Protocols and Architecture

• Starting Notes
  • From the book:
    • A Protocol Architecture is the layered structure of hardware and software that supports the exchange of data between systems and supports distributed applications, such as electronic mail and file transfer.
    • At each layer of protocol architecture, one or more common protocols are implemented in communicating systems. Each protocol provides a set of rules for the exchange of data between systems.
    • Key functions typically performed by a protocol include encapsulation, segmentaion and reassembly, connection control, orderly delivery, flow control, error control, addressing and multiplexing.
    • The most widely used protocol architecture is the TCP/IP protocol suite, which consists of the following layers: physical, network access, internet, transport and application.
  • We will go into more detail here, but we will still put portions of our discussion off until later chapters.
  • The most widely studied theoretical protocol architecture is the OSI model. We will look at this as well.

• Protocols
  • Characteristics
    • Direct, indirect
      • A description of the connection.
      • Four general classes here
        • Point to point
          • Directly connected
          • Example: Two people in an office communicating.
          • No intermediate agents.
          • Very simple
          • 
          • Example: Ethernet, two machines back to back, crossover cable.

        • Brodcast
          • All nodes share a common broadcast medium.
          • Example: This classroom.
          • Example: Ethernet with multiple machines
          • No itermediate agents, (direct)
          • Complexity -
            • Protocol needed to detect collisions
            • Ethernet: Line must be quiet for a period after you broadcast (Time for signal to travel the entire lenght of the media).
            • Protocol needed to correct collisions
            • Called access control.
            • Ethernet: if two machines have a collision, back off a random time amount and rebroadcast.
            • Simple hardware here ($)
          • 

        • Switched Network
          • A switched network connects the machines.
          • Example: People in offices calling within a building, using a single pbx. Note each person may have multiple phone lines, ie they can put someone on hold.
          • Example: ATM
          • 
          • Rely on intermediate hardware.
            • This hardware must understand how to set up a circuit or route a packet.
              • Note, this can be table driven.
            • It must understand the addressing scheme.
            • It must be able to read part of the packet.
            • $$

        • Combination of Networks (Internet)
          • Networks connected together
          • Example: Long Distance phone calls.
          • Example: Internet
          • Hardware is needed:
            • To decide where to route packets
              • Note, this must be dynamic
            • Possibly translate between protocols
            • Must understand multiple protocols
            • $$$$
          • High level of complexity.
          • 

    • Monolithic, structured (Modular)
      • Monolithic code performs all of the work its self.
      • No one does this
      • Well, sort of.
    • Symmetric, asymmetric
      • Is the communication (as defined by the protocol) two way (symetric) or one way (asymetric)?
    • Standard, nonstandard
      • Have we beat this dead horse yet?

  • Functions
    • What do we expect a protocol to do?
    • Some but not all (or even more):
      • Encapsulation
        • Protocol Data Units (PDU, packet)
        • As we have seen, each has a payload and a header (and possibly a trailer)
        • Control infromation (contained in the added parts)
          • Address - source, destination
          • Error Dection Code - sequence, check code
          • Protocol Control - priority, other info
        • This process is called encapsulation
        • This adds to packet size
        • This decreases throughput.
        • Example:

          Assume the three layer communication model from
          the previous chapter (application, transport, and
          communications).  Assume the network is capible of
          supporting 10mb/sec.  Assume that the following:
             
          
             
          Layer
           Control Information
          PDU Size
             
          Transport
           10 bytes
           100 bytes
             
          Network
           5 bytes
           125 bytes
             
          What is the maximum utilization possible with this 
          network with and without the protocol architecture?
          
          We will assume that without the protocol architecture,
          we can achieve 100% utilization.  IE all of the network
          is used, so that in one second we get 10x106 bits
          of application data.  
          
          With the protocol architecture as described, each packet is
          125 bytes (5 header + 120 payload), of this 15 bytes are used 
          for control, and a further 20 are padding.
          
          
          So in this case, 90 bytes of 125 are used for the application,
          or 72% of the data transmitted is application data.  This means
          that the maximum throughput we can achieve is 7.2 mb/sec.
          

      • Segmentation and reassembly
        • It is almost always necessary to break the data as described above.
        • Lower level protocols may break the data down further.
        • This is segmentation
        • We need to put these back together on the other end.
        • Sometimes the underlying network limits the packet size (ATM 53 Bytes or octets fixed, Ethernet 1526 octets variable)
        • Small packet sizes are good because: (Why drive a car?)
          • Error detection is easier with smaller packet size (fewer bits for error checking)
          • Less to retransmit in the case of an error
          • More equatible access to shared facilities. (Trucks vs cars at tole booths)
          • Smaller buffers
          • If checkpoints are required, these can occur more often.
        • Large packet sizes are good (Why drive a truck)
          Less overhead.
          • Fewer OS inturrupts.
          • Less time parsing control information.
        • Good design requires a balencing of these two.

      • Connecton Control
        • If each packet is completely independant from all other packets, the communication is called connectionless
        • We also refer to these packets as datagrams
        • An example of this is ping
          • Ping uses a portion of IP konwn as ICMP
          • Internet Control Message Protocol
          • P 546 - 549
          • A suite for communicating network problems
            • Destinaton Unreachable
            • Timestamp
            • Redirect
            • Echo, Echo Response
            • Address request
          • Ping sends an ICMP Echo Request
          • Recieves and ICMP Echo Response

            • [root@imladras bennett]# ping 10.1.1.100
              PING 10.1.1.100 (10.1.1.100) from 10.1.1.103 : 56(84) bytes of data.
              Warning: time of day goes back, taking countermeasures.
              64 bytes from 10.1.1.100: icmp_seq=0 ttl=255 time=1.845 msec
              64 bytes from 10.1.1.100: icmp_seq=1 ttl=255 time=1.380 msec
              
              --- 10.1.1.100 ping statistics ---
              2 packets transmitted, 2 packets received, 0% packet loss
              round-trip min/avg/max/mdev = 1.380/1.612/1.845/0.235 ms
              			  

            • [root@localhost sbin]# tcpdump ip -nn
              Kernel filter, protocol ALL, TURBO mode (575 frames), datagram packet socket
              tcpdump: listening on all devices
              14:10:37.630169 eth0 > 10.1.1.103 > 10.1.1.100: icmp: echo request (DF)
              14:10:37.630169 eth0 < 10.1.1.100 > 10.1.1.103: icmp: echo reply (DF)
              14:10:38.630169 eth0 > 10.1.1.103 > 10.1.1.100: icmp: echo request (DF)
              14:10:38.630169 eth0 < 10.1.1.100 > 10.1.1.103: icmp: echo reply (DF)
              			  

          • 
        • In other data transfer moded, a connection is established.
          • connection-oriented
          • Much better for longer transactions
            • Allow for establishing parameters of exchange
            • Allow for tuning of parameters during connection
          • At least three phases:
            • Establish connection , includes negotiation of parameteres.
            • Transfer Data
            • Terminate Connection.
            • 
          • Each numbers the packets and keeps track of the sequence.

      • Ordered Delevery
        • If packets do not arrive in order, they must be placed in order.

      • Flow Control
        • This is a producer/consumer model
        • If the producer is too fast, something may need to tell it to slow down.
          • This may be the receiver
          • Or something in the path to the receiver

      • Error Control
        • Additional bits to indicate error in a packet
        • Watch sequence to see that all packets arrive
        • No ack- resend the packet
        • Don't ack bad packets.

      • Addressing
        • In this discussion, we will look at how TCP/IP does addressing
        • Addressing Level: The level in the architecture at which an entinty is named.
        • In TCP/IP it is at the IP level.
        • Every device on the network has an unique IP address
        • IP address
          • 32 bit number broken into 4 octets
          • Assigned to entities.
        • This identifies the machine, but we still need to identify the service on the machine
          • In addition, a number of Service Access Points (SAP) are established
          • We call these ports
          • Only one connection may be established at a port at a time.
          • /etc/services
          • source:port-ip destination:port-ip uniquely identifies a connection
        • Global nonambiguity- a global address identifies a unique system
          • A system may have more than one address
          • No two have the same address
        • Global appplicability
          • At any point, you can identify any other address.
        • These two properties allow data to be routed on the internet
        • 
        • Addresses can be:
          • Unicast - between two hosts
          • broadcast - everything on the network
          • Multicast - everything subscribing to the channel

      • Multiplexing
        • Combining or breaking multiple connections in a single system/ multiple systems.

      • Transmission Services
        • Some examples:
          • Priority
          • Quality of Service
          • Security

• OSI
  • Open Systems Interconnection
  • Established in 1977, Published 1984
  • The Model
    • A set of layers
    • Each layer must perform a set of functions
    • Each relies on lower levels for services
    • Implementation at each layer is independant.
    • Changes in implementation of a layer should not effect any other layers.
    • Sufficient layers to make each layer manageable
    • Not too many layers, reduce overhead
    • Actual communications only occurs at lowest level
    • Any underlying network is permitted.
    • Encapsulaton through all layers.
    • Lower layers may segment data from higher levels.
  • Standardization within the OSI Framework
    • Layers are independant, after the interface with the other layers are developed:
      • Design can occur in parallel
      • Development can occur in parallel
      • New standards can be added to a layer without reworking other layers
    • Elements of standaradization
      • Protocol Specification - how entities at the same layer communicate
      • Service Definition - services provided by that layer
      • Addressing- both network address and port. (ports at each layer of OSI)
    • 
  • Service Primitives and Parameters
    • A communication consists of
      • Layer N invokes N-1 with a request
      • N-1 prepares an N-1 PDU to send to it's peer
      • Destination N-1 delivers the data to destination layer N
      • N sends an ACK if called for to N-1 (as above)
      • N-1 sends ACK to source N-1
      • source N-1 sends Ack to N
    • This is called confermed service.
    • If no ACK is sent, this is nonconfermed service.
  • The OSI Layers
    • Physical
      • Mechanical properties of how the transmission medium works
      • Electrical properties on how to represent bits
      • Functions performed by circuits in the interface between system and transmission media
      • Procedures on the sequence of events in exchanging data over the physical media
    • Data Link Layer
      • Make the physical layer relyable
      • Error detection and Control
    • Network Layer
      • Transmits data over a network
      • Network addresses
      • Network facilities
      • Network Properties
    • Transport Layer
      • Connection oriented layer
      • Exchange of data between machines
      • In order, error free, everything present, no duplication
    • Session Layer
      • Relibility
      • Grouping
      • Recovery
    • Presentation Layer
      • Data Compression
      • Security
    • Application Layer
  • The first three are network layers (to next node).
  • The top four are communications layers (End to end).

• TCP/IP
  • The TCP/IP Approach
    • Modular and heirarchical
    • No Attempt to combine common features at a level
    • No need to go through all of the layes.
  • TCP/IP Protocol Architecture
    • Application Layer
    • Transport Layer (TCP) - end to end
    • Internet Layer (IP) - to the next device
    • Network Acces Layer
    • Physical Layer
  • Operation of TCP and IP
    • Application hands a message to TCP with a destination address
    • TCP adds a destination port, sequence number, and checksum
    • TCP hands this message to IP
    • IP Adds routing control information to each datagram and hands them to the low level internet
  •