Book Cover
Computer Networking

:star: :star: :star: :star: :star:



  • Book PDFs:
  • More:
  • Slides:

1: The big picture: layers of protocols

K&R Ch 1: Computer Networks and the Internet

  • Start at the Internet and go down into the links and switches
  • After survey of implementations, look at things more abstractly - models for throughput, delay, transmission, propagation, queueing, architecting, etc.

  • Nuts and bolts of Internet
  • Internet is a network that interconnects a bunch of devices all over the world
  • Each device is a host
  • Connections made by communication links (cables, wires, fibers, etc., each with different transmission rates) and packet switches (takes a packet arriving on communication link and forwards packet on an outgoing communication link)
  • Commonest packet switchers are routers (in network core) and link-layer switches (in access networks)
  • Sequence of communication links and packet switches from start to end is the route through the network
  • Mental model: similar to networks of highways, roads, intersections. Packets are trucks, links are highways/roads, switches are intersections, end systems are buildings
  • Each ISP (internet service provider) is a network of switches and links. Conform to conventions so that they can all access internet similar way
  • Protocols control send/receive info. TCP (Transmission Control) and IP (Internet) are very important. Together, TCP/IP
  • Standards come from the IETF (Internet Engineering Task Force), whose RFCs define protocols and stuff

  • Services of Internet
  • It’s an infrastructure that provides services to applications like web apps, social networks, IMs, etc.
  • How do these apps communicate? By following an API (Application Programming Interface). Set of rules that sending program mnust follow so that the Internet can deliver data to the destination program
    • Postal service API: stamp, from address, return address, post office, etc.
  • Protocols
  • Human analogy: when meeting someone, “Hi how are you my name is Tiger”, etc etc.
  • There are specific messages we send and actions we take in response to the received reply messages or other events (like no reply)
  • Human analogy: “Got a question?” “Yes” “Yay, what is it?”
  • All Internet activity is governed by a protocol. Protocol defines the format and order of messages exchanged between two or more communicating entities, as well as the actions taken on the transmission and/or receipt of a message or other event
  • Mastering networking === understanding what/why/how of protocols

  • What’s on the edge of the network? Hosts and what they immediately talk to
  • Hosts can be clients (ask for stuff) and servers (return stuff)
  • Access network: network that physically connects edge router (first router) on a path from one host to a distant host. Can be home network, mobile network, ISP network, or enterprise network
  • DSL (digital subscriber line) piggybacks off the telephone company’s access, at different frequencies
  • Cable internet access piggybacks off of the cable TV company’s infrastructure. Bit faster. Shared network, so it’s slow if everyone’s downloading video files at the same time
  • Ethernet is a switch that is custom and connected to internet. Common for corporate/university/home networks
  • Finally, wireless LAN connects to access point that is in turn connected to the wired internet. Based on 802.11 tech, more colloquially known as WiFi
  • Wide-area access in the form of 3G and LTE. Use same tech as cell phones to send/receive packets through cell towers

  • Overview of physical media
  • Whenever you go from transmitter to receiver, you go through physical medium
  • Guided media: waves guided along solid medium. Fiber, twister-pair copper wire, coaxial cable
  • Unguided media: waves propagate in the atmosphere and outer space, such as in wireless communication
  • Cost of materials dwarfed by cost of labor, so in buildings you’ll lay down all three of twisted pair, fiber, and coaxial just in case another gets used in the future
  • Twister-pair copper: least expensive, most common, used by phones. Up to 10Gbps
  • Coaxial: used in cable TV. Shared medium
  • Fiber optics: up to hundreds of gigabits per second. Immune to electromagnetic interference, used for overseas links and stuff. Cost a lot tho
  • Satellites and terrestrial radio channels don’t require physical wire to be installed, can receive signals over the air

  • Okay, now the core of the network
  • Packet switching
  • Message broken down by source into chunks called packets
  • Packets transmitted as fast as the communication link can do it
  • store-and-forward: packet switch waits until it has the entire packet before it begins transmitting it
  • Each switch has an output buffer for each attached link. This stores packets while the link is busy. Output buffer time is a queuing delay
  • If you get new packet and output buffer is full, oops you got packet loss
  • How does router know which link to forward packet onto? Each router has a forwarding table that maps destination addresses (or portions of them) to the outbound links. When packet arrives, router examines the destination and forwards to correct outbound link
    • Analogy: car driver that keeps asking for directions of more and more specific parts of his destination address (Florida -> Orlando -> 145 Lakeside Dr)
  • How are forwarding tables set? Using the Internet’s standard routing protocols. Use shortest path or something
  • Packet switching is how internet operates

  • Circuit switching e.g. for telephone lines. Analogous to making a reservation at restaurant so you have dedicated space (vs. packet switching using available resources)
  • Packet switchers say it makes better use of resources and is easier to implement
    • Way more resource efficient!
  • Packet switchers say circuit switching is wasteful during dead periods
  • Circuit switchers say packet switchers can’t do real time stuff b/c it’s unreliable

  • Where are we now? The internet is a network of networks. Gotta connect everyone to everyone else!
  • Build up from the bottom!
  • Consists of a dozen or so tier-1 ISPs and hundreds of thousands of lower-tier ISPs. They have diverse coverage all over the globe. Low tier ISPs connect to higher tier ISPs, which interconnect to one another. Users are customers of lower-tier ISPs, who are customers of higher-tier ISPs. Sometimes the Googles of the world also create their own networks and connect to lower-tier ISPs

  • Delay/loss/throughput in packet switching :’(
  • Constraints make it so that we can’t always have the throughput we want

  • How do you get delayed?
  • Nodal processing: time to examine packet header and determine where to direct packet, error checking. Microseconds. Negligible
  • Queuing: wait for space on the outbound link. Micro to milliseconds. If traffic arrive much faster than output speed, this is large.
    • Traffic intensity: average rate at which bits arrive divided by the transmission rate
    • Don’t have traffic intensity greater than 1! Exponential curve as traffic intensity approaches 1 too, so be carefuls
    • Higher traffic intensity = packet loss = sad face
  • Transmission: time to push all packet’s bits from switch onto link. Micro to milliseconds. Can be significant if slow link
  • Propagation: time to move along the link. Supertiny, speed of light often? if a long area though, can be milliseconds. Longer for further away stuff
  • All of these make up the total nodal delay. Get a feel for how much each of them affect things
  • Total end-to-end delay is nodal delay for allll of the nodes in the path

  • Traceroute used to see your network path. Sends special packets that get a return message from every router along the way, so you can build up the path you took to get to your destination

  • Some stuff on network constraints. Today, it’s generally the access network that creates a bottleneck because of lower capacity

  • Okay, lots of words

  • Protocol layers and service models. Important!
  • Layers is best way to conceptualize complex systems
  • Analogy of traveling by plane
  • Layers are ticket, baggage, gate, takeoff/landing, and airplane routing. Each layer implements some functionality or service. Each layer has stuff happen below it that brings it back to that layer
  • Okay, now bring back to networking
  • Each network protocol belongs to a layer. What’s important is the service model of a layer: what service does the layer offer to the layer above it?
  • Layer performs service via actions and using services of layer below it

  • Five layers: Application > Transport > Network > Link > Physical
  • Application and transport in softwaare, link and physical in hardware, network is mixed
  • All of these layers are distributed across hosts. One piece of each of these layers found in hosts
  • Together, this is the protocol stack
  • Application: HTTP/SMTP/FTP. Distributed over multiple hosts, protocol used to exchange messages with application layer in another host
  • Transport: transport application-layer messages between endpoints. TCP and UDP. A transport-layer packet is a segment. TCP important for reliability and sequencing and throttling. Doesn’t solve security. UDP has no reliability or sequencing or throttling. UDP is basically IP with the ability to address ports. UDP used for stuff like streaming games and video
  • Network: move network-layer packets (datagrams) from one host to another. Takes a transport-layer segment and a destination address from transport layer, then provides the service of delivering segment to transport layer in the destination host. Generally uses IP here
  • Link: routes datagram through routers between source and destination. At each node, network layer passes datagram to link layer, which delivers datagram to the next node along the route. They differ - could be Ethernet, WiFi, custom cable protocol. Link layer packets are frames
  • Physical: move individual bits within a link layer frame from one node to the next. Dependent on which link layer is used. Depend on the actual physical medium as well. E.g. ethernet has a protocol for each physical layer - copper wire, coax, fiber, etc.
  • OSI model has two additional layers: presentation and session. If an app developer needs these for the Internet, they gotta build them into the application

  • Encapsulation is important. Applicatio-layer message sent to transport layer, which adds some information so it becomes a transport-layer segment. Thus, the segment encapsulates the message. Datagram encapsulates segment, and frame encapsulates datagram. Each layer adds its own headers

  • Shpeel on some security stuff. What can go wrong?
  • Malware can come across the Internet and do bad stuff to our device
  • Maybe it becomes part of a botnet, which is a bunch of devices controlled by bad guys to send spam or something
  • Viruses require some kind of user interaction to infect device. Worms can get in without user interaction (e.g. you’re on a bad network and stuff gets in the side door)
  • DDos: make a host unusable by legitimate users by slamming it with traffic
    • Vulnerability: send carefully-formed messages to a host to make it crash
    • Bandwidth flooding: send a shitt on of traffic so the host can’t keep up
    • Connection flooding: lots of half-open or slow TCP connections established, host too bogged down to accept real traffic
  • Packet sniffers can listen on your router or something and copies all the packets that flow through. Can deconstruct those later to see if there’s anything useful
  • Spooky

  • Okay time for some history
  • Packet switching as an alternative to circuit switching came up in the 60s. ARPANet also conceived with packet switchers. By 1972 ARPAnet had 15 nodes and was demo’d. It was a single, closed network
  • Now in the 70s lots more networks from lots more places
  • TCP, UDP, IP concepts in place by end of 70s. Ethernet protocol as well
  • In the 80s, more stuff developed, ARPANet grew. French Minitel was a public packet-switched network that a bunch of French people had and that had a decent amount of good stuff
  • Internet explodes in the 90s. Invented by da god Tim Berners-Lee. Developed HTML, HTTP, web server, and browser. We get email, the web, IMs, p2p file share. A ton of shit now of course

  • Summary time
  • Lots of stuff in the Internet
  • Edge of the network: end systems, applications, and the service provided to those applications
    • Link and physical media in the access network: routers, cables, and such
  • Now in the core, we’ve got different switches and the network-of-networks concept
  • Then we got to delays/throughput/packet loss
  • Key architectural principles of protocol layering and service models


  • OS abstractions for communication
  • File: file descriptor, basic read/write/offset functionality. Permanent storage. Seekable - you can jump around at different offsets
  • Pipe: returns two file descriptors, one for reading and one for writing. When you write to one the bytes are available for reading on the other end. No offset, once you read stuff it’s gone
  • Socket: Unix Domain Socket. Server process can bind to a socket address (e.g. /tmp/postgres.5123). Now clients can interact with that socket, and they get a r/w file descriptor to interface with
    • Like a phone call, cool
    • Does NOT specify connector vs. listener as client vs. server
    • The connector is usually the client, but doesn’t have to be. It can make a connection and then serve requests. K. Connector/listener are fundamental roles, client/server are “technical”, implemented roles
  • More history
  • Military needs to connect stuff better, relay stuff

  • Why is there a need for layers above IP? No security, no reliability

  • Kernel is between the application and transport layers
  • SSL sits between application and transport, but don’t write your own. Use a reliable one instead

  • Use file to get file info for binary

2 - Application Layer: HTTP

K&R 2.1 - 2.2

  • See slides


  • Why is text bad?
  • Parsing sucks
  • Wasted space
  • How do you transmit non-text??
    • Content-Type: octet-stream
    • Base64 encoding
  • Upside: human-parseable. -_-

  • Alternatives to request/response (HTTP model)
  • push/pull: client pushes messages onto a stream, many hosts pull from the stream
  • pub/sub: server publishes to a stream where the client is subscribed and can pull stuff off

  • Why do you need to include the Host header? Isn’t it redundant?
  • Two ppl sharing a webserver and IP address

3 - Application Layer: DNS, FTP, SMTP, etc.

K&R 2.3 - 2.5

  • See slides


  • What does DNS solve? People have a bunch of addresses but the addresses keep changing (they’re treated like home addresses), so we need a way to map names onto addresses
  • What is DNS? It’s a protocol for a distributed database
  • No forced cache invalidation is a tradeoff
  • Feature: extensible record types (hierarchical)
  • Who has authority?

  • There’s one root DB, but it pretty much just has NS records for the TLD’s

  • Dig is for looking at DNS

  • DNS requests happen in parallel with HTTP. Uses UDP because everything fits in one packet

4 - Transport Layer: TCP and UDP

K&R Ch 3

  • See slides


  • UDP always needs to fill source port field

  • SYN flood by forcing server to allocate a ton of resources that’ll never be used
  • Fix with Syn cookie - on initial SYN, don’t allocate anything, make a cookie. This cookie needs to come on the next ACK (since it’s on the SYN of the SYN ACK response) or else you won’t actually allocate resources

5 - Network Layer: IP

K&R Ch 4

  • See slides


  • How does a router’s forwarding table get established?
  • Procedures convert routing algorithm into forwarding table at each router
  • Why have a “transport protocol layer” in the IP datagram header? So you know which code to jump to after parsing out the IP header!
  • Risk-based problem-solving: where’s the biggest risk? Can I mitigate that first? If I can’t, there’s no point in continuing

K&R Ch 5

  • See slides


  • Didn’t go over much, there was a wireshark lab. See mostly stuff in slides. This is a protocol that operates over hardware /shrug

7 - Multimedia streaming and realtime communication

K&R Ch 7

  • See slides


  • Switching adds less noise to hubs, because not everything is broadcast
  • MAC address can identify a node as being on a specific VLAN
  • VLAN makes it even quieter, need specific translators between VLANs though
  • VPN tries to interpose one network over a set of other subnets, okay
  • If you TCP connect to a VPN, you can send requests from that VPN and it looks like your computer is just having one TCP connection /shrug

  • Streaming stored video time
  • TCP control channel for stuff like “start halfway through, send at this rate”, UDP channel for the actual data

8 - Security

K&R Ch 8

  • See slides


  • Everything is bolted on!
  • To brute force decrypt: helps to have a ciphertext-plaintext pair
  • Cipher block: not byte by byte, but chunk by chunk
  • Chaining: Start with just one random seed, XOR with first message block, encrypt that block, then XOR next with (first xor’d w/ random) then so on and so forth
  • So if you have symmetric encryption, how do you exchange keys?
  • Use asymmetric public/private key exchange to exchange symmetric keys for the rest of the session
  • RSA is 300x slower than AES to encrypt!

  • So now:
    • From Alice (plain)
    • Alice makes message key
    • Alice encrypts message with message key
    • Alice encrypts message key with Bob public key
    • Alice encrypts ^ with Alice public key
    • Alice hashes ^^ and encrypts the hash with her own public key and puts it at the end of the message (this is signing)
    • First thing Bob does is take the encrypted hash off the end of the payload, decrypt with Alice’s public key, and compare to hashing the payload
  • Certificate: 3rd party certificate authority gets your public key with a proof of identity. CA verifies this, and if you pass, they verify your public key by hashing it and signing it with their (super secret) private key, which becomes your certificate

9 - The Future of Networking

IPFS White Paper

  • See PDF in 09_future
  • In business, a white paper is closer to a form of marketing presentation, a tool meant to persuade customers and partners and promote a product or viewpoint

Van Jacobson on A New Way to Look at Networking

  • Network research is dead end
  • Everything is about paths

  • Phone number is a program for directing your line on a switchboard
  • How do you amortize setup cost and worry about reliability?
  • But everyone thought telephony was the only model for networking
  • But then people came along with packet switching: focus on endpoints, not paths
  • Lots of paths from one endpoint to another. Split data into packets
  • But just overlayed over telco

  • Then Stanford peeps came up with TCP/IP stack
    • To concatenate paths
  • Reliability increases exponentially with system size (more failure correction)
  • No call setup - high efficiency
  • No hot spots because topology can self adjust
  • Connecting is binary: either connected or not
  • Man I can’t take no notes, just listen

  • TCP/IP rescued us from plumbing at the wire level but we still have to do it at the data level. A dissemination based architecture would fix this
  • Many ad-hoc dissemination overlays have been created (CDN, BitTorrent, etc.) - there’s a need!


  • Packet switching had great diagrams, but how to implement?
  • At first had to piggyback on the telephone lines
    • Objection: “it’s basically the same thing”
  • Key: dissemination. Internet is designed for conversations, not for dissemination (disseminate information and stuff)
  • Bad at broadcast
  • Binary connection - 2 participants

  • Ideas: address data not machines. Sign/Auth data
  • Big buffers/caches
  • Immutable data