1. Multi Node Label Routing – A layer 2.5 routing protocol
Addressing routing scalability issues
NSF Ignite Project

2. Routing protocol challenges
All Routing Protocols face
Scalability issues – dependency on network size
Flooding on information changes
High Churn rates
Packet Looping
Unreliable and unstable routing
Work with IP and avoid impacts of IP forwarding when necessary

3. What should the Solution look like?
Forward the packets - transparent to IP at layer 3 –
at layer MPLS-like?
But MPLS is not adequate for emergency situations
Set up time, failure recovery, dependency on routing protocols etc.

4. WHAT DO WE HAVE? Every network has a structure
CORE (Backbone (BB) routers)
DISTRIBUTION (DB routers)
ACCESS (AC Routers)
Tiers of routers for specific operations
If they do not have - it is easy to set up a (virtual) structure
Packets between access networks have to be forwarded via distribution and core
Flat routing based on logical IP addresses OR
Use the structure to forward packets

5. Using the Structure – the Tier Structure
Capture the attributes of the structure
LABELS
Based on the structure and connectivity -
Routers can have multiple LABELS
We will use the LABELS to route and forward

6. TIER Structure and LABELS
Let us introduce routers and assign LABELS that capture the structure properties
1.1
TierValue.UniqueID
TIER 1
1.1
BB Routers
1.2
1.3
TierValue.UniqueID
UniqueID = parentID: ChildUniqueID
2.1:1
DB Router Set 1
2.1:1
2.3:1
DB Router Set 2
2.2:1
2.3:2
TIER 2
AC Router Set 1
3.3:1:1
3.1:1:1
AC Router Set 2
3.3:2:1
3.2:1:1
TIER 3
3.1:1:1
TierValue.UniqueID
UniqueID = grandparentID:parentID: ChildUniqueID
The label structure is TierValue UniqueID
Unique ID carries the parent child relationship Grandparent : Parent : Child
Tree like - Can be used for routing and forwarding
TierValue provides a level of aggregation

7. Routing and Forwarding in the Structure
Each node records a neighbor table
Neighbor Table of 2.1:1
1.1
1.2
1.3
Label
Port
3.1:1:1 1
2.3:1 2
1.1 3
2.1:1
2.3:1
2.2:1
2.3:2
3.3:1:1
3.1:1:1
3.3:2:1
3.2:1:1
Frame from Source 3.1:1:1 to Destination 3.3:1:1
At 3.1:1:1 check my neighbor table
3.3:1:1 is not my direct child or parent (compare uniqueIDs 1:1:1 with 3:1:1)
send to my parent
packet reaches 2.1:1

8. Routing and Forwarding in the Structure
Neighbor Table of 2.1:1
1.1
1.2
1.3
Label
Port
3.1:1:1 1
2.3:1 2
1.1 3
2.1:1
2.3:1
2.2:1
2.3:2
3.3:1:1
3.1:1:1
3.3:2:1
3.2:1:1
Frame from Source 3.1:1:1 to Destination 3.3:1:1
At 2.1:1 – this node checks if the destination is relation of any one of my neighbors –
3.3:1:1 is a child of 2.3:1 – so node 2.1:1 forward packet to port 2
if this was not the case – send to my parent

9. Routing and Forwarding in the Structure
1.1
1.2
1.3
2.1:1
2.3:1
2.3:2
2.2:1
3.1:1:1
3.3:1:1
3.3:2:1
3.2:1:1
At each node - comparisons with the neighbor table entries
The packet will be directed to the proper port
No match – send to my parent
No routing tables – like OSPF or BGP
No flooding of routing updates –
local neighbor information exchanged
Tier 1 may get more traffic – that normally happens
Tier 1 – partial mesh – max 2 hops
(Seattle POP – AT&T network 44 routers)
Neighbor table can record up-to max 2 hops (working)

10. Routing and Forwarding in the Structure
1.1
1.2
1.3
Link / node failure –
follow another decision path
no flooding, low churn rates
Example link between 2.3:1and 2.1:1 fails
only node 2.1:1 record a change
2.1:1 would send the packet to 1.1
Packet will take path 1.1,
1.3,
2.3:1
2.1:1
2.3:1
2.2:1
2.3:2
3.3:1:1
3.1:1:1
3.3:2:1
3.2:1:1

11. How to implement in current networks?
Solution should be Deployable and easy/smooth transition
As a layer 2.5 routing protocol – similar to MPLS
Forward traffic from IP networks connected at the edge
Learning edge IP Networks <-> Labels of nodes connected to the IP Networks
Disseminated this information to all edge nodes

12. Multi Node Label Routing (MNLR)
Implementation Details
DEMO available with 30 nodes on GENI - SAI

13. MNLR Domain – Edge IP Networks
End IP network 2
End IP network 1
MNLR Edge Node
MNLR Edge Node
MNLR Core Node
MNLR Core Node
MNLR Edge Node
MNLR Core Node
MNLR Edge Node
End IP network 4
MNLR Domain
End IP network 3

14. Tasks completed MNLR coding in C and implementation over GENI
Automation of script to setup in GENI and performance metric collection
BGP performance over Emulab

15. Demo 27 nodes MNLR vs BGP
MNLR operation in a 27 node scenario / BGP operation

16. Automated Process Flow
GIT HUB
Manifest Rspec.xml
Parse
Topology Info
GIT update on the Local Repo
Copy Code to all GENI Nodes
Trigger MNLR on all Nodes
Tier Allocation and Command Formation
Code Compilation/ Error Check
Trigger Performance Analysis Test Cases using Iperf
Notification

17. Software Defined Networks
SDN can be used for
Label assignment and connectivity information dissemination at tier 1
Kind of management
End IP address to Label map dissemination to all edge MNLR nodes
Placement of SDN controllers

18. SDN Controllers SDN C1 SDN C2 SDN C3 SDN C5 SDN C4 IP NW 1 IP NW 4

19. MNLR Routers (Edge and Core)
MNLR Edge Node
MNLR Edge Node
MNLR Core Node
MNLR Domain
End IP network 2
End IP network 3
End IP network 1
End IP network 4
CONFIGURATION
Labels assignment to all nodes in an MNLR domain.
Hello Messages (neighbor table) and Connect messages (label assignment at tiers 2 and 3
CORE CONNECTIVITY
Periodic ‘hello’ messages by MNLR nodes to neighbors -> Neighbor Table
EDGE CONNECTIVITY
End IP network address dissemination to all egress/ingress MNLR nodes
EDGE ENCAP/DE-ENCAP
Encapsulation of incoming IP packets in special MNLR headers
De-encapsulating MNLR packets to deliver IP packet to the destination IP networks
CORE FORWARDING
Forwarding of MNLR encapsulated packets towards egress MNLR nodes based on MNLR labels
MNLR SRC | MNLR DST IP PACKET

20. Optimization with MNLR
With one-hop neighbors in neighbor table
How many hops in the neighbor table
Optimization problem
Only same tier neighbors

21. Implemented at layer 2.5 in the geni testbeds
Evaluated for several topologies and compared with BGP
Quagga BGP run on Emulab

22. SETTING UP THE TOPOLOGY AND CONFIGURATIONS
TIME CONSUMING
AUTOMATION SCRIPTS
CAN SET UP LARGE TOPOLOGIES IN GENI

23. Test Cases IPN1  IPN2 IPN1  IPN4 IPN1  IPN6 IPN1  IPN7
N5 – N3 – N4 – N8
IPN1  IPN4
N5 – N3 – N0 – N1 – N13 – N16
IPN1  IPN6
N5 – N3 – N0 – N2 – N24 – N26
IPN1  IPN7
N5 – N3

24. Test Case – Link Failure
IPN1  IPN2
N5 – N3 – N4 – N8
After link failure between N3 and N4, follows the path:
N5 – N3 – N0 – N4 – N8

25. BGP routing table size
Routing table size is equal to the number of networks in the topology
All nodes have routing information for all networks in the topology
For this topology, the routing table size is 29

26. Link Failure Example We fail the link between Nodes 0 and 1:
Churn Rate: 18/27
Convergence Time: 156 secs (HELLO INTERVAL IS 60 secs)
In comparison with MNLR:
Churn Rate is equal to the number nodes that experience a link failure (2 in this case)
Convergence Time is equal to the number of hello times required to determine a link failure plus the time to update the affected nodes’ neighbor tables (2 second hello times, 3 hellos for link failure)

27. Future work Video images Files Round trip time Reliability
Demo in Early May

28. Demo
MNLR vs BGP

29. Discussions Truly clean slate technology
No distance vector, link state, path vector
Decouples dependency of routing and operations from network size
Scalable
Complements SDN – modularized control plane operations
MNLR is modular
Can be used for intra or inter-AS routing
Suggestions

30. BACKUP SLIDES

31. Routing Structure and Modularity
Let us introduce routers and assign IDs that capture the structure properties
ISP A
1.2
Seattle POP
1.2:1
New York POP
1.2:2
Tier 1
1.1
BB Routers
1.2
1.3
Tier 2
DB Router Set 1
2.1:1
2.3:1
DB Router Set 2
2.3:2
2.2:1
Tier 3
AC Router Set 1
3.3:1:1
AC Router Set 2
3.1:1:1
3.3:2:1
3.2:1:1
Devices 4. :::
Chicago POP
1.2:3
Forward between 3.3:1:1 – 3.3:2:1 – via 2.3:1 and 2.3:2
Forward between 3.3:1:1 in Seattle POP – and NY POP – packet leaves the Seattle cloud – address will be 1.2:1(3.3:1:1). The device in NY POP will accordingly have an address 1.2:2{3.3:1:1…) – Name services