Topics related to Application Protocols

The scope of this topic includes applications and application protocols from TCP/IP (ch 2 of Kurose) along with our continued study of simple sockets network programs

• datacom 102 - concepts:
  • communication and network errors--- what can prevent a frame that is transmitted to appear to the sender to be 'sent ok' but either arrive at the receiver damaged or not arrive at all?
    • First concept: detection and recovery :
      • Error Detection:
        • Within a frame (bit errors) using some sort of check sum or CRC
          • IP Checksum overview
        • Link level - if a frame never even arrives at a receiver, a sliding window protocol is typically used.
        • At the socket endpoint (the transport protocol OR the application)- might observe a gap in sequence numbers, OR, a sender's transmission timer might pop
      • Error Recovery -
        • Over wireless where bit errors are common, redundant data is added within frames so that the frame data can be received in spite of some number of bit errors
        • Reliable data transfer (RDT) protocols are used to allow a receiver to coordinate with a sender regarding what might need to be retransmit.
    • Second concept: cause of error
      • channel errors that lead to bit errors at a receiving physical layer
      • network congestion that lead to routers dropping arriving packets due to a 'buildup' or a queue as packets wait their turn for transmission
  • Common network measures to help monitor a network
    • packet delay - one way delay components: transmission time, propagation time, for each router in the path between the two hosts that form the socket. forwarded by routers over the path
      • Measureing one-way delay : if the endpoints are in perfect time sync, this is easy.... the packet can include the senders time of transmission and the receiver observes an end-to-end one-way latency of the current time minus the senders Tx time
        • (not easy to keep computers in accurate time sync)
      • Round trip time (RTT): Many applications or network protocols require a sending side to wait for an acknowledgement back from the receiving side that it received the data (we'll call this an ACK message!!).
        • A performance tool like ping is designed to help us estimate the aggregate delay of the system by sampling with periodic ping (send/receive) iterations. Each line of ping shows an RTT sample. After running for a number of iterations, and assuming the network system is in some form of 'steady state' , we should see a mean RTT and the standard deviation gets lower and lower as we include in the computation more and more samples. The average RTT is a measure of interest!!
    • loss rate - probability that a sent from one Host is correctly received by the destination host.
      • Typically modeled as a simple uncorrelated loss model- any given packet is not received correctly with a probability p.
      • performance tools such as ping can be used to sample the loss process and give us an estimate (the sample mean) of the average loss rate
• Protocols and Layering
• Review intro material on sockets - and in particular on the type of socket : streaming, datagram, raw.
  • Each maps to something different from a socket perspective and from a protocol perspective: TCP, UDP, ICMP.
    • TCP: provides error detection and recovery, handles all possible cases of errors, including the case when a 'fast sender ' potentially overruns a 'slow receiver'
      • The congestion point is the socket receive queue at the destination.... TCP has flow control that prevents a fast sender from overflowing the slow receiver's Rx socket buffer
    • UDP: bit error detection optional, some UDP applications as error detection by using sequence numbers, most UDP applications do not care if sometimes messages do not arrive (i.e., UDP does not provide recovery, and applications are ok with occasional packet loss)
    • ICMP is a strange thing....it's somewhere in between an application, transport, and network layer function. For now, just know that ping uses ICMP messages. It is a transport layer function but it does not do protocol level error detection or recovery.
• DNS overview
• NFS- The Network File System Tutorial

last updated: 3/25/2019