Evaluation of IP Fast Reroute Mechanisms

Evaluation of IP Fast Reroute
Mechanisms
Minas Gjoka,Vinayak Ram, Xiaowei Yang
University of California, Irvine
Outline





Motivation for IPFRR
Basic Concepts of IPFRR
IPFRR Mechanisms
Simulation Results and Analysis
Open Research Issues and Conclusion
COMSWARE 2007
Evaluation of IPFRR Mechanisms
The Need for IPFRR





Failure of a link/node results in disruption of network traffic.
These disruptions can last several seconds before the network
is re-converges
Emergence of low latency applications such as Video
Conferencing, VoIP etc.
The distributed nature of the network places an intrinsic limit on
the minimum re-convergence time.
How Fast ?



Sub Second : Achievable and a requirement for most IP n/ws
Sub 500ms : Low Latency Applications e.g : VoIP
Sub 50ms : Impossible 
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Convergence Times
Source : IPFRR Overview – NANOG 36.
COMSWARE 2007
Evaluation of IPFRR Mechanisms
IPFRR :: Introduction

IP Fast Reroute (IPFRR) refers to the set of technologies which allow a
link/node failure to be repaired locally using pre-computed backup routes

IPFRR provides a mechanism to compute and use backup/repair paths
computed using SPF/rSPF on IP networks

Analogous to MPLS FRR but differs in the mechanisms employed for the
backup routes

The aim of IPFRR mechanisms is to find a alternate path which would forward traffic
to the destination without using the failed link/node and without causing microloops.
COMSWARE 2007
Evaluation of IPFRR Mechanisms
IPFRR :: Approach

Node detects failure and invokes the pre-computed paths

Packet delivery restored (100%)

Node generated and floods LSP describing the failure

All nodes recompute SPF and load the new FIB using loop free
convergence

Maximum disruption < 100 mS
 Time to detect failure + a few mS
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Microloops
NOT CONVERGED YET
A
1
A
1
C
10
B
1
C
10
B
1000
1
5
5
D
D
Shortest path for
destination D
Shortest path for
destination D from B
Shortest path for
destination D from A,C
Possible Microloops
COMSWARE 2007
1
Classification of IPFRR Mechanisms
IPFRR Techniques
ECMP
COMSWARE 2007
LFA
Multi-Hop Techniques
U-Turns
Not-Via
Evaluation of IPFRR Mechanisms
Tunneling
ECMP



Simplistic Technique ..but it
works !
Exist when multiple equal
cost paths exist to the
destination traversing
different links.
Node Selection Criterion:

COMSWARE 2007
Evaluation of IPFRR Mechanisms
cost(Ni) =cost (Nj)
Loop Free Alternates (LFA)
Route to C: NH- C
Route to D: NH-C
Route to D: NHNH-C,
E, LFA–E
LFA-E



Uses the intuitive fact that traffic can be diverted via alternate unaffected
node towards the destination.
Works very well on networks having a high degree of connectivity.
On the failure of the B-C link, traffic is rerouted via
B-E-C.
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Loop Free Alternates


Must ensure that micro loops are not created either
during the link down or link up phase.
Conditions to be satisfied :
cos t(Ni)  link(Ni,S)  cos t(S)
cos t ( Ni)  cos t ( S )
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Loop Free Criterion
Downstream Path Criterion
Loop Free Alternates
COMSWARE 2007

On the failure of link S-E, S uses
its LFA i.e N to route traffic to D

Note : S is an invalid LFA for N as
it does not satisfy the
Downstream Path Criterion

Tradeoff between prevention of
microloops and number of LFA’s
observed
Evaluation of IPFRR Mechanisms
U-Turn Alternates

No LFA’s exist in this case, however
U-turn Alternates exist.

Selection Criteria:

cost(Ni) >= linkcost(Ni,S)+
cost(S)

S must be the primary next hop
for all N-D paths.

N has an LFA

U Turn traffic explicity or implicitly
marked
Improved coverage with increase in
complexity

COMSWARE 2007
Evaluation of IPFRR Mechanisms
Tunnels



COMSWARE 2007
The repairing node must ensure
that the packets sent through the
tunnel will not pass through the
failed component
The decapsulated packets
at the tunnel endpoint should
reach the destination router using
normal forwarding.
Directed forwarding allows the
repairing node to specify the
release point at the tunnel
endpoint.
Evaluation of IPFRR Mechanisms
Not-via Addresess
COMSWARE 2007

For each protected interface two
addresses are required, the normal IP
address and the Not-via address

Once a failure is detected, the repairing
router tunnels traffic towards the Not-via
address of the protected component

This mechanism requires the participation
of the intermediate routers on the repair
path since they must be able to tell the
link which they must avoid traversing from
the semantics of the Not-via address

They are able to provide full coverage with
an increased amount of complexity
Evaluation of IPFRR Mechanisms
Simulations

Failures simulated for all src-dst pairs for a
diverse number of topologies:

Real Topologies


Inferred Topologies


Abilene, Geant, Belnet
Rocketfuel
Synthetic Topologies

COMSWARE 2007
BRITE Topology Generator
Evaluation of IPFRR Mechanisms
Simulations (2)

Total coverage computed under link/node failure for







ECMP
LFA
U-Turns
LFA+U-Turns
Tunnels
Rule-2
Assumptions:


Link costs assumed to be symmetric
Topologies assumed to be static
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Methodology

For every source-destination pair we set the first link/node in
the primary path as failed

The set of neighbors “excluding the failed neighbor” is then
examined on whether it satisfies the necessary conditions for
an alternate routing path.

An IPFRR mechanism is said to provide coverage if the
packets can be safely routed towards the destination avoiding
entirely the failed link/node
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Link Failures
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Node Failures
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Diversity of Recovery Nodes – Sprint
(Link Failures)
Nodes : 315
Degree:6.17
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Diversity of Recovery Nodes – Geant
(Link Failures)
Nodes : 23
Degree:3.22
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Diversity of Recovery Nodes – Exodus
(Link vs Node Failures)
Nodes : 79
Degree:3.72
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Comparison
High
Not-Via
Tunnels
LFA
Complexity
LFA+U Turns
ECMP
Low
0
COMSWARE 2007
Coverage
Evaluation of IPFRR Mechanisms
100 %
Open Research Issues

What services need to protected and how fast?

How do we determine the coverage required ? Full /
Partial vs Complexity tradeoff.

How do we deal with multiple simultaneous failures ?

Behaviour of the IPFRR mechanisms for asymmetrical
link costs
COMSWARE 2007
Evaluation of IPFRR Mechanisms
Thanks !
COMSWARE 2007
Evaluation of IPFRR Mechanisms