Summary of Network Related Terms and Concepts

Networking Concepts

• OSI seven layer network model - understand intuitively what each layer does....especially the physical, link, network, and transport layers.
  • TCP/IP- usually refers to the network, transport and application layers. It is assumed that the network layer can work with any link layer over any top of physical network.
• Addressing/Identifiers-
  • Application level name spaces -
    • Example: web browsing assumes objects named by Uniform Resource Locator (e.g., http://www.cs.clemson.edu/~jmarty/index.html)
  • Unique network host names
    • Human readable: domain names
    • Network addresses: (IPv4 or IPv6 defined)
    • The Domain Name System (DNS) provides a service to dynamically learn the IP address of a Domain Name (or the reverse!!).
      
      • DNS uses a distributed database that contains the names/address binding
      • UDPEchoClient2.c - the call to inet_addr is passed a string that is either a domain name (koala2.cs.clemson.edu) or a IP address in dotted notation (130.127.48.77). Inet_addr converts an address in dotted notation to a sin_addr (32 bit number for IPv4) or a -1 if an error occurs....an error will occur if the serverIP is a domain name. It then calls DNS (gethostbyname) to resolve the domain name.
      • /* Construct the server address structure */
      • memset(&serverAddress, 0, sizeof(serverAddress));
      • /* Zero out structure */ serverAddress.sin_family = AF_INET; /* Internet addr family */
      • serverAddress.sin_addr.s_addr = (in_addr_t)inet_addr(serverIP); /* Server IP address */
      • /* If user gave a dotted decimal address, we need to resolve it */
      • if (serverAddress.sin_addr.s_addr == -1) {
        • thehost = gethostbyname(serverIP);
        • serverAddress.sin_addr.s_addr = *((unsigned long *) thehost->h_addr_list[0]);
      • }
      • serverAddress.sin_port = htons(serverPort); /* Server port */
      • whois : user interface to the distributed DNS database
        • See ..IntroToTheInternet page.
  • Machine addresses
    • computers that communicate directly on a network - requires machine addresses
      • Think MAC headers ---> Section 6.4.1 in kurose (slides 6-9) of datacom102.
      • The address resolution protocol (ARP) is used to automate learning the binding between a MAC address and a Network address.
        • ARP- client software called the resolver issues a broadcast to learn the MAC address of a specific destination Host (given a destination IP address)
        • UDPEchoClient2.c : The very first sendto() will cause an ARP exchange.
• Error detection - applicable at multiple layers. Errors might be bit errors within a frame or a message, or frames dropped in the network. The latter might be due to poor channel conditions prevent any port of a transmission from being received, or due to a congested router which might drop packets if they arrive at a router with a full queue.
• Error recovery -For our purposes, this simply means the receiver conveys to the sender a request to retransmit. Or, the sender learns of an error implicitly when a timeout occurs.
• Reliable Data Transfer Protocol - See Kurose, chapter 3.4. A RDP provides a service abstraction to applications- a reliable data service. This involves both error detection AND recovery.

Terms related to the performance and control of a network

• Performance metric: Conceptually, a performance metric is something (usually a quantitative result) that reflects how well the system (or the specific thing under study) is performing. It's important for performance metrics to be precisely identified/described so that they can be repeated by others in any similar system.
  • Examples:
  • Vehicles: fuel efficiency, reliability measures
  • The US postal system: average time to deliver mail, average number of items lost per day
  • Telephone networks: call quality (note....might be hard to nail this quantitatively as human perception comes into play), rate of call drops, rate of failed call attempts
  • Packet switched network:
    • Latency: the time it takes for data to be sent from one location to another.
    • Round trip time (RTT) - When it makes sense, we might want to assess the time it takes for one computer to send a message to a partner until when the desired response is received.
    • Loss rate: When sending packets from one computer to another over a network, it is possible that not all packets arrive at the receiver. The loss rate describes the ratio of lost packets (number lost / total attempts )
    • Rate at which data is transferred. There is a family of terms that fall in this general metric category. Examples include:
      • Bandwidth - still a general term. Let's define based on a EE or a CS background:
        • An electrical engineer will think of bandwidth at the physical medium layer. A ‘channel bandwidth’ might refer to the ‘analog bandwidth’. This corresponds to the ‘spectrum range’ of the communications channel. So it is the difference between the highest and lowest frequencies that can operate over the channel. Another way to state this is that the channel bandwidth is the fastest continuously oscillating signal that can be sent over the channel. It is measured in cycles per second or Hertz (hz). The baud rate is a measure of how the communications method actually uses the channel bandwidth. Therefore, the baud rate is the transmitter’s maximum signal rate. The baud rate will be at most the channel bandwidth. However, to ensure robust communications over possibly noisy channel conditions, the baud rate is likely lower than the channel bandwidth. A measure of the efficiency of a communications system is how close the baud rate comes to the channel bandwidth.
          • The baud rate along with the number of bits that are ‘encoded’ in each information symbol determines the data rate.
        • Data rate = baud * log2(#signal levels) (eq. 6.1 in the text).
      • A Computer Engineer or a CS (software engineer) might thinkof Bandwidth in one of several ways....depending on the layer or boundary of the system of interest. Note the units differ depending on the specifics...
        • In physical layer communications, the term bandwidth relates to the spectral width of electromagnetic signals or to the propagation characteristics of communication systems. In the context of data networks, the term bandwidth quantifies the data rate that a network link or a network path can transfer.
        • Other variants
          • Effective data rate: the physical layer data rate adjusted for the amount of physical layer overhead. •
          • Link rate: The rate at which a link layer protocol can send and receive data from higher layers. This takes into account the link layer overhead. •
          • Network bandwidth : the rate that a network can send and receive data. •
          • Available bandwidth : A network (large or small) might not be able to provide to a particular user all of capacity as there might be congestion, channel errors, or network service provider policy limits. At a given time instant, the amount of capacity that can be allocated to a user is referred to as the available bandwidth.
          • Throughput: This usually refers to the rate at which data can be transferred between two endpoints. It might assume transport layer overhead (or not).
          • Goodput: This usually implies the rate at which application data (excluding all overhead) is sent over a socket. The units usually assumed are bits per sec (bps).
• Flow control - Networking mechanism that allows a receiving side to indicate to the sending side to slow down. It is usually referred to as a speed-matching mechanism. It might be implemented at different layers - hop-by-hop, network entry to senders, network exit (egress) to entry (ingress), or end-to-end.
  • TCP's congestion control blends end-to-end flow control with network congestion control.
• Congestion control - Network congestion results from too much demand for network bandwidth- more precisely, demand exceeds capacity.
  • See slide 5 datacom102.
  • Network congestion typically means that one or more routers along a path has a queue of packets waiting for their turn to be transmit over the network interface.
    • Measurables: Increased latency and non-zero packet loss
  • All Hosts and Routers will allow packets to be queued if the output interface (the Network Interface Card, NIC) is busy.
    • Simple problem: h1---->R1-----h2
      • H1 sending at a rate of 1 Mbps, the h1-R1 link rate is 100Mbps, R1-h2 is 2 Mbps.
      • Let's say that for a period of 5 seconds, H1 sends at a rate of 10 Mbps.
      • At the end of the 5 second period, how many packets are queued?
        • Model this as: H1---> R1 sends at a rate of 8 Mbps (10Mbps-2Mbps) which represents the rate at which data is queued
        • X(amount Queued) = ((8*10000000 bps * 5 seconds ) / 8 bits/byte) = 50Mbytes
          • If packets sizes are 1000 bytes, this is a queue level of 50000 packets!!
        • What is the queue delay that packets experience if the queue level is 50000 packets (and if the 'drain rate' of the queue is 2 Mbps)?
          • 50Mbytes*8)/2 Mbps) = 200 seconds!
  • Hosts/Routers allow the maximum queue capacity to be set through configuration. At some point, the router will start dropping packets that arrive to a full queue.
  • End-to-end congestion control - responds to explicit or implicit congestion notifications from the network. The sending hosts are expected to 'react' appropriately - TCP congestion control defines the correct reaction for hosts in an Internet.
    • Explicit- the network sends a specific congestion indication message to sending hosts.
    • Implicit - the network does nothing in particular other than dropping packets once queue levels reach capacity. The end points can assume network congestion when either packet loss is observed or evidence of sustained queuing delay is observed.
    • The Internet method is based on TCP congestion control which uses implicit indications (packet loss) to tell sending hosts the network is congested and to slow down.
      • Basic TCP behavior: it reduces it's send window significantly by either reducing it by a factor or 2 OR by setting the window all the way down to 1 packet.
        

Application communication models

• client/server
  • client
  • server
  • interactive (sequential) server
  • concurrent server
• Peer-to-peer - Any node can be a client or server. Think bitTorrent.
• Broadcast oriented
  • Broadcast (one to all) on a single network is easy - use the all one's address (try ping 255.255.255.255 )
  • Multicast (one to a group) - think content providers using IP Video based on UDP/multicast.
  • Modern equivalent is HTTP-based Adaptive Streaming (HAS) - it involves unicast sessions to the end points. But it likely multicasts content out to edge content servers.
• Publish/subscribe - An old model but it's now relevant due to Internet of Things. Devices can publish data- well suited for power constrained devices.

Terms :

• Elastic application - an application that is 'OK' with whatever bandwidth it obtains from a network. Examples are file transfer or to a lesser degree, web browsing
• Traffic model - when analyzing systems, we need to characterize the network traffic.
  • Some applications are easy to model:
    • File Transfer - an 'always on ' sender
    • Ping - on/off for timescales of packet times
    • web browser - on/off over timescales of seconds
    • VoIP - on/off over timescales of 'talking' times
    • Network games (like call of duty) - on/off at the timescale of player input and output changes (typically based on default time intervals)
    • Internet Video (HTTP-based Adaptive Streaming) - on/off based on 'chunks of video' downloaded (although the download rate of chunks can change)
• Application level performance metrics
  • RTT's appropriate for web browsing and ping applications
  • Throughput appropriate for FTP and HAS video
  • One-way latency and loss appropriate for VoIP or network games

Data Transmission

• Terms
  • bandwidth
  • throughput
  • propagation dealy
  • full-duplex
• What's the general idea of Nyquist's theorem?
• List three different types of physical media.

Packet Transmission

• Terms
  • packet
  • multiplexing
  • transmission errors
  • frame
  • Ethernet
  • collision and CSMA/CD
  • Maximum transmission Unit (MTU)
• Frame versus Packet versus Datagram versus Message
  

TCP/IP Concepts (we did not cover these topics in detail....I won't ask specific questions on these topics but you should be familiar with what they roughly correspond to)

• IP addressing
• Classful address mechanism - given an address like 130.127.49.1, is it Class A, B, or C?
  • Difference between Class A,B,C
  • Intent of classful address
    
• broadcast addresses
• multicast addresses
• Private addresses
  
• IP
• ARP
• Datagram formats
• Forwarding
• Fragmentation
• UDP/TCP
• Conceptually what do these layers do?
• What does an Ethernet MAC frame look like that contains a IP Packet that contains a UDP message?
  

Socket Programming

• What is a socket? What is it used for?
• The following code is from our UDPEcho client
  • clientAddressSize = sizeof(clientAddress);
  • alarm(TIMEOUT); //set the timeout
  • /* Recv a response RTT_MODE , client address/port will be filled in on return */
  • recvMsgSize = recvfrom(sock, echoBuffer, MAX_DATA_BUFFER, 0, (struct sockaddr *) &clientAddress, &clientAddressSize);
  • alarm(0); //clear the timeout
  • You should be prepared for questions such as:
    • on the recvfrom, why is a pointer to clientAddress provided? What are we exepting the system to provide in clientAddress?
      • Answer: a recvfrom pulls the next IP datagram from the sockets Rx buffer that was targeted (dst IP/port to this socket endpoint) to process. The system copies the data from the IP datagram (if any) to the caller's echoBuffer), and returns the number copied in recvMsgSize
    • How does the client know if the recvfrom returns but something abnormal might have happened?
      • The recvMsgSize return code is '-1' (the standard Unix system call error indication), and then details about the error are in the global variable errno.
        
• Related to UDPEcho and messages.c: Let's say our client sends a DataMessage with 100 bytes of application data. Draw a rough picture of the MAC frame that gets sent over the wire?

Probability and Stats topics

• Review concepts including
  • sample spaces
  • events
  • probability of events based on counting techniques
    • From a freshly shuffled deck of cars, you are dealt 2 cards (from a deck of 52). What is the probability that you are dealt a blackjack?
  • established probabiltiy models
    • Continuous - a uniform distribution
    • Discrete - You roll two die.What is the probability you roll two 6's?
  • multiple, identical experiments
  • random variable
    • sample space of the random variable
    • standard RVs :
      • Continuous - a uniform RV
      • Discrete - roll two die. Define a RV X to represent the sum of both dice. Find the probability that X is greater than 12.
    • Expected value, E[X]
  • measurements and experimental statistics
    • Time series
    • Independent samples (vs correlated)
    • Sample mean, sample variance
    • Visualizing time series data
      • histograms and sample distribution plots

last updated: 3/11/2019