CS 382: Network-Centric Computing

Distance Vector Routing

Zartash Afzal Uzmi

Spring 2023-24

ACK: Slides use some material from Scott Shenker (UC Berkeley) and Jim Kurose (UMass)
 Agenda
⚫ Finish Up: Distance Vector Routing
⚫ Complexity
⚫ Count to infinity
⚫ Split horizon and Poisonous Reverse

⚫ Comparison of DV and LS
⚫ Scale and complexity

2
Recap: DV Routing (generally: links have costs)
d(x,z) represents the total cost
y over the path from x to z
2
x 7 u z
1 cost(x,y) represents the cost
w from x to a directly attached
neighbor y

d(x,z) = min{ cost(x,y) + d(y,z),

cost(x,u) + d(u,z),
cost(x,w) + d(w,z) }

Bellman-Ford equation 3
 Three Node Network
to
x y z
x 0 2 7
from

y 2 0 1 y
z 7 1 0 2 1
x 7 z

d(x,z) = min{ cost(x,z),

cost(x,y) + d(y,z) }

4
 Three Node Network
to
x y z
x 0 2 3
from

y 2 0 1 y
z 7 1 0 2 1
x 7 z

d(x,z) = min{ cost(x,z),

cost(x,y) + d(y,z) }

5
 From Algorithm to Protocol
⚫ Algorithm:
⚫ Nodes use Bellman-Ford to compute distances

⚫ Protocol:
⚫ Nodes exchange distance vectors
⚫ Update their own routing tables
⚫ And exchange again….
⚫ Details: when to exchange, what to exchange, etc….

6
 Overall Approach
⚫ Each node 𝒙 begins with an estimate 𝒅(𝒙, 𝒚) for all nodes 𝒚
⚫ An “estimate” of the cost of the least-cost path from itself to all nodes 𝒚
⚫ Why are we calling it an estimate?
⚫ It’s the cost along the currently known least-cost path – may get updated

⚫ Each node, 𝑥, maintains the following routing information:

1. For each neighbor 𝑣, the 𝑐𝑜𝑠𝑡(𝑥, 𝑣) from 𝑥 to a directly attached neighbor, 𝑣
2. Node 𝑥’𝑠 distance vector, containing 𝑥’𝑠 estimate of its cost to all destinations
3. The distance vectors of 𝑥’𝑠 neighbors

7
 Overall Approach (Cont’d)
⚫ Each node sends a copy of its distance vector to each neighbor

⚫ Node 𝑥 receiving a new distance vector from a neighbor 𝑤

⚫ Saves 𝑤’𝑠 distance vector
⚫ Uses the Bellman-Ford equation to update its own distances vector

⚫ If node 𝑥’𝑠 distance vector changes, it sends its updated distance vector
to each of its neighbors

8
 x y z
This is y’s x - - -
“distance vector” y 2 0 1
z - - -

y
2 1
x 7 z
to
x y z x y z
x 0 2 7 x - - -
from

y - - - y - - -
z - - - z 7 1 0
9
 x y z
x 0 2 7
y 2 0 1
z 7 1 0

y
2 1
x 7 z
to
x y z x y z
x 0 2 7 x 0 2 7
from

y y 2 0 1
z z 7 1 0
10
 x y z
x 0 2 7
y 2 0 1
z 7 1 0

y
2 1
x 7 z
to
x y z x y z
x 0 2 3 x 0 2 7
from

y 2 0 1 y 2 0 1
z 7 1 0 z 7 1 0
11
 x y z
x 0 2 7
y 2 0 1
z 7 1 0

y
2 1
x 7 z
to
x y z x y z
x 0 2 3 x 0 2 7
from

y 2 0 1 y 2 0 1
z 7 1 0 z 3 1 0
12
 x y z
x 0 2 3
y 2 0 1
z 3 1 0

y
2 1
x 7 z
to
x y z x y z
x 0 2 3 x 0 2 3
from

y 2 0 1 y 2 0 1
z 3 1 0 z 3 1 0
13
 A More Complicated Case
⚫ The three-node network:
⚫ Everyone was neighbors with everyone else

⚫ What happens in a larger network?

14
 x y v z Four Node Network
x - - - - x y v z
y 1 0 - 3
y - - - -
z - - - - v - - - -
y z - 3 1 0
1 3
x z
2 1
to
x y v z v
x x y v z
0 1 2 -
from

y - - - - x - - - -
v - - - - v 2 - 0 1
z - - - -
15
 x y v z After first exchange
x 0 1 2 - x y v z
y 1 0 - 3
y 1 0 - 3
z - 3 1 0 v 2 - 0 1
y z - 3 1 0
1 3
x z
2 1
to
x y v z v
x y v z
x 0 1 2 -
from

y 1 0 - 3 x 0 1 2 -
v 2 - 0 1 v 2 - 0 1
Let’s find d(x,z), which
z - 3 1 0
represents the total cost over 16

the path from x to z

 x y v z Recomputing Tables
x 0 1 2 - x y v z
y 1 0 - 3
y 1 0 - 3
z - 3 1 0 v 2 - 0 1
y z - 3 1 0
1 3
Let’s now find d(z,x) at router
z to update its DV
x z
2 1
to
x y v z v
x y v z
x 0 1 2 3
from

y 1 0 - 3 x 0 1 2 -
v 2 - 0 1 v 2 - 0 1
z - 3 1 0
17
Let’s now find d(y,v) to x y v z Recomputing Tables
update the DV of y
x 0 1 2 - x y v z
y 1 0 - 3
y 1 0 - 3
z - 3 1 0 v 2 - 0 1
y z 3 3 1 0
1 3
x z
2 1
to
x y v z v
x y v z
x 0 1 2 3
from

y 1 0 - 3 x 0 1 2 -
v 2 - 0 1 v 2 - 0 1
z - 3 1 0
18
 x y v z Recomputing Tables
x 0 1 2 - x y v z
y 1 0 3 3
y 1 0 - 3
z - 3 1 0 v 2 - 0 1
y z 3 3 1 0
1 3
x z
2 1
to
x y v z v
x y v z Finally, update the DV of v
x 0 1 2 3
from

by finding d(y,v)
y 1 0 - 3 x 0 1 2 -
v 2 - 0 1 v 2 - 0 1
z - 3 1 0
19
 x y v z Recomputing Tables
x 0 1 2 - x y v z
y 1 0 3 3
y 1 0 - 3
z - 3 1 0 v 2 - 0 1
y z 3 3 1 0
1 3
x z
2 1
to
x y v z v
x y v z
x 0 1 2 3
from

y 1 0 - 3 x 0 1 2 -
v 2 - 0 1 v 2 3 0 1
z - 3 1 0
20
 x y v z Vectors distributed
x 0 1 2 3 x y v z
y 1 0 3 3
y 1 0 3 3
z 3 3 1 0 v 2 3 0 1
y z 3 3 1 0
1 3
x z
2 1
to
x y v z v
x y v z
x 0 1 2 3
from

y 1 0 3 3 x 0 1 2 3
v 2 3 0 1 v 2 3 0 1
z 3 3 1 0
21
Distance Vector Routing

Cost to neighbor routers

Distances to all destinations
Send own DV to all neighbors

Receive DVs from neighbors – apply Bellman-Ford

Send updated own DV to all neighbors

22
Questions?

23
 Other Aspects of Protocol
⚫ When to send messages?
⚫ When any of your distances d(u,v) change or the link cost changes to some
neighbor
⚫ Periodically, to ensure consistency between neighbors

⚫ What information to send?

⚫ Could send entire vector
⚫ Or just updated entries

⚫ Do we send every neighbor the same information?

⚫ Before answering: consider the following slides
24
 x y z

y 2 0 1
z 3 1 0

Failed Link 2 y
1
x 7 z

x y z
What happens now?
y 2 0 1
z 3 1 0
25
 x y z

y 4 0 1
z 3 1 0 routing loop!

y
1
x 7 z

x y z

y 2 0 1
z 3 1 0
26
 x y z

y 4 0 1
z 3 1 0 routing loop!

y
1
x 7 z

x y z

y 4 0 1
z 3 1 0
27
 x y z

y 4 0 1
z 3 1 0 routing loop!

y
1
x 7 z

x y z

y 4 0 1
z 5 1 0
28
 x y z

y 4 0 1
z 5 1 0 routing loop!

y
1
x 7 z

x y z

y 4 0 1
z 5 1 0
29
 x y z

y 6 0 1
z 5 1 0 routing loop!

y
1
x 7 z

x y z

y 4 0 1
z 5 1 0
30
 x y z

y 6 0 1
z 5 1 0 routing loop!

y
1
x 7 z

x y z

y 6 0 1
z 5 1 0
31
 x y z

y 6 0 1
z 5 1 0 routing loop!

y
1
x 7 z

x y z

y 6 0 1
z 7 1 0
32
 x y z

y 6 0 1
z 7 1 0 routing loop!

y
1
x 7 z

x y z

y 6 0 1
z 7 1 0
33
 x y z

y 8 0 1
z 7 1 0 routing loop!

y
1
x 7 z

x y z

y 6 0 1
z 7 1 0
34
 x y z

y 8 0 1
z 7 1 0 routing loop!

y
1
x 7 z

x y z

y 8 0 1
z 7 1 0
35
 x y z

y 8 0 1
z 7 1 0 routing loop!

y
1
x 7 z

“Count-to-Infinity” Scenario if c(x,z) x y z

was infinite
y 8 0 1
z 7 1 0
36
 x y z

y 8 0 1 Routing loop are possible in

z 7 1 0 case of link failures and cost
changes!
y
1
x 7 z

“Count-to-Infinity” Scenario if c(x,z) x y z

was infinite
y 8 0 1
z 7 1 0
37
 Why the loop?
⚫ Why would a router advertise a path back to the router which advertised
it?
⚫ Which doesn’t tell them anything new
⚫ Perhaps misleads the other router to think there is an independent path
⚫ And in case of failure (or link cost change), can lead to a loop
x y z

y 2 0 1 x y z
z 3 1 0
2 y
1 y 2 0 1
z 3 1 0
x 7 z 38
 How can we fix this?
⚫ Solution: if a router is using a next-hop’s path, then:

⚫ Do not tell them that you independently have a path to destination

⚫ Called “split horizon”

⚫ OR…

⚫ Tell the other router not to use your path (by telling them cost of ∞)
⚫ Called “poisoned reverse”
⚫ A node is poisoning its route for a destination in its update message to the other
node
39
 x y z

y 2 0 1
poisoned reverse z ∞ 1 0

2 y
1
x 7 z

x y z

y 2 0 1
z 3 1 0
40
 x y z

y ∞ 0 1
poisoned reverse z ∞ 1 0

y
1
x 7 z

x y z

y 2 0 1
z 3 1 0
41
 x y z

y ∞ 0 1
poisoned reverse z ∞ 1 0

y
1
x 7 z

x y z

y ∞ 0 1
z 3 1 0
42
 x y z

y ∞ 0 1
poisoned reverse z ∞ 1 0

y
1
x 7 z

x y z

y ∞ 0 1
z 7 1 0
43
 x y z

y ∞ 0 1
poisoned reverse z 7 1 0

y
1
x 7 z

x y z

y ∞ 0 1
z 7 1 0
44
 x y z

y 8 0 1
z 7 1 0

y
1
x 7 z

x y z

poisoned reverse y ∞ 0 1
z 7 1 0
45
 Homework Problem

Is it possible that we may still have a count-to-

infinity problem even with poisoned reverse?
Hint: Consider network topologies with more than 3 routers, and try to create
a count-infinity problem with poisoned reverse

46
 More on Loops …
⚫ Routing loops can be alleviated by limiting the maximum number of hops
a packet can take
⚫ Customary for the packet header to include a hop-limit
⚫ Source sets this to some value larger than the number of hops it believes are
needed to get to the destination
⚫ Each switch/router decrements hop limit field by 1
⚫ If this field reaches 0, then it does not forward the packet

⚫ Loops can be avoided by exposing complete paths, instead of just

distance vectors
⚫ This method is used by BGP (next lecture)
47
 DV Routing Complexity
⚫ Algorithm Complexity
⚫ In the worst case, how many computations
would a router have to perform when a
distance vector is received?

⚫ Message Complexity
⚫ How many messages would be exchanged
as neighbors share their distance vectors
(given e links)?

48
 DV Routing Complexity
⚫ Algorithm Complexity
⚫ In the worst case, how many computations
would a router have to perform when a
distance vector is received?
⚫ O(N)

⚫ Message Complexity
⚫ How many messages would be exchanged
as neighbors share their distance vectors
(given 𝑒 links)?
⚫ At least 𝟐𝒆 messages. Why?

49
 LS and DV: Key Difference
⚫ In Distance Vector Routing, each node
⚫ Shares routing information only with its directly connected neighbors
⚫ Routing information is the Distance Vector! What does that mean?
⚫ The shortest currently known distances to every node
⚫ These shortest distances may get updated when information propagates

⚫ In Link State Routing, each node

⚫ Floods/broadcasts routing information to all nodes in the entire network
⚫ What is routing information in this case?
⚫ Link state! What does that mean?
⚫ Cost of each directly connected link – only what a router knows for sure!
50
 Summary: LS and DV Comparison
⚫ Link-State:
⚫ Global flood: each router’s link state (# ports)
⚫ Send it once per link event, or periodically
⚫ Message complexity: 𝑶(𝒏. 𝒆), Algorithm complexity = 𝑶(𝒏𝟐)
⚫ Distance-Vector:
⚫ Send longer vector (# destinations) just to neighbors
⚫ But might end up triggering their updates
⚫ Send it every time DV changes (which can be often)
⚫ Can run into count-to-infinity problems
⚫ Overall message and algorithm complexity can vary

51
 Summary: LS and DV Comparison
⚫ LS vs. DV Tradeoff:
⚫ LS: Send it everywhere and be done in a predictable time
⚫ DV: Send locally, and perhaps iterate until convergence

52
 What happens when routers lie?
⚫ What if router claims a 1-hop path to everywhere?
⚫ All traffic from nearby routers gets sent there

⚫ How can you tell if they are lying?

⚫ You can’t in DV
⚫ In LS both ends of link send updates

⚫ Can this happen in real life?

⚫ It has, several times….

53
Q: Can we scale either distance-vector or
link-state routing to the scale of the Internet?
Problem: neither distance-vector nor link-state
routing will scale to the size of the Internet
Preview of Next Lecture

56
Questions?

57