Networking background

Internet components and structure

The Internet is:

PoP (Point of Presence): router(s) in the provider’s network where customer ISPs can connect
multi-homing: customer ISPs may connect to more than one provider, for fault tolerance
peering: connect directly to another ISP at the same level, instead of going through a provider, usually without any cost
IXP: third-party location where ISPs can peer with each other
content provider networks: large content providers (e.g. Google) have large networks, connect directly to lower-level ISPs and IXPs

Interoperability
- each network is independent
- they interoperate due to standards, especially: IP, BGP, TCP, DNS
- a network can do whatever it wants internally
packet switching
- data is divided into packets
- all packets from all sources share each link
- each packet uses full link bandwidth
- packets are stored before being forwarded
- link never idle if some packets in the queue

circuit switching:

packet switching:

makes Internet routers very simple, with complexity at the edges
see End-to-end arguments in system design
enables the Internet to support a wide variety of applications
for a long time, circuits were considered best for voice and video but we now use Skype and watch movies over the Internet

→ never underestimate the power of lots of bandwidth and caching

rough consensus and working code
- implementations available
- design influenced by experience: performance • fluid and open standardization body (IETF)
open (rather than proprietary) architecture

layering helps to build complex systems
- split large system into smaller pieces
- identify each layer’s functionality and interfaces
- can change a layer’s implementation as long as interfaces remain the same
encapsulation vs decapsulation

→ security was not considered Brief History of the Internet

people like to use names for computers (www.byu.edu), but computers need to use numbers (128.187.16.184)
- the Domain Name System (DNS) is a distributed database providing this service
- a program sends a query a local resolver
- the resolver contacts DNS servers as needed
many DNS services
- host name to IP address translation
- host aliasing (canonical name versus alias names)
- look up mail server for a host
- load distribution - can provide a set of IP addresses for one canonical name
demonstrate: dig
domain name: top-level domain (TLD) + one or more subdomains
- example: cs.byu.edu
- may be 127 levels deep, 63 characters per label, 255 characters per name

resolver contacts root name server if it doesn’t have the mapping
each server that doesn’t know the mapping tells the local name server the identity of the next server in the hierarchy that can answer the query
any name server that learns a mapping caches it
TLD servers usually cached in a local name server, so root name server (theoretically) not visited often

best-effort service
- connectionless: no state setup
- unreliable: lost packets are not re-sent • no flow control
- no congestion control (won’t slow down, you can send as much and as fast as you want)

defined in RFC 768, from 1980
- port numbers (16 bits each)
- length of UDP segment, including header (bytes)
- checksum: detect some bit errors in transmitted segment
- application data (sometimes called ADU)
binary protocol

pipelining: multiple data segments outstanding at once
VERY important to use the high bandwidth available on the Internet
uses congestion control to slow down when the network is overloaded, converge to a roughly fair allocation of bandwidth among compeeting flows

sequence number: byte number of this segment within the byte stream
ACK number: sequence number of next byte expected from sender
- an ACK tells the sender that the receiver has EVERYTHING up to but NOT INCLUDING the byte in the ack number
- “this is the next in-order byte I am expecting”
the power of cumulative ACKs
- if an ACK is ever lost, a subsequent ACK synchronizes the state of the receiver with the sender
- no need to retransmit lost ACKs, which would get messy

source IP address (easily spoofed)
destination IP address
TTL (hop count)
protocol: typically UDP or TCP
IP addresses
- 32 bits
- dotted decimal notation
- byu.edu is 128.187.16.184
forwarding process: check the destination address
- is this one of my addresses? send to next protocol specified in IP header
- is this one of my subnets? send to link layer to forward to the destination
- do I have a route? send to link layer to forward to next IP hop
- destination unknown! send an ICMP error to the source of the IP packet

simplified
result of negotiation
traffic class: enable routers to map traffic into classes (delay, loss guarantees, etc)
flow label: uniquely identify all packets for a particular flow/application, used for QoS
Next Header: upper layer protocol or option
IPv6 addresses
- 128 bits

MAC addresses
- used to get frame from one interface to another physically-connected interface (on the same network)
- most are 48 bits long, depends on link-layer protocol
- address burned into the adapter ROM
- broadcast address usually all ones (FF-FF-FF-FF-FF-FF)
address assignment administered by IEEE
- manufacturer buys portion of MAC address space (prefix)
- uses that prefix for all MAC addresses and ensures it does not reuse the suffix
- uniqueness provides address portability

protocol that queries for IP addresses and maps to the associated MAC address
each host builds a table
- IP address
- MAC address
- TTL (e.g. 20 minutes)
if host A has no entry for IP address B in table
- A broadcasts ARP query for B
- all hosts on LAN receive query
- host with address B responds by unicast to A with its MAC address
- all hosts hear query and response, cache translations for A and B in their ARP tables
all hosts process all ARP packets, even if not addressed to themselves