Hierarchical Routing

Problem: as size of network grows, routing table, complexity grows

millions of nodes (hosts, routers) in Internet

Solution: hierarchically aggregate nodes into "regions" (domain)
node have full knowledge of routes, topological structure within region
one (or more) nodes in region responsible for routing to the outside

Teminology:
intradomain routing: within domain
interdomain routing: between domains
autonomous system (AS): domain, region, administrative domain
gateway: routes to/from domain, a.k.a. border router

Hierarchical Routing (cont)
Three domains: A, B, C

A.a, A.b A.c run interdomain routing
protocol

A.c, B.a, B.b, C.a run intradomain
routing protocol among themselves
Hierarchical Routing (cont)
Different routing protocols can be
used for interdomain and
intradomain routing

A.a routing table:
A look inside A.c:
destination next hop
h6 A.b
. A.b
h9 A.b
all other
(default route)
A.c
Hosts and routers
Hosts (end systems) typically perform no routing
start packets on their way
send packets to nearest router

Q: how do hosts learn identity of nearby router:
A1: IP address of router hard-coded into file (see /etc/networks
on many UNIX systems)
A2: router discovery: RFC 1256
router periodically broadcasts its existence to attached hosts
host (on startup) broadcasts query (who is my router) on attached
links/LANs
Network Layer Case Study: the Internet
Fields in IP packet:
version number: (of IP protocol),
current version is 4, new version is 6
header length: because of options,
length of header is variable
TOS: not used, idea was to allow
different levels of reliability, real-time,
etc
packet length: header plus data
identifier: used with IP fragmentation
to identify fragments belonging to
same original IP packet
flags: 2 bits: do not fragment, more
fragments
fragmentation offset: if this a fragment, where it belongs in original packet
time-to-live: decremented by each router, so a packet will not loop forever in
the net
protocol: which upper layer protocol to demultiplex to. See RFC 1700
header checksum: recomputed at each hop, as TTL changes
source, dest IP address: of original sender, and eventual recipient

IP fragmentation and Reassembly
transport layer packet may be too big to send in single IP packet
underlying data link protocol will constraint maximum IP length
fragmentation: IP packet divided into fragments by IP
each fragment becomes its own IP packet
each address has same identifier, source, destination address
fragment offset gives offset of data from start of original packet
more fragment bit: 0 means last bit in this fragment
fragments not reassembled until final destination
Internet Intradomain Routing: RIP
RIP: Routing Information Protocol, uses distance vector algorithm, with
link costs of 1
shortest path
routing table sent to neighbors every 30 seconds, or when route costs change

Implemented as a daemon (user-level process)
communicates with other attached router using UDP packets
note: UDP packets can be lost!
if route via neighbor not updated in 3 minutes, timeout route (set cost to infinity)
called routed on UNIX systems
A RIP routing table
Example table taken from freya.cs.umass.edu:

~ netstat -rn (note: on freya.cs.umass.edu)
DestinationGateway FlagsRefcntUseInterface
127.0.0.1127.0.0.1 UH252260Io0
Default 128.119.40.254UG5 15223In0
128.119128.119.40.195U28188671In0
Internet Intradomain Routing: OSPF
OSPF: open shortest path first
open: a published standard (RFC 1247)
interior gateway protocol: for intradomain outing within an autonomous
system (AS)
uses link state algorithm to determine routes
each outgoing link (interface) assigned dimensionless cost
different cost can be used for different TOS
load balancing: with several equal-cost-paths to destination, will distribute
load across both paths

Support for hierarchy:
autonomous system divided into "areas"
one area designated "backbone"
area border routers in backbone route between areas
other routers in backbone also
AS boundary router talks to outside world
Internet Intradomain Routing: OSPF
(cont)







Intra-area routing:
never cross backbone

To get from one area to another:
source area -> backbone -> destination area


area router: red
boundary router: blue
Interdomain Internet Routing: BGP
BGP: Border Gateway Protocol
routing between nodes in different autonomous systems (i.e.,
routing between networks)
RFC 1267, 1268
uses a distance verctor approach

Policy-Based Routing
rather than costs to destinations, BGP routers exchange full
path information (networks crossed) to destination router can
decide on policy basis which route to take
e.g. "traffic from my AS should not cross AS's a,b,c,d"

BGP implementation:
implemented as a daemon (user-level process)
communicates with other BGP routers using TCP
ICMP: Internet Message Control Protocol
used to communicate network-
level error conditions and info
to IP/TCP/UDP protocols or
user processes
often considered part of IP, but
ICMP message sent within IP
datagram
IP demultiplexes up to ICMP
using IP protocol field
ICMP message contains IP
header and first 8 bytes of IP
contents that causes ICMP
mesage to be generated
I C M P t y p e co d e d es cri p t i o n
0 0 ech o r ep l y ( t o p i n g )
3 0 d es t i n at i o n n et w o r k
u n r ea ch ab l e
3 1 d es t i n at i o n h o s t
u n r ea ch ab l e
3 2 d es t i n at i o n p r o t o co l
u n r ea ch ab l e
3 3 d es t i n at i o n p o r t
u n r ea ch ab l e
3 6 d es t i n at i o n n et w o r k
u n k n o w n
3 7 d es t i n at i o n h o s t
u n k n o w n
4 0 s o u r ce q u en ch
( co n g es t i o n co n t r o l )
8 0 ech o r eq u es t
9 0 r o u t er ad v er t i s em en t
1 0 0 r o u t er d i s co v e r y
1 1 0 T T L ex p i r ed
1 2 0 I P h ead er b ad
IPv6: next generation IP
Changes to Ipv4:
128 bit addresses (so we don't run
out of IP addresses)
header simplification (faster
processing)
more support for type of service
priorities
flow identifier: identifiy packets in a
connection
security
Notes:
no fragmentation in network
packet too big generates ICMP error to source
source fragmentation via extension header
no checksum (already done at transport and data link layer)

Transitioning from IPv4 to IPv6
Internet too big for "flag day":
can't turn off all IP routers, install IPv6 and reboot
IPv4 nodes will be legacy
IPv6 nodes can route IPv4 packets
IPv4 nodes can not route IPv6 packets

Tunneling:
source and destination speak network protocol X
physically intermediate nodes speak network protocol Y
source takes protocol X packet, sticks it inside (encapsulates) protocol Y
packet
intermediate nodes route using protocol Y
destination receives packet using protocol Y, removes protocol X packet
network between source and destination looks like a single link to
protocol X

Tunneling: a pictorial view

Case Study: ATM Network Layer
ATM: packet (cell) format:
UNI: user-network interface
(host-to-switch)
NNI: network-network interface
(switch-to-switch)
GFC: generic flow control
(unused)
VPI: virtual path identifier
VCI: virtual circuit identifer
VPI and VCI together a call/connection identifier
PTI: payload type: 3 bits
111: RM cell (recall RM congestion control)
000: user cell
010: user cell, congestion experienced (recall EFCI)
CLP: cell loss priority (1 bit)
priority bit for discarding
HEC: header error correction
DATA: 48 bytes of data
Observations about ATM Cell
very small
reflecting telephony origins
48 bytes a compromise, halfway 64 and 32
no explicit source/destination address
VCI/VPI used instead
faster switching (VPI/VCI can index into table)
28 bit VPI/VCI for switching instead of 128 bit IP address in IPv6
(savings)
fixed length for faster switching
minimal priority
ATM networks: Virtual-circuit Oriented
VCI/VPI together identify call
multiple calls (VCI) bundled into same VP
network can switch on VP basis only
less state (network only sees VP's)
all VC's in VP follow same path
Connection Setup in ATM
messages ("signaling") used to setup up call through network
state info (VP switching info - which output line to switch
incoming VC) set up in switches
meaning of call setup messages:
message sent byhost tonet receivedfromnet
SETUP establish
connection
incomingcall
CALL
PROCEEDING
host seescall Networkattemptingcall
CONNECT I accept incoming
call
Your call accepted
CONNECT
ACK
ACKreceived
CONNECT
ACKreceived
CONNECT
ATM Call Setup (cont)
Observations:
unlike Internet, switches involved in call setup
state creation
ACKing between switches
wait one RTT before sending data
unlike UDP
same as TCP
what if connection breaks?
other switches must remove state
ATM standard does not specify a routing protocol
Switches and Routers: What's Inside
Input interface cards:
physical layer processing
memory buffers to hold incoming packet

Switch fabric:
to move packets from input to output

Output interface cards:
memory buffers to hold outgoing packets
physical layer processing


Control processor: routing table updates, supervisory (management)
functions
will typically not touch the packets being switched

Switching Fabrics
Two popular ways to switch:
switching via memory: input line ports write to memory, output
ports read from memory
switching via a bus: bus (backplane) connects input and
output ports
e.g.: Cisco AGS+ has 533 Mbps backblane bus
Network Layer: Summary
Network service: datagram versus VC

Theory of routing protocols
link state and distance vector
multicast
broadcasting

Case studies:
Internet
IPv4, IPv6
protocols for exchanging routing information: RIP, OSPF, BGP
ATM