul networks.

  2. LANs: a shared communication medium.

  3. Principle of locality: temporal, physical/spatial.

LAN Topologies

  1. Star: what you have with a hub.

    Hub broadcasts everything.

  2. Ring.

    Nodes on a ring retransmit data.

  3. Bus.

    How can a star be a bus?

Ethernet

  1. Shared bus.

  2. Speeds: 10 Mbps, 100 Mbps, gigabit.

  3. CSMA --- ``carrier'' sense multiple access. Listen before transmitting. Idle.

  4. CD --- collision detection. Listen to what you've transmitted.

  5. Binary exponential backoff. Upon collision, wait for up to d time units, randomly chosen, before retransmitting.

    The exponential part.

IBM Token Ring

  1. What is a token?

  2. How do I transmit on the token ring?

  3. ``Pass'' nodes.

  4. How do I receive? How do I know it's for me?

    Why is this efficient: Layering. The NIC is responsible for determining which host a message is for, etc., not the CPU.

  5. Transmitter can do error detection.

ATM

(Asynchronous Transfer Mode)

  1. Star switch, not a bus.

  2. Full duplex, so two fiber strands needed. (Fiber doesn't allow for FDM?)

  3. Fast: 155 Mbps and up.



Thomas P. Kelliher
Mon Feb 7 10:02:56 EST 2000
Tom Kelliher