Optimum Implementation of TI-LFA

and Segment Routing on SURFnet 8

RP #22

Peter Prjevara & Fouad Makioui

Supervisors: Marijke Kaat & Wouter Huisman

The Goals of Networks

ARPANET - 1974
2
https://en.wikipedia.org/wiki/ARPANET
 What IGPs Currently Offer?
● Sub-second convergence times (< 1000ms)

○ If effects BGP -> can take up to 3 minutes

● Reactive Approach

○ Fault Recognition

○ Information Flooding
3
Source: T Anji Kumar and MHM Prasad. Enhanced multiple routing configurations for fast ip network recovery from multiple failures.
The Goals of Networks Today
● Real time services
○ VoIP / Video
○ Cloud Software
○ Financial Trading
○ Experimental
Where might virtual reality lead us?
Not good enough David Ramos/Getty Images

4
Network in a Normal State

5
Failure Occurs

6
The Reactive Approach: Step 1

7
The Reactive Approach: Step 2

8
Segment Routing (or SPRING)
● Every node is labelled
○ Node ID
● Every link is labelled
○ Adjacency ID
● MPLS labels
● IGP to distribute Segment IDs (SIDs) creating a full mesh
9
 Protective Fast Reroute Solutions
Point of Local
rLFA TI-LFA
Repair 5
(PLR)
5
5

6
5 5 100

Juniper Networks, 2017. Juniper Tech Library - Fast Reroute Overview. 10

Feature Link / Node Protection

11
Link Protection

12
Node Protection

13
Link / Node Protection Summary

14
Fate Sharing

15
16
SURFnet8 Topology
● Interfaces that share the same fate due to:
○ Line card sharing
○ Optical path sharing
● Juniper Routers used that support:
○ TI-LFA
○ SPRING
○ Node Protection
○ Fate Sharing 17
 Optical cable sharing

Line card
sharing

18
Research Questions
1. How do different TI-LFA configurations perform when
implementing Node / Link Protection and Fate Sharing?

2. How do they affect the proposed metrics in IGP?

3. Is fate sharing necessary for all links that share the same
line card or optical layer?
19
Methodology
● Desk research
○ Understand novel concepts
● Define experiments
○ Create topology
● Analyse results
● Draw conclusions
20
Our Test Topology

21
Our Test Topology

22
Our Test Topology

23
Our Test Topology

24
List of Experiments
Experiment Sub Experiment

Baseline SR Without TI-LFA

With TI-LFA

Baseline SR with extra hop Without TI-LFA

With TI-LFA

Multiple link failures with source as PLR With a single backup path

With equal cost multi paths

With fate sharing

Link/Node Protection Observe the routing table on PLR

ECMP Metric Calculation Python Script Simulation / Paper analysis 25

Baseline SR
● SR without TI-LFA vs SR with TI-LFA

26
Baseline SR with Extra Hop
● SR without TI-LFA vs SR with TI-LFA (without crosslink)

27
Results

28
Multiple Link Failures 1

29
Multiple Link Failures 2

30
Multiple Link Failures 3

31
Multiple Link Failures 4

32
Multiple Backup Paths
Route output

145.125.124.6/32 (2 entries, 1 announced)

*L-ISIS Preference: 14
Next hop: 145.125.176.59 via ge-2/3/0.0 weight 0x1, selected
Next hop: 145.125.176.18 via xe-2/0/2.0 weight 0xf000
Next hop: 145.125.176.0 via et-1/1/0.0 weight 0xf000

● Maximum 8 backup paths

○ Equal Cost Multi Path (ECMP)

33
Experiment: Fate Sharing
● TI-LFA with fate sharing

34
Results

35
Multiple Broken Links

Average ~500ms

36
Multiple Broken Links

Average ~500ms

37
Multiple ECMPs

Average ~52ms

38
Fate Sharing Enabled

Average ~30ms

39
Link | Node Protection
Link protection

145.125.124.6/32 (2 entries, 1 announced)

*L-ISIS Preference: 14
Next hop: 145.125.176.59 via ge-2/3/0.0 weight 0x1, selected
Next hop: 145.125.176.18 via xe-2/0/2.0 weight 0xf000
Next hop: 145.125.176.0 via et-1/1/0.0 weight 0xf000

Node protection

145.125.124.6/32 (2 entries, 1 announced)

*L-ISIS Preference: 14
Next hop: 145.125.176.59 via ge-2/3/0.0 weight 0x1, selected
Next hop: 145.125.176.61 via ge-2/3/1.0 weight 0xf000
Age: 51 Metric: 25

40
41
42
43
Discussion
● TI-LFA works well with ECMPs, so ECMPs should be

implemented on SURFnet8

● Node protection effects ECMPs

● If fate sharing is enabled, routers might not use the post

convergence backup path

44
Recommendations
● Use low metrics on links between core routers

○ Default metric on the daisy chain (default 10)

○ Increase number of ECMPs

● Implement fate sharing

● Do not use node protection

45
Future Work
● Improve failure detection speed

○ Bidirectional Forwarding Detection (1 - 10ms)

● How will SRv6 perform in comparison with SR on MPLS?

○ Currently not implemented yet

46
Acknowledgements
● Special thanks:
○ Marijke Kaat and Wouter Huisman
○ SURFnet Team

47
Q&A

48