Final Review

the future of networking
IP, routing, protocol implementation
network address translation
transport layer, TCP, reliable transmission, flow and congestion control
lower layers, including wireless and Aloha, Ethernet, learning bridges, switches
ATM and public networks
basic issues

Potential Futures

how fast can we go?
wireless: 802.11 and cellular. What are the limits?
Ethernet seems to be dominating over ATM: what is the next hot new technology?
fragmentation of the Internet: intranets, firewalls, NAT, ...
when (if ever) will we switch to IPv6?
will the Web look like television? what will replace the Web and the Internet, and when?
peer-to-peer networks

NAT is nonstandard -- what is standard is using a router
an endpoint router (gateway) does the following:
- forwards packets from the Internet to the internal network
- forwards packets from the internal network to the Internet
- does not modify the source and destination
- does not look at the higher layer (transport layer or ICMP) headers
- usually runs a routing protocol
such a router has disadvantages for a SOHO (Small Office/Home Office):
- the IPs on the inner network must be routable, that is, recognizable throughout the Internet and assigned to this network, which can be expensive
- IP addresses alone do not provide enough information for useful access control
- in particular, SOHOs usually want to restrict access only to outgoing connections and to servers known to the administrator

a NAT box performs the forwarding functions of a router, but not the general routing functions. In particular, the "inside" addresses are not Internet routable addresses, e.g. are from 10.0.0.0/8, 172.16.0.0/12, or 192.168.0.0/16 (RFC 1918)
when a NAT box with external IP number IP_official is forwarding a packet with the SYN bit set, it:
- in -> out: records inner IP (source IP) and inner port (TCP source port), then rewrites the source IP number to be IP_official, and may rewrite the source port number to be a port on the NAT box, Port_NAT
- out -> in: checks the destination port to make sure this connection is authorized, then rewrites the destination IP (and perhaps the destination TCP port) to be the inner IP and port of the server
in each of these cases, an entry is made in a table recording the mapping (inner IP, inner port <-> Port_NAT for external IP, external port)
this table to rewrite packets from the inside to be valid (routable) on the internet, and packets outside outside to be delivered to the correct host inside
the hosts on the inside are configured to use the NAT box as their default router, so they forward all their external traffic to the inner interface of that box

NAT box with IP_official = 1.2.3.4, IP_inside = 192.168.100.1
client on the inside with IP = 192.168.100.102 sends to "next hop" 192.168.100.1 a packet with IP src = 192.168.100.102, IP dst = 2.3.4.5, TCP src port = 4567, TCP dst port = 993 (IMAP over SSL)
NAT box sends a packet to its next hop router with IP src = 1.2.3.4, IP dst = 2.3.4.5, TCP src port = 4567, TCP dst port = 993
TCP packets received by the NAT box with IP src = 2.3.4.5, IP dst = 1.2.3.4, TCP src port = 993, TCP dst port = 4567 are rewritten to IP src = 2.3.4.5, IP dst = 192.168.100.102, TCP src port = 993, TCP dst port = 4567
the inner TCP port may also be modified if necessary to avoid confusion of two connections, e.g. another client connecting to TCP port 993 of 2.3.4.5 from its port 4567 would have its inner port changed
a server on the inside with IP = 192.168.100.101, serving HTTP on port 80
incoming packets received with any source IP and TCP port for port 80 of the NAT are rewritten to have destination IP 192.168.100.101
a server on the inside with IP = 192.168.100.102, serving HTTP on port 80 might be configured to receive packets sent to the NAT's port 8080
also functions as a firewall, discarding unsolicited packets to any port that is not explicitly configured
UDP functionality is similar, but since there are no connections, requires caching and timing out of table entries

TCP, UDP, demultiplexing, sockets/WinSock APIs
headers, pseudo-header for end-to-end reliability
reliable transmission
flow control: stop-and-wait, sliding windows, sequence number wrap-around, time-to-live
congestion control: linear increase/multiplicative decrease

framing, addresses, and access control
wireless and Aloha
Ethernet: header, hubs, switches, speed ranges, half and full duplex, collision detection, binary exponential backoff
learning bridges/switches, spanning trees, comparison of routing and switching, wormhole routing
serial lines, SLIP

ATM:
- 53-byte cells (5+48)
- virtual circuits: VPI, VCI identify connection, only have meaning locally to one link
- IP over ATM: virtual LAN (everything multicast to the nodes on the LAN), Classical IP over ATM (CLIP, connection set up for each IP destination, ATMARP)
- connection signaling, Quality of Service, adaptation layers (AAL-5)
SONET: synchronous, 8000 frames per second
Frame Relay: virtual connections, packet oriented