Outline

• network performance
• layered design

network performance

• many independent measures of network performance
• fundamental measures: delay (RTT), throughput, packet loss
• for example, a satellite link may have very high throughput (good) but also very high delay (not good) and loss (not good)
• a 56KB modem may provide guaranteed minimum throughput (good) but low average throughput (not good) and high delay and packet loss
• a fiber provides low delay, low packet loss, and high throughput (all good) but is expensive

Delay

• many sources of delay
• transmission delay: the time to write a packet to the network (packet size / link capacity)
• propagation delay: the time for a bit to go from one system to the next (some fraction of the speed of light)
• processing delay: the time for a system to process a packet (looking up in the routing/forwarding table, error checking, etc)
• queueing delay: the time a packet must wait before prior traffic is deleted (varies over time)

Transmission Delay

• if a link can carry B bits per second (the capacity or bandwidth of the link)
• and a packet is made of of P bits (not bytes)
• it takes P/B seconds to write the packet to the network
• in a store-and-forward network, each packet must be received completely before it is processed
• so transmission delay is experienced every time the packet is retransmitted
• the transmission delay is less on faster (high bandwidth) links
• the transmission delay is maximum on the link with lowest capacity
• for example, 8/56th of a second (14ms) for a 1-KB packet on a 56K modem
• for example, 10,000/100,000,000 of a second (100us) for a 1250-byte packet on a 100Mb/s ethernet

Processing Delay

• when a system (host or router) receives a packet, it must do something with it:
  • error checking: verify that the packet has no bit errors, no truncation, etc
  • forwarding: look up the destination in the routing or forwarding table, assigning it to the proper queue
  • other bookkeeping and overhead: incrementing internal counters, managing memory, running routing protocols, etc
• the processing delay is the time from receiving the last bit of the packet, to the first of:
  • delivering a packet to the application
  • beginning to send the packet on the outgoing link
  • putting the packet in a queue
• typically on the order of microseconds

Propagation Delay

• this is the time between when a bit is placed on a medium, and when that bit is received
• the time is equal to the physical length of the medium divided by the speed of light on the medium
• the physical length may be more than the shortest distance, but never less
• the speed of light on the medium may be less than the speed of light in vacuum, but never more
• for example, Hawaii to California, 3800km, takes 12.7ms at the speed of light in vacuum (radio waves)
• the same distance takes 19ms at a speed of 2/3c (e.g. on an optic fiber)

Queueing Delay

• queueing delays depend entirely on the traffic being carried
• at each node, the queueing delay is the the time that a packet waits to be sent
• queueing delays can vary a lot, but are rarely more than a hundred ms or so per node.
• queueing vary over time, reflecting the amount of traffic encountered by a packet
• the lowest delay measured between two endpoints gives an estimate for the total delay minus the queueing delay
• as traffic intensity (percentage of traffic carried over a given link bandwidth) increases, queueing delay tends to increase more than linearly

Total Delay

• the total delay is the sum of the delays at each node
• in the case of a homogeneous network with no queueing delay, it is simply N times the delay at each node
• sometimes, one link will have sufficiently low bandwidth that
  • it often ends up having substantial queueing delays
  • in any case, its transmission delay dominates the uncongested end-to-end delay
• such a link is known as a bottleneck link
• other delays can be identified, for example, the medium access delay on an Ethernet or WiFi network

Packet Loss

• as the queueing delay increases, eventually
  • the system has no more physical memory to store incoming packets, or
  • the system designer or administrator has made a decision about maximum queue length
• once the queue is filled, newly received packets are discarded

Throughput

• a network has a well-defined capacity, which is the maximum possible number of bits it can carry in a second
• the number of bits per second actually transmitted through a network is the network througput
• the number seconds used for the measurement is significant:
  • if the measurement is taken over a very short time interval, we are measuring the instantaneous throughput
  • if the measurement is taken over a long time interval, for example, the transfer of a whole group of files, we are measuring the average throughput
  • average throughput is usually more consistent than instantaneous throughput
• some applications, for example VoIP, require a minimum throughput to work well
• many applications benefit from high throughput, but work even with low throughput. For example, the web.
• the bottleneck link puts an upper limit on the througput of a connection

Protocol Layers

• in a layered design, the protocol at each layer
  • logically communicates with the protocol implementation at the same layer in a different node
  • actually provides a service to the protocol in the layer above it (on the same node)
  • and uses the service provided by the protocol in the layer in the layer below it (on the same node)
• the top layer is the application protocol: HTTP, email (SMTP, IMAP), ssh, DNS
• the lowest layer is the physical layer protocol, which define ways of encoding the bits using electromagnetic signals, and the actual physical medium used

Internet layers

• much of the functionality is in the application layer
• the application layer uses a transport protocol:
  • TCP provides reliable transmission
  • UDP provides timely transmission
• the protocol at the transport layer uses the functions of IP, the Internet Protocol, which is at the networking layer
• IP provides best-effort end-to-end packet delivery
• the networking layer uses one of many possible link-layer protocols to deliver data across a single network
• the link-layer protocol usually provides Medium Access Control, and so is also known as MAC layer
• the link-layer protocol is layered above the physical layer protocol, which knows how to send and receive bits

OSI layers

• the OSI standard has two additional layers between the application and the transport layer:
  • the presentation layer provides functions to represent the data in different ways, e.g. compression or image encodings
  • the session layer provides functions such as login and checkpointing
• in the internet, the application is responsible for these functions,
• except TCP connections can be seen as being in the session layer

layers and packets

• the structure of the data that is sent reflects the layering of the protocols
• when sending a packet, each layer adds a header (some layers also add a trailer)
• an application-level message is joined with a transport-level header to become a segment (one message may be sent as many segments)
• a transport-level segment is joined with a network-level header to become a datagram (one segment could result in many datagrams)
• a network-level datagram is joined with a link-level header (and, usually, trailer) to become a frame
• when a packet is received, the headers (and trailers) are stripped one at a time as the packet moves up the stack

Layers in nodes

• generally, a packet being processed by a node only traverses the layers needed for processing
• in an end-system, a packet would traverse all layers
• in a router, a packet would only traverse the physical, link, and network layers:
  • at the network layer, the network header (IP header) is examined to find the destination address
  • the destination address is used to index the routing table
  • the packet then traverses the outgoing network, link, and physical layers to be sent from the router
• in a switch, a packet would only traverse the physical and link layers, and the link-layer address is used for forwarding