Network and Network Routing

October 28, 2009 , 9

Ye Wang Ye Wang
1
Outline Outline
k Network overview
2
Network Layer Network Layer
Transport packet from source to Transport packet from source to
dest.
Network layer in every host,
router router
Basic functions:
Data plane: forwarding p g
move packets from
routers input port to
router output port router output port
Control plane: path p p
determination and call setup
determine route taken by
packets from source to
3
packets from source to
destination
Forwarding: Illustration Forwarding: Illustration
routing and call
setup
4
Network Layer: Complexity Factors Network Layer: Complexity Factors
For users: quality of service For users: quality of service
guaranteed bandwidth?
preservation of inter-packet timing (no jitter)? preservation of inter packet timing (no jitter)?
loss-free delivery?
in-order delivery? y
Interaction between users and network
providers providers
signaling: congestion feedback/resource reservation
For network providers For network providers
efficiency
policy of route control
5
policy of route control
scalability
Network Layer Quality of Service Network Layer Quality of Service
Network Service
Congestion
Guarantees ?
Network
Architecture
I t t
Service
Model
b t ff t
Bandwidth
none
Loss
no
Order Timing
Congestion
feedback
(i f d
Internet
ATM
best effort
CBR
none
constant
no
yes
no
yes
no
yes
no (inferred
via loss/delay)
no
ATM VBR
rate
guaranteed
rate
yes yes yes
congestion
no
congestion
ATM
ATM
ABR
UBR
guaranteed
minimum
none
no
no
yes
yes
no
no
g
yes
no
ATM UBR none no yes no
no
Internet model being extended: Intserv, Diffserv
multimedia networking
6
multimedia networking
ATM: Asynchronous Transfer Mode; CBR: Constant Bit Rate; V: Variable; A: available; U: User
Network Layer: Protocols Network Layer Protocols
Network layer functions:
Transport layer
Routing protocols
path selection
e.g., RIP, OSPF, BGP
Control protocols
error reporting
e.g. ICMP
Control protocols
- router signaling
e.g. RSVP
Network layer protocol (e g IP)
Network
layer
forwarding
Network layer protocol (e.g., IP)
addressing conventions
packet format
packet handling conventions
Link layer
7
physical layer
Outline Outline
k Network overview
Control plane: routing overview p g
8
Control Plane: Routing Control Plane: Routing
Routing
Goal: determine good paths
(sequences of routers) thru
network from source to dest
g
Graph abstraction for the
network from source to dest.
p f
routing problem:
5
graph nodes are routers
graph edges are physical
A
C B
F
2
5
3
5
g p g p y
links
links have properties: delay,
it $ t li
A
E D
F
2
1
3
1
1
2
9
capacity, $ cost, policy
1
Key Desired Properties of a Routing Algorithm Key Desired Properties of a Routing Algorithm
Robustness Robustness
Optimality
find good path find good path
(for user/provider)
Simplicity p y
10
Example: Latency Optimal Routing
- Robustness
- Optimality
Si li it
Example: Latency Optimal Routing
h l k h l f l ( )
- Simplicity
Assume each link has a latency function l
e
(x):
latency of link e when x amount of traffic
h h goes through e,
e.g., prop. + 1/( x), where prop. is link
d l d h l k propagation delay and the link capacity
Objective: min E
e
x
e
l
e
(x
e
)
l(x) = x
s t
total demand = 1
11
l(x) = 1
Routing Design Space
- Robustness
- Optimality
Si li it Routing Design Space
R ti h l d i
- Simplicity
Routing has a large design space
who decides routing?
s tin : nd h sts m k d isi n source routing: end hosts make decision
network routing: networks make decision
how many paths from source s to destination d? how many paths from source s to destination d?
multi-path routing
single path routing
ill i d k ffi d d? will routing adapt to network traffic demand?
adaptive routing
static routing static routing

12
Routing Design Space:
User based Multipath Adaptive
- Robustness
- Optimality
Si li it
User-based, Multipath, Adaptive
R ti h l d i
- Simplicity
Routing has a large design space
who decides routing?
s tin : nd h sts m k d isi n source routing: end hosts make decision
network routing: networks make decision
how many paths from source s to destination d? how many paths from source s to destination d?
multi-path routing
single path routing
ill i d k ffi d d? will routing adapt to network traffic demand?
adaptive routing
static routing static routing

13
User Optimal Multipath Adaptive User Optimal, Multipath, Adaptive
User optimal: users pick the shortest routes User optimal: users pick the shortest routes
(selfish routing)
14
Braesss paradox
Price of Anarchy Price of Anarchy
F k h l l f For a network with linear latency functions

total latency of user (selfish) routing for

given traffic demand g
4/3
total latency of network optimal routing total latency of network optimal routing
for the traffic demand
15
Price of Anarchy Price of Anarchy
F k h For any network with continuous, non-
decreasing latency functions
total latency of user (selfish) routing
f ff d d for given traffic demand

l l f k i l i total latency of network optimal routing
for twice traffic demand
16
- Robustness
- Optimality
Si li it
Routing Design Space:
Internet
R ti h l d i
- Simplicity Internet
Routing has a large design space
who decides routing?
s tin : nd h sts m k d isi n source routing: end hosts make decision
network routing: networks make decision
- (applications such as overlay and p2p are trying to bypass it) ( pp y p p y g yp )
how many paths from source s to destination d?
multi-path routing
i l th ti ( ith ll t f lti th) single path routing (with small amount of multipath)
will routing adapt to network traffic demand?
adaptive routing adaptive routing
static routing (mostly static; adjust in larger timescale)

17
Routing: Single-Path, Adaptive Routing g g , p g
Ass m li k sts
Link costs reflect current traffic intensity
Assume link costs
reflect current
t ffi
D
A
C
2+e
0
0
1+e1
0
D
A
C
1
1+e
0
0
0
0
0
traffic
B
0
0
1 e1
t=1
B
e
0
e
1
1
0
D
D
A
C
0
2+e
0 0
t=1
e
t=0
D
A
C
2+e
0
0
D
A C
A
B
C
1+e
1
0 0
A
B
C
2+e
e
0
1+e1
B 1
1
t=2
Solution: Link costs are a combination of current
traffic intensity (dynamic) and topology (static). To
b l h l h ld b
t=3
e
18
improve stability, the static topology part should be
large. Thus less sensitive to traffic; thus non-
adaptive.
Shortest Path Routing Protocols Shortest Path Routing Protocols
Static (positive) costs assigned to Static (positive) costs assigned to
network links, the path from a source
to a destination chosen by the routing to a destination chosen by the routing
protocol is a shortest path among all
possible paths possible paths
h i li k b dj d i the static link costs may be adjusted in a
longer time scale: this is called traffic
engineering engineering
19
Outline Outline
k Network overview
Control plane: routing overview p g
Distance vector protocols
20
Distance Vector Routingg
Basis of RIP IGRP EIGRP routing Basis of RIP, IGRP, EIGRP routing
protocols
Based on the Bellman-Ford Based on the Bellman Ford
algorithm (BFA)
Conceptually runs for each Conceptually, runs for each
destination separately
Distance Vector Routing: Basic Idea g
At node i, the basic update rule t n , t a up at ru
) ( min
) ( j ij i N j i
d d d + =
e
d
where
) (
) ( j ij i N j i e
destination
where
- d
i
denotes the distance
estimation from i to the
destination,
- N(i) is set of neighbors of
d i d
j
i
d
ij
d
node i, and
- d
ij
is the distance of
the direct link from i to j;
i
j
d
j
22
the direct link from i to j;
assume positive
Outline Outline
k Network overview
Control plane: routing overview p g
Distance vector protocols
synchronous Bellman-Ford (SBF) synchronous Bellman-Ford (SBF)
23
Synchronous Bellman-Ford (SBF)
C B
7
1
Synchronous Bellman Ford (SBF)
Nodes update in rounds:
A
E
D
8
10
2
2
there is a global clock;
at the beginning of each round, each node sends
2
its estimate to all of its neighbors;
at the end of the round, updates its estimation
)) ( ( min ) 1 (
) (
h d d h d
j ij i N j i
+ = +
e
A
? ) 0 ( d
A
BB
E
24
C
D
Outline Outline
k Network overview
Control plane: routing overview p g
Distance vector protocols
synchronous Bellman-Ford (SBF) synchronous Bellman-Ford (SBF)
- SBF/
25
SBF/
C B
7
1
SBF/
l ( 0)
A
E
D
8
10
2
2
Initialization (time 0):
2

=
=
dest i 0
) 0 ( d

=
otherwise
) 0 (
i
d
26
Example
C B
7
1
Example
A
E
D
8
10
2
2
Consider D as destination; d(t) is a vector consisting of
ti ti f h d t d t
2
A B C E D
estimation of each node at round t
d(0)
0
d(1)
2 2 0
( )
2 2 0
d(2)
12 3 2 2 0
d(3)
10 3 2 2 0
d(4)
10 3 2 2 0
27
Observation: d(0) > d(1) > d(2) > d(3) > d(4) =d*
A Nice Property of SBF: Monotonicity
)) ( ( min ) 1 (
) (
h d d h d
j ij i N j i
+ = +
e
A Nice Property of SBF: Monotonicity
d f d( ) d d( ) Consider two configurations d(t) and d(t)
If d(t) > d(t)
i.e., each node has a higher estimate in one , g
scenario (d) than in another scenario (d),
then d(t+1) > d(t+1)
i.e., each node has a higher estimate in d than in d i.e., each node has a higher estimate in d than in d
after one round of synchronous update.
28
Correctness of SBF/
)) ( ( min ) 1 (
) (
h d d h d
j ij i N j i
+ = +
e
Correctness of SBF/
Claim: d
i
(h) is the length L
i
(h) of a Claim: d
i
(h) is the length L
i
(h) of a
shortest path from i to the destination
using h hops using h hops
base case: h = 0 is trivially true base case: h = 0 is trivially true
assume true for h assume true for h,
i.e., L
i
(h)= d
i
(h), L
i
(h-1)= d
i
(h-1),
29
Correctness of SBF/
)) ( ( min ) 1 (
) (
h d d h d
j ij i N j i
+ = +
e
Correctness of SBF/
consider h+1 hops: consider h 1 hops:
= + ) 1 (h L
i
))) ( ( min ), ( min(
) (
h L d h L
j ij i N j i
+
e
) (
i
))) ( ( min ), ( min(
) (
h d d h d
j ij i N j i
+ =
e
))) ( ( ) ( (
) ( j ij i N j i e
)) 1 ( ), ( min( + = h d h d
i i
))) ( ( ), ( (
) ( j ij i N j i e
)) 1 ( ), ( min( + h d h d
i i
since d
i
(h) d
i
(h-1)
) ( )) 1 ( ( min )) ( ( min ) 1 (
) ( ) (
h d h d d h d d h d
i j ij i N j j ij i N j i
= + s + = +
e e
30
) 1 ( ) 1 ( + = + h d h L
i i
Bellman Equation q
We referred to the equations as We referred to the equations as
Bellman equations (BE):
) ( min
) ( j ij i N j i
d d d + =
e
where d
D
= 0.
- SBF/ solves the equations in a
distributed way d str buted way
- Does the equation have a unique
s luti n (i th sh rt st p th n )?
31
solution (i.e., the shortest path one)?
Uniqueness of Solution to BE
) ( min
) ( j ij i N j i
d d d + =
e
Uniqueness of Solution to BE
h l d ll h h Assume another solution d, we will show that
d = d*
case 1: we show d > d*
Since d is a solution to BE we can construct Since d is a solution to BE, we can construct
paths as follows: for each i, pick a j which
satisfies the equation; since d* is shortest d > d* satisfies the equation; since d* is shortest, d > d*
1
C B
1
2
3
1
1
Dest.
A
C B
7
8
10
1
2
10
32
1
1
E
D
10
2
2
Uniqueness of Solution to BE
) ( min
) ( j ij i N j i
d d d + =
e
Uniqueness of Solution to BE
h d d* Case 2: we show d d*
assume we run SBF with two initial
configurations: configurations
one is d
another is SBF/ (d

) another is SBF/ (d ),
-> monotonicity and convergence of SBF/ imply m n n y an n rg n f F/ mp y
that d d*
33
Outline Outline
k Network overview
Control plane: routing overview p g
Distance vector protocols
synchronous Bellman-Ford (SBF) synchronous Bellman-Ford (SBF)
- SBF/
- SBF/-1 F/
34
SBF at another
Initial Configuration: SBF/ 1
C B
7
1
Initial Configuration: SBF/-1
A
E
D
8
10
2
2
Initialization (time 0):
2

=
=
dest i 0
) 0 ( d

=
otherwise 1
) 0 (
i
d
35
Example
C B
7
1
Example
A
E
D
7
8
10
2
2
Consider D as destination
2
A B C E D
d(0)
-1 -1 -1 -1 0
d(1)
6 0 0 2 0
d(2)
7 1 1 2 0
d(2)
7 1 1 2 0
d(3)
8 2 2 2 0
d(4)
9 3 3 2 0
d(5)
10 3 3 2 0
d(6)
10 3 3 2 0
36
Observation: d(0) s d(1) s d(2) s d(3) s d(4) s d(5) = d(6) = d*
Bellman Equation and
Correctness of SBF/ 1
)) ( ( min ) 1 (
) (
h d d h d
j ij i N j i
+ = +
e
Correctness of SBF/-1
SBF/-1 converges due to monotonicity SBF/ 1 converges due to monotonicity
At equilibrium, SBF/-1 satisfies the set of equations
called Bellman equations (BE) q ( )
) ( min
) ( j ij i N j i
d d d + =
e
where d
D
= 0.
Another solution is shortest path solution d*
Since there is a unique solution to the BE equations; Since there is a unique solution to the BE equations;
thus SBF/-1 converges to shortest path
Question: will SBF converge under other non-negative initial conditions?
37
Question: will SBF converge under other non negative initial conditions?
Problems of running synchronous BF?
Outline Outline
k Network overview
Control plane: routing overview p g
Distance vector protocols
o synchronous Bellman-Ford (SBF) o synchronous Bellman-Ford (SBF)
asynchronous Bellman-Ford (ABF)
38
Asynchronous Bellman-Ford (ABF) Asynchronous Bellman Ford (ABF)
f l b l No notion of global iterations
each node updates at its own pace
Asynchronously each node i computes
) ( i
i
d d d
using last received value d
i
j
from neighbor j.
) ( min
) (
i
j ij i N j i
d d d + =
e
g
j
f g j
Asyncrhonously node j sends its estimate to Asyncrhonously node j sends its estimate to
its neighbor i:
there is an upper bound on the delay of estimate there is an upper bound on the delay of estimate
packets (no worry for out of order)
Asynchronous Bellman-Ford (ABF) Asynchronous Bellman Ford (ABF)
l d d ff d In general, nodes are using different and
possibly inconsistent estimates
j
d
i
d
j
i
d
i
j
d
i
d
i
Distance Table: Example
C B
7
1
Distance Table: Example
A
E
D
7
8
10
2
Below is just one step! The protocol repeats forever!
E
D
2
d ()
distance tables
from neighbors
computation
Es
distance
t bl
distance
table E sends
A B D
A B D
d ()
A
E
table
to its neighbors
A: 10
10 15
A B D
0 7
A B D
A: 10
B
A: 10
B: 8 7 0
B: 8
17 8
C
C: 4 1 2 D: 4 9 4
D
D: 2 0
D: 2
2
E: 0 10 8 2
Asynchronous Bellman-Ford (ABF) Asynchronous Bellman Ford (ABF)
BF ll ll h h ABF will eventually converge to the shortest
path
links can go down and come up but if topology is
stabilized after some time t, ABF will eventually
t th sh t st th ! converge to the shortest path !
If h k i d h ABF If the network is connected, then ABF
converges in finite amount of time, if
di i conditions are met
ABF Convergence ABF Convergence
h d ff f BF There are too many different runs of ABF,
so need to use monotonicity
Consider two sequences: Consider two sequences
SBF/; call the sequence U()
SBF/-1; call the sequence L()
System State System State
j
i
where can distance estimate from node j appear?
System State System State
three types of three types of
distance estimates
from node j:
dj
from node j
- d
j
: current
distance estimate
d
i
d
i
j
d
i
j
at node j
- d
i
j
: last d
j
that
neighbor i received
d
i neighbor i received
- d
i
j
: those d
j
that
are still in transit are still in transit
to neighbor i
ABF Convergence ABF Convergence
d h h h l Consider the time when the topology is
stabilized as time 0
U(0) and L(0) provide upper and lower bound U(0) and L(0) provide upper and lower bound
at time 0 on all corresponding elements of
states states
L
j
(0) d
j
U
j
(0) for all d
j
state at node j
L
j
(0) d
i
j
U
j
(0) L
j
(0) d
j
U
j
(0)
L
j
(0) update messages d
i
j
U
j
(0)
ABF Convergence ABF Convergence
d d
j
after at least one
d d update at node j:
d
j
falls between
L (1) d U (1)
dj
L
j
(1) d
j
U
j
(1)
d
i

d
i
d
i
j
d
i
j
d
i
j
:
eventually all d
i
j
that
l b d d b are only bounded by
L
j
(0) and U
j
(0) are
replaced with in replaced with in
L
j
(1) and U
j
(1)
Asynchronous Bellman-Ford: Summary Asynchronous Bellman Ford: Summary
Distributed: each node Distributed: each node
communicates its routing table to
its directly attached neighbors its directly-attached neighbors
Iterative: continues periodically or p y
when link changes, e.g. detects a link
failure
Asynchronous: nodes need not
exchange info/iterate in lock step! exchange info/iterate in lock step!
Convergence in finite steps,
independent of initial condition if independent of initial condition if
network is connected
Properties of Distance-Vector Algorithms Properties of Distance Vector Algorithms
G d f Good news propagate fast
Properties of Distance-Vector Algorithms Properties of Distance Vector Algorithms
Bad news propagate slowly Bad news propagate slowly
This is called the counting-to-infinity
problem
Question: why does counting-to-infinity happen?
Backup Slides
51
Using Virtual Circuit to Implement
N t k S i Network Services
I d t id f ti liti In order to provide some functionalities, a
network may choose virtual circuit, e.g., Virtual
Private Network (VPN) ( )
52
Using Datagram to Implement the
M t B i N t k S i Most Basic Network Service
A datagram network generally provides simple
services: the forwarding of packets from src to
dest.
We will focus on datagram networks which provide
best effort service
extensions to provide more services will be discussed in
53
extensions to provide more services will be discussed in
the multimedia networking part of the course