Restoration Routing in MPLS Networks Zartash Afzal Uzmi Computer Science and Engineering Lahore University of Management Sciences

Restoration Routing in MPLS Networks

Zartash Afzal UzmiComputer Science and Engineering

Lahore University of Management Sciences

Dec 20, 2005 Lahore University of Management Sciences 2

Outline Background

Network Services and QoS Architectural Requirements IP and MPLS

Introduction to restoration routing Local Restoration: Types of Backup Paths Local Restoration: Fault Models Backup Bandwidth Sharing Activation sets

Restoration routing framework Components Typical example Evaluation and Experimentation


Outline Background





Network Traffic and Services

Network Traffic today Not what it was 10 years ago Multimedia intensive

New and interactive applications are emerging Internet telephony Videoconferencing Streaming media (voice and video) Remote collaboration (e.g., remote desktop)

Many new applications are real-time More and more users of these applications

Burstiness behavior has changed over the years!


Current Network Architecture Internet is popular because

It is inexpensive Internet is inexpensive because

It uses resource sharing by means of statistical multiplexing

Current Internet architecture Uses packet switches with buffers Required buffer size is primarily determined by a random

traffic pattern Buffer size optimization

Too low High drop rate Too high High delay


Architectural Requirements

Emerging applications Two-way interactive communications One-way streaming media type applications

Under normal conditions We are worried about the two-way interactive applications

When resources fail We are also worried about the one-way applications

Current Internet architecture is not suitable for new and emerging applications New architectures are being researched


Architectural Requirements

New network architectures All circuit-switched? Mix of packet-switch and “circuit-switch-like”

Experience with networks Bigger buffers are required when there is more

randomness and more aggregation Should use circuits at places where we see more

randomness Example: 100x100 project

Edge network is packet-switched Core network is virtual-circuits


IP versus MPLS

In IP Routing, each router makes its own routing and forwarding decisions

In MPLS: source router makes the routing decision Intermediate routers make forwarding decisions A path is computed and a “virtual circuit” is established

from ingress router to egress router

An MPLS path or virtual circuit from source to destination is called an LSP (label switched path)


Outline Background





Restoration in IP network

In traditional IP, what happens when a link or node fails? Information needs to be disseminated in the

network During this time, packets may go in loops Restoration latency is in the order of seconds

We look for restoration possibilities in an MPLS network


QoS Requirements Bandwidth Guaranteed Primary Paths

Bandwidth Guaranteed Backup Paths BW remains provisioned in case of network failure

Minimal “Restoration Latency” Restoration latency is the time that elapses between the

occurrence of a failure and the diversion of network traffic on a new path

Path Restoration More LatencyLocal Restoration Less Latency


Restoration in MPLS

S 1 2 3 D

Primary Path

Backup Path

Path Protection

This type of “path Protection” still takes 100s of ms.

We may explore “Local Protection” to quickly switch onto backup paths!


Local Restoration: Fault Models

A B C DLink Protection

A B C D

A B C D

Node Protection

Element Protection


nhop and nnhop paths

Primary Path

Backup Path All links and all nodes are protected!

A B C D E

PLRPLR: Point of Local Repair: Point of Local Repair

nnhop

nhop


Opportunity cost of backup paths

Local Protection requires that backup paths are setup in advance Upon failure, traffic is promptly switched onto

preset backup paths

Bandwidth must be reserved for all backup paths This results in a reduction in the number of Primary LSPs

that can otherwise be placed on the network

Can we reduce the amount of “backup bandwidth” but still provide guaranteed backups?


BW Sharing in backup Paths

Example:

max(X, Y)

BW: Y

A B

C D

E F G

L1L1

L2L2

BW: XBW: X

Primary Path

Backup Path

XX XXXX

YY YYX+Y

SharingSharing


Activation Sets

A

B

C

D

E

Activation set for node B Activation set for link (A,B)

A

B

C

D

E


Outline Background





Restoration Routing Frameworks

We look to answer the following questions? Who computes the primary path? What is the fault model (link, node, or element protection)? Where do the backup paths originate? Who computes the backup path? At what point do the backup paths merge back with the primary

path What information is stored locally in the nodes/routers What information is propagated through routing protocols What if a primary path can not be fully protected

The goal is almost always to maximize bandwidth sharing Performance criteria is almost always the maximum number of

LSPs that can be placed on the network


Evaluation & Experimentation

Traffic Generation Use existing or emerging traffic models Consider call holding times and multi-service traffic

Rejected Requests Experiments Measure the number of rejected requests Simulate on various topologies

Network Loading Experiments Set link capacities to infinity Measure the total bandwidth required to service a given set

of requests Simulate on various topologies


Recent Trends Preemption of lower class traffic Multilayer recovery

We can “almost” deal with recovery at a single protocol layer

What if we intend to provide recovery at multiple protocol layers?

For multilayer recovery, we need to consider these additional issues: Interworking of layers Local information stored at each node of each layer Recovery provided by each individual layer Signaling mechanism from one layer to another Effects on bandwidth sharing (if sharing is used)


Thank You!

Questions & Answers


Extra Stuff!


Extent of BW Sharing: oAIS

Aggregate Information Scenario (AIS) Fij: Bandwidth reserved on link (i, j) for all primary LSPs Gij: Bandwidth reserved on link (i, j) for all backup LSPs

Optimized AIS (oAIS) – (Hij instead of Fij) Hij: Maximum bandwidth reserved on any one link by all

backup paths spanning link (i, j)

More Information propagated More potential for BW sharing


oAIS versus AIS: ExampleLSP Request-1 (src, dst, bw) = (A, C, 4)

A

F

D E

B C

G

FAB=4

HAB=4

GAF=4


oAIS ExampleLSP Request-2 (src, dst, bw) = (A, C, 5)

A

F

D E

B C

G

FAB=9

HAB=5

GAF=4

GAG=5

FAB=4

HAB=4


oAIS ExampleLSP Request-3 (src, dst, bw) = (D, E, 7)

A

F

D E

B C

G

FAB=9

HAB=5

GAF=4

GAG=5

FDE=7

GAF=7


oAIS ExampleLSP Request-4 (src, dst, bw) = (A, C, 6)

A

F

D E

B C

G

FAB=9

GAF=7

GAG=5

FDE=7Need to Evaluate cost of all possible backup paths?How much BW is shareable on (A, F)?

AIS:Shareable = max(0, GAF - FAB) = GAF - min(GAF, FAB) = 0Additional resv = 6

oAIS: (HAB ≤ FAB)Shareable = GAF - min(GAF, HAB) = 2Additional resv = 6 - 2 = 4

CIS: (link (A,B) knows BWred)Shareable = GAF - BWred = 7 - 4 = 3Additional resv = 6 - 3 = 3

HAB=5


Single Link Protection: Network 1








Single Node Protection: Network 1


Single Element Protection: Network 1


A Bandwidth Sharing Model

Primary Path

Backup Path All links and all nodes are protected!

(Simplified for the Link Protection Fault Model)Recall the definition of nhop paths

A B C DLink Protection


Bandwidth Sharing Model

Previous: Aij:= Set of all primaries traversing through (i, j)

Buv:= Set of all backups traversing through (u, v)

New definition (specialized for link protection case): Aij:= Set of all primaries traversing through (i, j)

Buv:= Set of all nhop paths traversing through (u, v)

µij:= Set of all nhop paths that span (i, j)

ijuv:= Buv ∩ µij (set of paths falling on (u,v) if (i,j) fails)



i

u v

j k

RED=7BLU=2

3

OLD MODEL:Aij = {R, B}Buv = {R, B, …}Aij ∩ Buv= {R, B}|| Aij ∩ Buv || = 2+7 = 9Un-shareable = 9Shareable = 10 - 9 = 1

GRN=3 (New Request)Guv = 10

NEW MODEL:Aij = {R, B}Buv = {nhij

r, nhijb, …} (nhops through (u, v))

µij = {nhijr, nhij

b, …} (nhops spanning (i, j))ij

uv = µij ∩ Buv= {nhijr, nhij

b}|| ij

uv || = 2 + 7 = 9 (Un-shareable)Shareable = Guv - || ij

uv || = 10 - 9 = 1



i

u v

j k

RED=7BLU=2

3

OLD MODEL:Aij = {R, B}Buv = {R, B, …}Aij ∩ Buv= {R, B}|| Aij ∩ Buv || = 2+7 = 9Un-shareable = 9Shareable = 10 - 9 = 1

NEW MODEL:Aij = {R, B}Buv = {nhij

r, nhjkb, …} (nhops through (u, v))

µij = {nhijr, nhij

b, …} (nhops spanning (i, j))ij

uv = µij ∩ Buv= {nhijr}

|| ijuv || = 7 (Un-shareable)

Shareable = Guv - || ijuv || = 10 - 7 = 3

GRN=3 (New Request)Guv = 10


Restoration in MPLS

Primary Path

Backup Path

Path Protection

MPLS path Protection may take 100s of ms, whereas MPLS Local protection takes less than 10 ms.

A B C D E

Documents

Restoration Routing in MPLS Networks Zartash Afzal Uzmi Computer Science and Engineering Lahore University of Management Sciences