dealing with network message lost

handling lost message: acknowledgements

handling lost message

protocol so far

  • on sender: until ACK received:

    • (re)send frame of data
    • wait fixed amount of time for ACK

  • on receiver: continuously:

    • wait for frame of data
    • send ACK back

problem

  • really want to send multiple frames
  • example: data split in multiple pieces

splitting messages: try 1

reconstructed message:
The meeting is at 12pm.

splitting messages: try 1 — problem 1

reconstructed message:
The meetingThe meeting is at 12pm.

exercise: other problems?

  • sending ‘The meeting’, ‘is at 12pm’
  • what would be received for each of these scenarios?
    1. message (instead of acknowledgment) is lost
    2. first message from machine A is delayed a long time by network
    3. acknowledgment of second message lost instead of first

aside: message delays

  • long message delays not possible with direct link

  • but are possible with:

    • multiple paths from A to B
    • doing this kind of acknowledgment + resending hop-by-hop

splitting messages: try 2

reconstructed message:
The meeting is at 12pm.

splitting messages: try 2 — missed ack

reconstructed message:
The meeting is at 12pm.

splitting messages: try 2 — problem

A thinks: part 0 + part 1 acknowleged!

splitting messages: version 3

sequence numbers

  • call the ‘part’ label sequence number

    • for now: sequence number = message (or segment) number
    • in TCP: sequence number = byte number
  • important question: how large can they get?
  • if we never reuse them — infinite!
  • so really want to reuse them

1-bit sequence number

‘stop and wait’

  • machine A is only sending one thing at a time
  • never start sending next thing until after sending previous thing

stop-and-wait exercise (receive, 1)

  • machine B receives 0: X
  • machine B sends ACK 0
  • machine B receives 0: X
  • what should machine B do now?
    A. send ACK 0 B. send ACK 1 C. send nothing

stop-and-wait exercise (receive, 2)

  • machine B receives 0: X
  • machine B sends ACK 0
  • machine B receives 1: X
  • what should machine B do now?
    A. send ACK 0 B. send ACK 1 C. send nothing

stop-and-wait exercise (receive, 3)

  • machine B receives 0: X
  • machine B sends ACK 0
  • machine B receives 1: Y
  • machine B sends ACK 1
  • machine B receives 0: X
  • what should machine B do now?
    A. send ACK 0 B. send ACK 1 C. send nothing

stop-and-wait exercise (send, 1)

  • A trying to send ‘X’, then ‘Y’, then ‘Z’
  • machine A sends 0: X
  • machine A sends 0: X
  • machine A receives ACK 0
  • machine A sends 1: Y
  • machine A receives ACK 0
  • what should machine A do now?
    A. send 0: X again B. send 1: Y again
    C. send 0: Z D. something else

stop-and-wait exercise (send, 2)

  • A trying to send ‘X’, then ‘Y’, then ‘Z’
  • machine A sends 0: X
  • machine A sends 0: X
  • machine A receives ACK 0
  • machine A sends 1: Y
  • machine A receives ACK 1
  • what should machine A do now?
    A. send 0: X again B. send 1: Y again
    C. send 0: Z D. something else

stop-and-wait issues

  • two issues with stop-and-wait:
  • doesn’t use close to full capacity of network
  • not clear how to set timeouts

looking at metrics

  • several important metrics we’ll care about
  • (both for this and future topics)
  • throughput and bandwidth (\(\sim\) how much capacity used/available)
  • latency and round-trip time (RTT) (\(\sim\) what timeouts needed)
  • jitter (\(\sim\) safety margin for timeouts)

stop-and-wait performance

  • stop-and-wait protocol
  • assuming no packets lost/corrupted
  • about one packet per round-trip time
    • (assuming packet tranmission time short relative to round-trip time)

example: local ethernet

  • my home wired network: 0.6 ms round trip time
  • typical packet has about 1400 bytes = 11200 bits of data
  • throughput with stop-and-wait: \(11200 \text{b} / 0.6 \text{ms} \approx 19000 \text{b/ms} = 19\;000\;000 \text{b/s} = 19 \text{Mbit/s}\)
  • available bandwidth is about \(1\) Gbit/s

backup slides

dealing with network message lost

handling lost message: acknowledgements

handling lost message

protocol so far

  • on sender: until ACK received:

    • (re)send frame of data
    • wait fixed amount of time for ACK

  • on receiver: continuously:

    • wait for frame of data
    • send ACK back

problem

  • really want to send multiple frames
  • example: data split in multiple pieces

splitting messages: try 1

reconstructed message:
The meeting is at 12pm.

splitting messages: try 1 — problem 1

reconstructed message:
The meetingThe meeting is at 12pm.

exercise: other problems?

  • sending ‘The meeting’, ‘is at 12pm’
  • what would be received for each of these scenarios?
    1. message (instead of acknowledgment) is lost
    2. first message from machine A is delayed a long time by network
    3. acknowledgment of second message lost instead of first

aside: message delays

  • long message delays not possible with direct link

  • but are possible with:

    • multiple paths from A to B
    • doing this kind of acknowledgment + resending hop-by-hop

splitting messages: try 2

reconstructed message:
The meeting is at 12pm.

splitting messages: try 2 — missed ack

reconstructed message:
The meeting is at 12pm.

splitting messages: try 2 — problem

A thinks: part 0 + part 1 acknowleged!

splitting messages: version 3

sequence numbers

  • call the ‘part’ label sequence number

    • for now: sequence number = message (or segment) number
    • in TCP: sequence number = byte number
  • important question: how large can they get?
  • if we never reuse them — infinite!
  • so really want to reuse them

1-bit sequence number

‘stop and wait’

  • machine A is only sending one thing at a time
  • never start sending next thing until after sending previous thing

stop-and-wait exercise (receive, 1)

  • machine B receives 0: X
  • machine B sends ACK 0
  • machine B receives 0: X
  • what should machine B do now?
    A. send ACK 0 B. send ACK 1 C. send nothing

stop-and-wait exercise (receive, 2)

  • machine B receives 0: X
  • machine B sends ACK 0
  • machine B receives 1: X
  • what should machine B do now?
    A. send ACK 0 B. send ACK 1 C. send nothing

stop-and-wait exercise (receive, 3)

  • machine B receives 0: X
  • machine B sends ACK 0
  • machine B receives 1: Y
  • machine B sends ACK 1
  • machine B receives 0: X
  • what should machine B do now?
    A. send ACK 0 B. send ACK 1 C. send nothing

stop-and-wait exercise (send, 1)

  • A trying to send ‘X’, then ‘Y’, then ‘Z’
  • machine A sends 0: X
  • machine A sends 0: X
  • machine A receives ACK 0
  • machine A sends 1: Y
  • machine A receives ACK 0
  • what should machine A do now?
    A. send 0: X again B. send 1: Y again
    C. send 0: Z D. something else

stop-and-wait exercise (send, 2)

  • A trying to send ‘X’, then ‘Y’, then ‘Z’
  • machine A sends 0: X
  • machine A sends 0: X
  • machine A receives ACK 0
  • machine A sends 1: Y
  • machine A receives ACK 1
  • what should machine A do now?
    A. send 0: X again B. send 1: Y again
    C. send 0: Z D. something else

stop-and-wait issues

  • two issues with stop-and-wait:
  • doesn’t use close to full capacity of network
  • not clear how to set timeouts

looking at metrics

  • several important metrics we’ll care about
  • (both for this and future topics)
  • throughput and bandwidth (\(\sim\) how much capacity used/available)
  • latency and round-trip time (RTT) (\(\sim\) what timeouts needed)
  • jitter (\(\sim\) safety margin for timeouts)

stop-and-wait performance

  • stop-and-wait protocol
  • assuming no packets lost/corrupted
  • about one packet per round-trip time
    • (assuming packet tranmission time short relative to round-trip time)

example: local ethernet

  • my home wired network: 0.6 ms round trip time
  • typical packet has about 1400 bytes = 11200 bits of data
  • throughput with stop-and-wait: \(11200 \text{b} / 0.6 \text{ms} \approx 19000 \text{b/ms} = 19\;000\;000 \text{b/s} = 19 \text{Mbit/s}\)
  • available bandwidth is about \(1\) Gbit/s