Transport layer

Transport layer

• Provide logical communication between processes running on different hosts
  • Sender host: Break message into segments
  • Destination host: Reassemble segments into message, and pass to application layer
• Rely on network layer services
  • network layer
    • Provide logical communication between hosts
• Two types of protocol
  • Transmission Control Protocol (TCP)
  • User Datagram Protocol (UDP)

Reliable data transfer protocol: pipelined protocols

Go-back N

• Sender
  • Can send N unacknowledged packets in pipeline
• Receiver
  • Only send cumulative acknowledgment
    • This means an acknowledgment for packet n indicates all packets with sequence number up to and including n have been received correctly
    • No buffer
• Timeout
  • Sender has a single timer for oldest unacknowledged packet
  • When timer expire, retransmit all sent but unacknowledged packets
    • If window size is large
      • There can be many unacknowledged packets in pipeline, so a single packet loss can cause sender to retransmit many packets

Selective repeat

• Sender
  • Can send N unacknowledged packets in pipeline
• Receiver
  • Send acknowledgment for each packet
  • Has buffer
• Timeout
  • Sender has timer for each unacknowledged packet, so it has many timers
  • When timer expire, retransmit the unacknowledged packet only
    • Address the disadvantage of Go-back-N

Go-back-N

Maintain a window with size N. A sender can send N unacknowledged packets in pipeline. When sender receive an unacknowledged packet, it slide forward window so that a sequence number change from not usable to usable, not yet sent. Each packet has sequence number.

Receiver

Always send ACK for correctly-received packet with the highest in-order sequence number. Generate duplicate ACK when receive out-of-order packet, discard out-of-order packet because receiver does not have a buffer

Re-acknowledge packet with the highest in-order sequence number

• Receiver
  • Expect to receive packet n
  • However, it receives packet n + 1. This means that packet n is lost.
  • Discard packet n + 1
  • Resend acknowledgment for packet n - 1
• Sender
  • Receive duplicated ACKs
  • Resend packet n and n + 1

Selective repeat

Sender

• If there is usable, not yet sent in window, sent it
• Resend packet n, restart timer of packet n
• When sender receive an acknowledgment for send_base packet, it slide forward window so that a sequence number change from not usable to usable, not yet sent.

Receiver

• Send acknowledgment for packet n
• If
  • Out-of-order packet
    • Buffer the packet
  • In-order packet
    • Deliver (also deliver buffered, in-order packets) to upper layer

Flow control

Control a source host’s sending rate so that is does not overflow receive host’s buffer. Inform sender the received window value rwnd

• rwnd
  • Unit type: byte
  • Indicate the buffer space availability at receiver
• Sender
  • Send packets of size < rwnd bytes to receiver

TCP congestion control

• End-to-end
• Destination host infer congestion from observed packet loss and end-to-end delay
• End-to-end delay = $\frac{RTT}{2}, $ where RTT is Round Trip Time
• Control a source host’s sending rate so that each connection traversing a congested link get an equal share link of bandwidth
• Sender
  • Send packets of size < min { rwnd, cwnd } bytes to receiver
    • rwnd - is receive window
    • cwnd - is congestion window
  • Sending rate $ \frac{cwnd}{RTT} bytes/second $
    • This means, sender send cwnd bytes, wait RTT for ACK, then send more
• Two main concept
  • Slow start
  • Congestion avoidance