1
OSPF and MANET WG meetings, IETF64
OSPF MANET Design Team outbrief
November, 2005
Tom Henderson
Design team members: Emmanuel Baccelli, Madhavi Chandra, Thomas Clausen, Padma Pillay-Esnault, David Green, Acee Lindem, Joe Macker, Richard Ogier, Tony Przygienda, Abhay Roy, Phil Spagnolo
2
Outline
• Brief History• Problem Overview• Current Status• Recommendation
3
MANET and OSPF• A Mobile Ad Hoc Network (MANET) is a wireless
network operating in absence of (much) fixed infrastructure– multi-hop, time-varying wireless channels
• MANET WG produced four Experimental RFCs– none integrated with a commercial IGP
• Why MANET and OSPF?– Interest in using MANETs in transit network scenarios
(requiring redistribution)– Layer-2 MANET routing/bridging not always possible
or optimal
4
A brief history
• Initial problem statement drafted– draft-baker-manet-ospf-problem-statement-00
(expired)
• Initial drafts on an OLSR-like adaptation of OSPF, and database exchange optimizations
• WG decides to charter a design team (2004)– Meetings in San Diego and Washington, and design-
team mailing list
•Note: Expired drafts available at http://hipserver.mct.phantomworks.org/ietf/ospf/
5
Initial problem statement1. Focus on OSPFv3 and not OSPFv22. Compatibility with non-wireless OSPFv33. Intra-area extensions only4. Not focusing on transit network case, but
should not be precluded5. Scaling goal is 50-100 nodes on wireless
channel6. Leverage existing MANET work where possible7. Use RFC 3668 guidance on dealing with IPR
claims
6
Benchmark results• Current OSPF benchmarked in MANET
environment– draft-spagnolo-manet-ospf-design-00 (expired)
0
20
40
60
80
100
120
140
160
180
Hello LSU LSAck LSR D_DESC
Ove
rhea
d (
Kb
/s)
LSU overhead evenly divided between floods and retransmissions
7
Consensus reached
• Working on defining a new MANET interface type rather than a MANET area type– in parallel with existing OSPF interface types
• Focusing first on designing an optimized flooding mechanism for new LSA generation– using acknowledged (reliable) flooding– use Link Local Signaling (LLS) hello extensions
• Focus on two active I-Ds– draft-chandra-ospf-manet-ext-03.txt– draft-ogier-manet-ospf-extension-05.txt
• New complementary draft: – draft-roy-ospf-smart-peering-00.txt
8
Current status
• Two developed approaches, no consensus on single approach forward– Not a lot of debate, either
• Let’s look at the two approaches...
9
Overview of different approaches
• Both drafts focus on selecting more efficient Relay Node Sets (RNS) for flooding– A “Connected Dominating Set” (CDS)
• Both approaches perform topology reduction– MANET Designated Routers uses the CDS– Overlapping Relays via Smart Peering extension
• Differences– Source Independent vs. Source Dependent CDS– Use of Hellos or LSAs for dissemination of two-hop neighborhood
information– Differential (Incremental) Hello implementations
10
Review of draft-chandra*
* from Proceedings of OSPF WG, IETF-60
11
Review of draft-ogier*
* from Proceedings of OSPF WG, IETF 62
12
Design team evaluation software
• Based on quagga open source OSPFv3 routing daemon– http://www.quagga.net
• Runs as Unix implementation, or as GTNetS simulation (same quagga code)– http://www.ece.gatech.edu/research/labs/MANIACS/GTNetS/
• Implements both drafts, plus July version of Smart Peering
quagga
zebraUser Space
KernelIP
netlink, sysctl, ioctl
drivers
GTNetS (discrete event
network simulator)
modifiedlib files
glue to GTNets
quagga
modifiedospf6d
modifiedospf6d
Same Code
Implementation Simulation
13
Simulation findings
• Note: Technical Report and software available at – http://hipserver.mct.phantomworks.org/ietf/ospf
• Combination of flooding efficiency and topology control seems necessary– Both approaches produce comparable gains in flooding
efficiency – Topology reduction can make overhead scaling nearly linear
with number of nodes
• Topology reduction more straightforward with MDRs– MDR adjacencies anchored by CDS, similar to OSPF DR– Smart Peering uses heuristics to accomplish this, but
currently published approach has limitations
14
Simulation findings • OSPF MANET Interface overhead improvements
• GTNetS simulations
Routing Traffic Overhead
0
1000
2000
3000
4000
5000
6000
10 20 30 40 50 60
Number of Nodes
Ov
erh
ea
d (
kb
ps
)
Legacy OSPF PTMP
Gains due to efficientflooding only
Cisco’s Overlapping Relays
MDRs with full adjacencies and full LSAs
Efficient flooding plusadjacency reduction
MDRs with bi-connected adjacencies and full LSAs
15
Route Quality
1
1.4
1.8
2.2
2.6
3
10 20 30 40 50 60
Number of Nodes
(Rec
v p
kt +
Fw
d p
kt)/
(Rec
v P
kt)
Simulation findings
• Improvements do not sacrifice routing performance.
User Data Delivery Ratio
0.8
0.84
0.88
0.92
0.96
1
10 20 30 40 50 60
Number of Nodes
Del
iver
y R
atio
User data delivery ratio is highwith all three proposals
Reduced topology stillyields good routes
Legacy OSPF PTMP
Cisco’s Overlapping Relays
MDRs with bi-connected adjacencies and full LSAs
Legacy OSPF PTMP
Cisco’s Overlapping Relays
MDRs with bi-connected adjacencies and full LSAs
16
Neighbor Adjacencies
0
5
10
15
20
25
30
10 20 30 40 50 60 70 80 90 100 110
Number of Nodes
Nu
mb
er o
f A
dja
cen
cies
per
No
de
Routing Traffic Overhead
0
1000
2000
3000
4000
5000
6000
10 20 30 40 50 60 70 80 90 100 110
Number of Nodes
Ove
rhea
d (
kbp
s)
Simulation findings• Nearly linear overhead scaling made possible by
controlling the density of neighbor adjacencies
User data delivery ratio is highwith all three proposals
Reduced topology stillyields good routes
MDRs with bi-connected adjacencies and (B)MDR full LSAs
MDRs with bi-connected adjacencies and (B)MDR full LSAs
MDRs with full adjacencies and full LSAs
MDRs with full adjacencies and full LSAs
17
State of consensus
• We do not have complete consensus on moving forward with either draft’s core approach– Henderson, Ogier, Spagnolo have recently voiced
preference for MDR approach– Emmanuel Baccelli suggested that MPR flooding
offers better properties than CDS flooding, with at least as good topology reduction capabilities, and better routing stretch performance
– A few days ago, Acee Lindem communicated some new simulation results in support of Smart Peering
– Everyone else has abstained (per recent mailing list query)
18
Next steps
• Design team not making further progress– Two viable approaches have been specified
to a level of interoperability– Lack of agreement on the core approach for
flooding (MDR vs. Overlapping Relays)– Either approach could consider some
orthogonal elements from the other• e.g. two-hop neighbor discovery
– Suggest to open this discussion somehow to broader OSPF/MANET WG community