Internet Telephony Lecture Notes

3 June 2008 • The Internet

Outline

• Networks and network isolation.

• Overlay networks.

• The Internet.

  • Routing

  • Congestion

• IP Version 6

Networks

• A network is a system for moving bits between hosts.

  

  • “System” means hardware and software (protocols).

    • Links are the hardware; hosts are the software.

Network Isolation

• Isolated networks can't communicate.

• Sometimes isolated networks are useful.

  Isolation improves security (usually). Smaller is easier. Reflects work practice.

• But mostly, network isolation is undesirable.

  • The network effect: more is better.

Overcoming Network Isolation

• Internetworking overcomes network isolation by connecting separate networks.

  Internetworking can occur at several protocol levels. We are interested in internetworking at transport and above.

• The two main impediments to internetworking are heterogeneity and scale.

Network Heterogeneity

• Two networks can differ in both hardware and software.

  • Hardware differences include fiber, twisted pair, wireless, …

  • Software differences include services offered, host addressing, data formats, …

• Internetworking has to resolve the differences between heterogeneous networks.

Network Scale

• Economics pushes networks to be as big as possible.

  • Per-user network costs decrease (amortization).

  • Users have more to access (the network effect).

• Network autonomy and management push against big networks.

  • Non-linear interconnection costs and growth.

• Internetworking must balance the desire for big with the need for small.

Gateways

• A gateway is a translation device between heterogeneous networks.

  • The translation involves both hardware and software differences.

  The network on the other side looks like the network on this side. Address space, data formats, services, …

Problem Solved?

• Gateways deal with heterogeneity, but at great cost.

  • And the costs grow with the differences.

  • Software differences are particularly costly.

• Gateways don't handle scale well.

  • Each interconnection needs a gateway.

  • N different networks need O(n²) gateways.

• Gateways are necessary, but aren't sufficient.

If I Were King…

• Eliminating heterogeneity would make internetworking much easier.

• The remaining scale problems are mostly solvable.

  • Managing complexity via hierarchical structure, for example.

• But mandating homogeneity is impractical, for technological, economic and political reasons.

  • Or is it?

Overlay Networks

• An overlay network is a network implemented on top of existing networks.

  Conceptually, each existing network is treated as a host in the overlay network.

• The overlay network defines its own set of hardware and software specs.

  • One network deals with another network via the overlay network's services.

Overlay Network Advantages

• Gateway count is linear in the different constituent networks.

  If the overlay network is simple, then the gateway translations should be less difficult too.

• Overlay-network facilities can manage scaling.

  • Which is the proper place to put it.

  • An overlay network is a network. (Recursion!)

Something in the Air

• Several interesting networking ideas were swirling around during the early 1960s.

  • Galactic networking: unconstrained, world-wide access (J.C.R. Licklider).

  • Survivable networking: what happens if the bomb drops here? (Paul Baran)

  • Packet switching: little chips of data flowing everywhere (Leonard Kleinrock, Donald Davies).

Enter the Internet

• The ideas coalesced around a Defense Advanced Research Project Agency (DARPA) project.

  1962: Licklider's proposal.
  1963: DARPA gets interested.
  1965: Packet-switched network demonstrated.
  1966: ARPANET project started.
  1969: Four-node ARPANET brought on-line.
  1972: E-mail, telnet, and ftp released.
  1983: Unix BSD 4.2 sockets interface released.
  1991: WWW released.
  1992: One million hosts.
  

Internet Service

• To simplify the heterogeneity problem, the Internet provides a simple service: best-effort datagram transmission.

  • Datagram: data to be sent is placed in bounded-size packets.

  • Best-effort: packet delivery is not guaranteed; packets may be lost.

    • Delivery is unreliable: arrival times are uncertain and data may be corrupted.

Dumb Networks

• A dumb network pushes intelligence to the edge.

  • End-points deal with lost, corrupted, or out-of-order delivery.

  • The network is fast, cheap, and simple.

• This leads to the end-to-end principle:

  • Do things in the end-points or the network, but not both.

  • Do things in the network that all end-points use.

Internet Operation

Observations

• Data to be sent are fragmented into packets or datagrams.

  • This transport style is called packet switching.

• Each packet is sent through the network independently of all other related packets.

  • This transport style is called connectionless.

• Packets may be lost or corrupted during transport.

  • The end-point hosts deal with it.

Internet Routing

What's the Good way for A to send some bits to B? “Good” is some combination of “fast” and “reliable.”

• Finding a good path from A to B is the routing problem.

• A good routing-problem solution is crucial for scalability.

Routers

• Some hosts in a network aren't interested in sending or receiving bits.

  Rather, they're interested in helping other hosts send and receive bits.

• Such hosts are known as routers.

  • Non-router hosts are called end-points.

Router Structure

• The buffer holds packets yet to be routed.

• The CPU moves packets from the buffer to the output links through the routing table.

Packet Routing

• Each packet contains the receiver's address (among other information).

• A routing table is a list of (receiver, output link, cost) triples.

  • Sending a packet to receiver along output link costs cost.

• The router CPU forwards each incoming packet along the output link with the best (usually cheapest) cost to the packet's receiver.

Here Comes Trouble

• Any connected host can dump any number of packets into the network (no admission control).

• Routers are shared resources for (potentially) any host on the network.

• Router buffers are finite.

• These three facts make buffer overflow a possiblity.

  • Packets are coming in faster than they're going out.

Congestion

• When a router buffer's full, the only thing to do is discard packets.

• A router discarding packets is congested.

• Given unconstrained shared access, congested routers are a fact of life.

• Best-effort delivery acknowledges congestion by allowing for discarded packets.

• Any other congestion handling must be done by the end-points.

IP Version 6

• IP version 6 (IPv6) is the next revision of IPv4.

  • Bring IPv4 up to date and fix some shortcomings.

• Among the many IPv6 features, ones of particular interest are

  • Authentication and privacy capabilities.

  • Quality of service mechanisms.

  • A larger and more structured address space.

IPv6 Addressing

• IPv6 addresses are 128 bits (16 bytes) as opposed to 32 bits (4 bytes) for IPv4 addresses.

  • Plenty of addresses for everything.

  • Lots of internal structure for hierarchical addressing.

• New address types, leading to new services.

  • Multicasting, and anycasting.

• Improved routing services for mobility and end-point selection.

Quality of Service

• Indicate how packets should be treated in the network.

  • IPv4's mechanism was little understood and mostly ignored.

• A 16-bit flow label identifies packets belonging to the same data stream.

• A 4-bit priority field suggests packet handling within the network.

  • Relative discard rates for constant-rate packets, for example.

Security

• The Authentication Header reliably identifies the sender.

• The Encapsulating Security Header provides data integrity and confidentiality.

• Both features are implementation independent, specifying frameworks within which particular mechanisms operate.

• The IPv6 security features have been split off into a separate set of IPSec protocols.

References

• Computer Networks, forth ed. by Larry Peterson and Bruce Davie, Morgan Kaufmann, 2007.

• End-To-End Arguments in System Design by Jerome Saltzer, David Reed, and David Clark in ACM Transactions on Computing Systems, November 1984.

• The Rise of the Middle and the Future of End-to-End by the Internet Architecture Board, James Kempf and Rob Austein, editors, RFC 3724, March 2004.

• IP Next Generation Overview by Robert Hinden, Communications of the ACM, June 1996.

• Where the Wizards Stay Up Late by Katie Hafner and Matthew Lyon, Touchstone Books, 1996.