Exam Review

Packet Networking

sharing an expensive medium
assumes nobody needs to send all the time (is it correct?)
shorter packets mean shorter delay for small amounts of information
bandwidth sharing mostly implicit, though many attempts to make it more explicit, manageable, controllable

a very simple system can be built on top of point-to-point connections which can transfer one byte at a time
alternately, shared media have conventions (Medium Access Control, MAC) for transferring entire packets and controlling access to a shared medium
an incoming packet must get from the hardware to the application: downcall or upcall?
sending a packet only require queueing if the medium is busy

structured, hierarchical name space
arbitrary mapping to IP addresses
mapping supported by distribute database
different "types" of mapping, e.g. A for address, MX for mail server, NS for Name Server
recursive or iterative queries
queries elicit answers, which may include authority and additional information
idempotent protocol means queries can be repeated for reliability

end-to-end addressing, protocol, length, fragmentation
hop-by-hop TTL, checksum
best-effort: packets can be discarded, reordered, misdelivered, delayed, or have bit errors
once the routing tables are in place, forwarding is straightforward:
- verify header checksum, discard if incorrect
- deliver to socket if destination is local
- decrement TTL, discard if zero
- look up destination in routing table, select (a) longest match, (b) best metric, or (c) first entry
- recompute checksum
- queue on output interface
network masks allow the division of networks into subnetworks
all-1s host part is broadcast address, all-0s host part is network number

ICMP used to report errors (e.g. when TTL=0 during traceroute) or probe network (ping)
IPv6: much bigger addresses, fewer processing steps at router
path MTU discovery
fragmentation: why it is needed, packet identifiers, fragment offsets

Distance-Vector: acquire routing tables from neighbors, use in computing own routes
RIP, RIPv2
BGP uses a variant, path-vector, that allows more administrative control
information received from neighbors may reflect old state that is no longer correct
split-horizon, with or without poisoned reverse, helps prevent some of the routing loops
other link failures lead to counting-to-infinity

Link-State: the networks each router is connected to are broadcast to all other routers
routers use Dijkstra's shortest path algorithm to compute the best route to each destination
OSPF, IS-IS use LS
OSPF divides the AS into areas
routers in each area receive complete information about the other routers in the same area, and summary information about routers in other areas
OSPF, IS-IS, and RIP are Interior Gateway Protocols, IGPs
BGP is an EGP

leaf-initiated join and reverse-path forwarding/broadcasting
source-specific vs. any-source
PIM-SM, PIM-SSM
rendezvous points change any-source multicasting to a single-source multicasting
as an optimization, can build a new source-specific multicasting tree back to one or more sources
standards could still evolve
soft state in routers must be refreshed to maintain subscription

no cutting-and-pasting: all answers must be in your own words, except for an occasional, well-marked quote with appropriate citation
answer the question that is asked
answer each question completely
5 minutes before turning in, run spellchecker (if not already running)