Download Document

Network Layer
Provides for the transparent transfer of data between transport
entities.
Responsible for:
– Establishing,
– Maintaining,
– Terminating connections
Across the intervening communication facility.
– Routing packets within subnet,
– Congestion and deadlock problems,
– Subnet-host interface?
1
Network Layer (cont.)
Service Provided:
Issue
Initial setup
Destination
address
Packet
sequencing
Error
control
Connectionoriented
Required
Only needed
during setup
Guaranteed
Connectionless
Done by
network layer
(e.g., subnet)
Done by
transport layer
(e.g., hosts)
Not possible
Needed on
every packet
Not guaranteed
2
Network Layer (cont.)
Issue
Flow
control
Is option
negotiation
possible?
Are
connection
identifiers used?
Connectionoriented
Provided by
network layer
Connectionless
Not provided
by network layer
Yes
No
Yes
No
Summary of the major differences between
connection-oriented and connectionless service.
3
E-mail
FTP
…
TCP
IP
ATM
Data link
Physical
Running TCP/IP over an ATM subnet
4
Virtual Circuits vs Datagrams
• Trade-off between IMP memory, space and
bandwidth
• Virtual circuits (used in SNA)
– Attractive for interactive terminal environment
– Packet contains circuit # vs full destination address
– Problem:
If IMP fails and loses its memory, then every thing is
lost.
– Error control and sequence control
(handled by data link layer)
5
Virtual Circuits vs Datagrams
(cont.)
• Datagrams (used in DECNET, ARPANET)
– Allow IMPs to balance load
– Host (transport layer) inserts error control and sequence
control.
6
Comparison of datagram and
virtual circuit subnets
Connectionless
Circuit setup
Addressing
State information
Not possible
Each packet contains
the full source and
destination address
Subnet does not hold
state information
Each packet is
routed independently
Routing
Effect of
node failures
Congestion
control
Complexity
Suited for
None, except for packets
lost during the crash
Difficult
In the transport layer
Connection-oriented and
connectionless service
Connection-oriented
Required
Each packet contains
a short vc number
Each established vc
requires subnet table space
Route chosen when vc is
setup; all packets follow
this route
All vcs that passed through
the failed equipment
are terminated
Easy if enough buffers can
be allocated in advance
for each vc set up
In the network layer
Connection-oriented service
7
Virtual Circuit Subnet
H
Host
H
B
8 Simplex virtual circuits
H
C
A
D
H
E
H
F
IMP
(a) Example subnet
H
Originating at A
0 - ABCD
1 - AEFD
2 - ABFD
3 - AEC
4 - AECDFB
Originating at B
0 - BCD
1 - BAE
2 - BF
(b) Eight virtual circuits
through the subnet
8
Virtual Circuit Subnet (cont.)
(c) IMP tables for the virtual circuits in (b)
9
Virtual Circuit Subnet (cont.)
Virtual circuit field
Line
packet
H to A
4
A to E
3
E to C
1
C to D
2
D to F
0
F to B
0
B to H
0
(d) The virtual circuit changes as a packet progresses
10
Routing Algorithms
• Concern:
– Correctness
– Robustness
– Simplicity
– Fairness
– Stability
– Optimality
Note: Trade-off between fairness
and optimality.
11
Routing Algorithms (cont.)
• 2 classes of routing algorithm
– Nonadaptive
– Adaptive
• Centralized
• Isolated
• Distributed
12
Example packet-switched
network
5
To node
2
3
3
5
2
1
2
3
6
1
2
1
4
1
5
1
2
From 3
Node 4
5
6
1
=
1
5
1
4
5
2
2
=
2
2
4
5
3
4
3
=
5
3
5
4
4
4
5
=
4
5
5
4
4
5
5
=
5
6
4
4
5
5
6
=
Central Routing Directory
13
Example packet-switched
network:
Fixed
routing
Node 1 Directory
Node 2 Directory
Node 3 Directory
Dest.
2
3
4
5
6
Next node
2
4
4
4
4
Node 4 Directory
Dest. Next node
1
1
2
2
3
5
5
5
6
5
Dest.
1
3
4
5
6
Next node
1
3
4
4
4
Node 5 Directory
Dest. Next node
1
4
2
4
3
3
4
4
6
6
Dest.
1
2
4
5
6
Next node
5
2
5
5
5
Node 6 Directory
Dest. Next node
1
5
2
5
3
5
4
5
5
5
14
Dijkstra's Algorithm
B
2
A
2
6
C
7
E
2
F
1
3
3
D
2
2
4
G
H
B (2,A)
2
A
2
6
C (-)
7
E (-) 2
F (-)
1
3
2
4
G (6,A)
3
2
D
(-)
H (-)
Shortest path from A to D
15
Dijkstra's Algorithm (cont.)
B (2,A)
2
A
2
6
C (9B)
7
E (4B) 2
F (-)
1
3
2
4
G (6,A)
B (2,A)
2
A
2
6
3
2
D
(-)
H (-)
C (9B)
7
E (4B) 2
F (6E)
1
3
2
4
G (5,E)
3
2
D
(-)
H (-)
Shortest path from A to D
16
Dijkstra's Algorithm (cont.)
B (2,A)
2
A
2
6
C (9B)
7
E (4B) 2
F (6E)
1
3
2
4
G (5,E)
B (2,A)
2
A
2
6
3
2
D
(-)
H (9,G)
C (9B)
7
E (4B) 2
F (6E)
1
3
2
4
G (5,E)
3
2
D
(-)
H (8,F)
Shortest path from A to D
17
Dijkstra's Algorithm (cont.)
B (2,A)
2
A
E (4B) 2
2
6
C (9B)
7
F (6E)
1
3
2
4
G (5,E)
3
D(10H)
2
H (8,F)
Shortest path from A to D
18
Multipath Routing
Advantage over shortest path routing:
Possibility of sending different classes of traffic over
different paths.
e.g., Short packets routed via terrestrial lines,
whereas a long file routed via satellite.
Note: Disjoint paths improve reliability and
performance.
19
Multipath Routing (cont.)
A
B
E
C
G
F
I
J
D
K
H
L
Dest
A
B
C
D
E
F
G
H
I
K
L
First
Choice
A 0.63
A 0.46
A 0.34
H 0.50
A 0.40
A 0.34
H 0.46
H 0.63
I 0.65
Second
Choice
I 0.21
H 0.31
I 0.33
A 0.25
I 0.40
H 0.33
A 0.31
K 0.21
A 0.22
Third
Choice
H 0.16
I 0.23
H 0.33
I 0.25
H 0.20
I 0.33
K 0.23
A 0.16
H 0.13
K 0.67
K 0.42
H 0.22
H 0.42
A 0.11
A 0.16
Routing Table for Node J
20
Centralized Routing
• Adaptive routing scheme based on routing control
center (RCC).
• Use global information to optimize individual
IMPs.
• Lots of overhead.
• Vulnerable to failure of RCC.
• Timing problem during update.
• Heavy traffic through RCC.
• Advantage; Bursty traffic.
21
Centralized Routing
22
Isolated Routing
• Adaptive, but fewer vulnerability problems than
centralized routing.
• Routing information is not passed among IMPs,
but IMPs ``learn'' from past I/O schemes.
• Hot potato -- IMP tries to get rid of packet quickly,
by putting it on shortest output Q.
• Backward learning -- use frame headers to pass
information among IMP about history of packet.
23
Isolated Routing (cont.)
• Delta routing -- hybrid
between isolated and
centralized.
• Each IMP measures the
cost of each line (i.e.,
some function of delay,
Q length, utilization,
etc) and periodically
send a packet to the
RCC giving these
values.
To A
To H
To I
To K
Queueing within the IMP
24
Flooding
Every incoming packet is sent out on all outgoing
lines except the one it arrived on.
Note: Not practical in most applications. However,
flooding produces shortest delay (if overhead is
ignored)
– Selective flooding
– Random walk
Static routing fixed tables. Each possible route is
weighted
25
Flooding
example
(hop count=3)
Assume that a packet is
to be sent from node 1
to node 6.
26
Distance Vector Routing
• Each IMP exchanges information with its
neighbors.
• Routing table in each IMP contains
Destination
Best line
Distance or time
to destination
All IMPs
in network
27
Distance Vector Routing (cont’d)
A
B
E
C
G
F
I
J
D
K
(a)
Subnet
H
L
A
B
C
D
E
F
G
H
I
J
K
L
A
0
12
25
40
14
23
18
17
21
9
24
29
JA delay = 8
I
24
36
18
27
7
20
31
20
0
11
22
33
H
20
31
19
8
30
19
6
0
14
7
22
9
10 12
K
21
28
36
24
22
40
31
19
22
10
0
9
New estimated
delay from J
8
20
28
20
17
30
18
12
10
0
6
15
Line
A
A
I
H
I
I
H
H
I
K
K
6
(b) Input from A,I,H,K and the new
routing table for J
28
Link State Routing
• Discover its neighbors and learn their network
addresses
• Measure the delay or cost to each of its
neighbors
• Construct a packet telling all it has just learned
• Send this packet to all other routers
• Compute the shortest path to every other router
29
Learning about the Neighbors
•Send a special HELLO packet on each point -to point line
•Two or more routers are connected by a LAN
Routers
H
B
D
E
D
G
A
C
I
F
B
E
H
G
C
A
F
I
N
LAN
(a) Nine routers and a LAN
(b) A graph model of (a)
30
Measuring Line Cost
• Estimate delay of neighbors
– Send special ECHO over line
– Measure the roundtrip time and divide by 2
– Load consider?
31
Building Link State Packets
2
B
C
4
3
A
1
D
6
5
7
E
8
A
Seq.
Age
B 4
E 5
B
Seq.
Age
A 4
C 2
F 6
C
Seq.
Age
B 2
D 3
E 1
D
Seq.
Age
C 3
F 7
E
Seq.
Age
A 5
C 1
F 8
F
Seq.
Age
B 6
D 7
E 8
F
(a) A subnet
(b) The link state packets for this subnet
32
Distributing the Link State Packets
•Use flooding
Send flags
Source Seq. Age A
A
F
E
C
D
21
21
21
20
21
60
60
59
60
59
0
1
0
1
1
ACK flags
C F A C F
1
1
1
0
0
1
0
0
1
0
1
0
1
0
0
0
0
0
1
1
Data
0
1
1
0
1
The packet buffer for router B
33
Computing the Routes
• Use Dijkstra’s Algorithm
• OSPF Protocol (used in Internet)
34
Optimal Routing
(a) A subnet
(b) The sink tree for IMP B,using
number of hops as metric
35
Flow-Based Routing
Basic Idea:
For a given line, if the capacity and average flow are
known, it is possible to compute the mean packet
delay on that line from queueing theory. The
routing then reduces to finding the routing
algorithm that produces the minimum average
delay for the network.
36
Flow-Based Routing (cont.)
Destination
A
A
B
20
Source
C
20
B
10
20
A
D
20
20
E
50
10
F
(a) A network with line capacity
shown in Kbps
C
D
E
F
B
E
C
D
9
7
4
1
AB ABC ABFD AE
9
2
8
3
BA
BC BFD BFE
4
8
3
3
CBA CB
CD CE
1
3
3
3
DFBA DFB DC
DCE
7
2
3
3
EA EFB EC ECD
4
4
5
2
4
FEA FB FEC FD FE
F
4
AEF
4
BF
2
CEF
4
DF
5
EF
(b) The traffic in packets/sec
and the routing matrix
37
Flow-Based Routing (cont.)
i
1
2
3
4
5
6
7
8
l
Line (pkts/sec)
AB
14
BC
12
CD
6
AE
11
EF
13
FD
8
BF
10
EC
8
C
mC
(kbps) (pkts/sec)
20
25
20
25
10
12.5
20
25
50
62.5
10
12.5
20
25
20
25
T
(ms)
91
77
154
71
20
222
67
59
Analysis of the above network using a mean packet
size of 800 bits. The reverse traffic (BA, CB, etc) is
the same as the forward traffic
38
Hierarchical Routing
Dest. Line hops Dest. Line
Region 1
Region 2
1B
1A
2A 2B
1C
2D
2C
5B
4A
5C
3B
3A
4B
4C
5A
5D
5E
Region 3
Region 4
Region 5
1A
1B
1C
2A
2B
2C
2D
3A
3B
4A
4B
4C
5A
5B
5C
5D
5E
1B
1C
1B
1B
1B
1B
1C
1C
1C
1C
1C
1C
1C
1B
1C
1C
hops
1A
1
1B
1B
1
1
1C
1C
1
2
2
1B
2
3
3
1C
2
3
4
1C
3
4
5
1C
4
3
Hierarchical
2
Table
3
For 1A
4
4
4
5
5
6 Full Table
5 For 1A
39
Routing For Mobile Hosts
Wireless
cell
Home
agent
Mobile
host
Home LAN
Foreign LAN
WAN
MAN
A WAN to which LANs, MANs, and wireless cells are attached
40
Routing For Mobile Hosts (cont.)
The registration procedure for mobile hosts:
1. Periodically, each foreign agent broadcast a packet
announcing its existance and address. A newly
arrived mobile host may wait for one of these
messages, but if none arrives quickly enough, the
mobile host can broadcast a packet saying: “Are
there any foreign agents around?”
2. The mobile host registers with the foreign agent,
giving its home address, current data link layer
address, and some security information.
41
Routing For Mobile Hosts (cont.)
3. The foreign agent contacts the mobile host’s home
agent and says: “One of your hosts is over here.”
The message from the foreign agent to the home
agent contains the foreign agent’s network address.
It also includes the security information, to
convince the home agent that the mobile host is
really there.
4. The home agent examines the security information
which contains a timestamp, to prove that it was
generated within the past few seconds. If it is
happy, it tells the foreign agent to proceed.
42
Routing For Mobile Hosts (cont.)
5. When the foreign agent gets the acknowledgement
from the home agent, it makes an entry in its table
and informs the mobile host that it is now
registered.
43
Routing For Mobile Hosts (cont.)
1. Packet is sent to the
mobile host’s home address
4. Subsequent packets are
tunneled to the foreign
agent
3. Sender is given foreign
agent’s address
2. Packet is tunneled to the
foreign agent
Packet routing for mobile users
44
Broadcast Routing
45
Broadcast Routing (cont.)
• Source to send a packet to each destination.
• Flooding.
• Multidestination routing -- each packet contains a
list of destinations.
• Spanning tree -- copy an incoming packet onto all
the spanning tree lines except the one it arrived on.
46
Traffic Control
• Flow control
e.g., Sliding window technique
• Congestion control
e.g., Managing input and output buffers at
IMPs.
– Packets may be discarded or
– A choke packet sent.
• Deadlocks
e.g., No buffers are available.
47
Traffic Control (cont.)
To 2
To 3
To 1
Node 4
Output
Buffer
To 5
Intput
Buffer
Input and Output
Queues at node 4
To host
48
Traffic Control (cont.)
2
3
1
6
4
5
Host
The interaction of queues in a packet-switching network
49
The effects of congestion
throughput
(packets
delivered)
(a) Throughput
50
The effects of congestion (cont)
Average
Packet
Delay
(b) Delay
51
Congestion Prevention Policies
Layer
Transport
Network
Data Link
Policies
 Retransmission policy
 Out-of-order caching policy
 Acknowledgement policy
 Flow control policy
 Timeout determination
 Virtual circuits versus datagram inside the subnet
 Packet queueing and service policy
 Packet discard policy
 Routing Algorithm
 Packet lifetime management
 Retransmission policy
 Out-of-order caching policy
 Acknowledgement policy
 Flow control policy
Policies that affect congestion
52
Traffic Shaping
• Regulating the average rate (and burstiness) of data
transmission
• monitoring a traffic flow is reffered as traffic
policing
• Example: The leaky bucket algorithm used in ATM
networks (Turner 1986)
• In contrast, sliding window protocols limit the
amount of data in transmission
53
Traffic Shaping (cont.)
Host
Computer
Packet
Interface
containing
a leaky
bucket
Unregulated
flow
The bucket
holds
packets
Regulated
flow
Network
A leaky bucket with water
A leaky bucket with packet
54
Types of deadlock
(a) Direct store-and-forward deadlock
55
Types of deadlock (cont.)
(b) Indirect
storeandforward
deadlock
56
Types of deadlock
(c) Reassembly deadlock
57
Structured buffer pool for
deadlock avoidance.
Class N


Class K


Class 2
Class 1
Buffer space
for packets
that have
traveled k hops.
Command pool
(Class 0)
58