Transcript PPT Version
Design Considerations for a
Wireless OSPF Interface
draft-spagnolo-manet-ospf-design
Tom Henderson, Phil Spagnolo, Gary Pei
{[email protected]}
IETF-60 MANET WG meeting
August 2004
Problem statement
(draft-baker-manet-ospf-problem-statement-00)
• OSPF does not have suitable interface type
for MANET (wireless, multi-access subnet)
operation
– Leads to scalability problems with respect to
overhead (primarily flooding overhead)
• OSPF seems extensible to cover this case
– proposals have centered on a new interface type
– could be for IPv4 or v6, or both
Purpose
Design Considerations for a Wireless OSPF Interface
draft-spagnolo-manet-ospf-design
• examine fundamental performance
problems of OSPF in this environment
• study the performance trends of different
OSPF MANET proposals
OSPF analysis (Sec. 3)
• Multicast-capable Point-to-Multipoint
interface type is the benchmark
• Finding: LSU flooding and acknowledgment
is by far the dominant contributor to
overhead
– backed up by simulations as well
Methodology
• Simulation-based study using QualNet 3.7
– 802.11-based and Rockwell Collins USAP
TDMA
– Ricean fading model, no power control
– OSPFv2 implementation (validated against Moy
ospfd implementation)
– random waypoint mobility on square grid
• Performance metrics
– OSPFv2 overhead measured at IP layer
– User data delivery ratio
Scenario-independent parameters
• Number of nodes
• Number of neighbors per node
– averaged over all nodes
Network size
Network density
• Number of neighbor state changes per unit
Network churn
time
– averaged over all nodes
• (Number of external LSAs)
– not included in this study
OSPFv2 benchmark simulations
Mobility |
Low
Medium
High
-----------------------------------Hello
|
2.20
2.00
1.71
LSU-flood| 43.55
66.33
67.59
LSU-rxmt | 35.62
72.04
87.28
LSAck
|
3.70
7.28
9.16
LSR
|
0.04
0.10
0.20
DDESC
|
2.67
4.91
6.80
Total
| 87.80
152.70
172.70
Figure 8: Summary of overhead (kbps) at
the three mobility levels.
Dominant
overhead
factor
(reliable) Flooding optimizations
• Lin’s SI-CDS reduced overhead by 23%
against benchmark
• Lin’s SI-CDS plus ….
– Multicast ACKs reduced additional 32%
– Ogier’s receiver-based ACK suppression reduced
overhead by 8% (created more overhead)
– Originator-based LSA suppression reduced
overhead by 28%
– Retransmit-timer backoff reduced overhead by
24%
Unreliable flooding advantage
(draft-spagnolo-manet-ospf-wireless-interface)
| best SI-CDS
MPR w/out flag
| (reliable) (unreliable)
MPR w/ flag
(unreliable)
-----------------------------------------------------Total
|
110.0
17.70
28.40
Hello
|
1.65
1.79
1.79
LSA Flood
|
34.94
15.97
26.63
LSA Rxmt
|
57.13
-
-
LSAck
|
8.07
-
-
LSR
|
0.39
-
-
DDESC
|
7.81
-
-
Deliv ratio |
0.78
0.78
0.78
Figure 17: Summary of overhead (kbps) for comparison of
reliable and unreliable flooding.
Summary
• LSU flooding is by far the dominant
contributor to overhead
– can reliable flooding optimizations do better
than 50% reduction?
• unreliable flooding can provide up to 10x
reduction without sacrificing performance
– large numbers of external LSAs are a concern
• Database exchange optimization also may
be important in a frequently partitioning
network
Next steps
Fundamental design choices
• Broadcast-based interface
Network LSA
(desig. rtr.)
– provides abstraction
– may be most scalable for large networks
• Point-to-multipoint-based interface
– provides visibility into structure of MANET
– important for picking good entry points into
network, over bandwidth-constrained links
Layer-2 triggers
• Should we specify how implementations
might make use of layer-2 information?
– neighbor discovery suppression
– link quality issues
• How does this affect interoperability?
• Examples:
– A Triggered Interface: draft-corson-triggered00.txt (expired)
– PPPoE interface for link metrics: draft-bberrypppoe-credit-01.txt