Review

data Link layer: Aloha, Ethernet, switching, and 802.11
network layer: IP and routing
transport layer: TCP and reliable transmission (and flow and congestion control)
application layer: HTTP, DNS, bittorrent, and email
networking principles: performance, error detection, layering, security

data link layer

responsible for transferring data between directly connected systems
"directly connected" extended to mean "connected by a network using only link-layer headers" (circular definition)
MAC layer is concerned with Medium Access Control, framing, and identifying interfaces (MAC addresses)
Ethernet switches learn on which port they receive data from a sender
and only forward data to that sender on that interface
forwarding tables are similar to routing tables, but are managed automatically
all other packets are broadcast
loops in the network can cause broadcast storms, unless a spanning tree protocol is in use

Aloha

two frequencies (channels), one from end systems to hub, one for the hub to broadcast to the end systems
end systems transmit when ready
if hub broadcasts packet, assume all have received it
otherwise, retransmit
low efficiency, at best about 18% (36% for slotted Aloha)
so only used when end system is unable to do carrier sensing

Ethernet

header has destination address, source address, ethernet type (x800 for IP packets, x806 for ARP packets)
Carrier Sense Multiple Access with Collision Detection (CSMA/CD): interface listens for carrier before transmitting, and can detect collisions
reliable, in-order delivery of packets if the transmission does not cause more than 15 collisions
binary exponential backoff for the first 10 transmissions
Ethernet addresses are globally unique, used also for other systems, e.g. 802.11
Ethernet addresses have a 24-bit OUID and a 24-bit part assigned by the organization
range of speeds, so far, 10Mb/s, 100Mb/s, 1Gb/s, 10GB/s
"plug and play"

802.11

designed as a wireless equivalent to Ethernet
Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA):
Request to Send (RTS)/Clear to Send (CTS) exchange to reserve the medium around both sender and receiver
ACK to confirm receipt
small data and broadcast data sent without RTS/CTS
ESSID identifies network to connect to, not necessarily the individual access point
802.11 header provides up to 4 addresses, to keep track of original source and destination while intermediate hosts are relaying on a wired or wireless mesh network
encryption provided by WEP (very weak), WPA, or WPA2

Network Layer

separate networks, identified by network part of the address
network part of the address defined by netmask (CIDR) or class
packet forwarding based on routing table
routing protocol builds and maintains routing table
end-to-end protocol requires participation from all intermediate routers (or overlay network)

IP

connectionless, best-effort, lightweight, "stateless" forwarding
caching improves performance: ARP cache, route cache
end-to-end: source address, destination address, protocol number (1 for ICMP, 6 for TCP, 17 for UDP)
hop-by-hop: TTL, checksum
throughout: header length, version, total length
fragmentation: packet ID, fragment offset, More Fragments and Don't Fragment flags
IPv6: larger addresses (128 bits, 16 bytes), fewer header fields, constant header size, optional functionality shifted to extension headers
multicast: leaf-initiated join, soft state is remembered as long as it is refreshed frequently enough

Routing

link-state (and flooding): OSPF
distance-vector: RIP, variant in BGP
Routing Information Protocol: Distance Vector with Split Horizon and Poisoned Reverse, infinity is 16 hops
Open Shortest Path First: Link State with flooding, areas used to decrease the amount of information being distributed, ways of distributing summary information. Backbone area connects to all other areas
BGP: path-vector allows policy decisions, cost considers each Autonomous System as one hop
one Exterior Gateway Protocol (BGP), two Interior Gateway Protocols (IGPs)

Transport Layer

assumes existence of unreliable, end-to-end network layer protocol
demultiplexing to identify applications or services (TCP/UDP port numbers)
reliable transmission can only be done end-to-end
stop-and-wait protocol
window-based flow control
congestion control: assume data loss means congestion (safe assumption even when not accurate) and Multiplicatively Decrease, while Additively Increasing when no loss, AIMD
UDP: source and destination port, length, checksum

TCP

source and destination port, 32-bit sequence and acknowledgement, header length, flags, window, checksum, urgent pointer
state machine uses flags to keep track of three-way handshake and closing exchange
SYN to open, FIN to close (one side), RST to terminate, ACK to confirm that the acknowledgment field is valid
cumulative acks acknowledge all that has gone before
three duplicate acks interpreted as a NAK
Selective Acknowledgement (SACK) TCP header option
window scaling options
carefully tuned adaptive retransmission timer
TCP is generally optimized for high performance, except for congestion control

Application Layer

many, many protocols
have to solve all issues left unsolved at the lower layers, e.g. security
sometimes by providing an overlay network
client-server or peer-to-peer architecture, or a mix
functionality, even debuggability is more important than efficiency, e.g. ASCII-based encodings
generally interactive (telnet, ssh) or batch transfer (everything else), though now also hybrid (e.g. AJAX)

HTTP

client/server, ASCII encoded request and reply protocol over TCP
header followed by empty line possibly followed by content, and perhaps additional headers
main HTTP request methods: GET, HEAD, POST
response shows status code and readable explanation (similar to FTP)
many fields allow much flexibility, e.g. languages, multimedia formats, many different kinds of content
cookies allow server to store state on client
many applications built on top of this basic web mechanism
URL includes DNS name, and system-dependent path

DNS

Domain Name Service maps names to IP addresses
request-reply client/server protocol, usually over UDP
domain name is made up of labels separated by dots
DNS and routing are essential for the functionality of the web and email
record types include A, MX, NS
recursive queries on behalf of requester, iterative queries: "go ask X"

Peer-to-peer data exchanges

Bit-Torrent: data transfers from the "best" peer that has the data
downloader becomes available as source
data split into chunks, each chunk can be downloaded independently
centralized trackers keep track of which node has which chunks

Electronic Mail

distributed system using TCP, neither purely client-server nor purely P2P
transfer of text, and other media encoded using text
email address includes system-dependent name @ DNS name
DNS MX records often different from DNS A record

Networking Principles: Performance

overall time to complete an operation matters to the user
for some operations, latency is most important
latency usually measured as the (average, minimum, or maximum) round-trip time of small packets (lower latency than large packets)
for other operations, throughput is most important
latency measured as the maximum data per second that can be sent, usually in large packets (lower overhead than small packets)
caching pervasive, makes testing harder
DNS resolution can be a substantial component of a web access

Networking Principles: Error Detection

parity, checksums, CRCs
mathematically define which error patterns are detected
efficient computation of error check code
efficient verification of error check code
forward error correction: n-bit-error detection, m-bit-error correction, n > m

Networking Principles: Security

encryption, authentication, integrity, intrusion prevention
symmetric key protocols such as AES, fast, but require key management
public-key protocols such as RSA, slower, but only require distribution of public keys
SSL/TLS, WEP/WPA/WPA2, IPSec, SSH

the future

when do you expect to see IPv6 deployed?
when do you expect to be able to web surf at Waikiki beach? Other beaches?
which of facebook, youtube, twitter, and second life is your favorite?
what is the next hot application?
how many computers do you expect to have simultaneously connected to the Internet at some point in the next 10 years?