E E 681 - Module 16 Path-oriented Survivable Mesh Networks W.D. Grover TRLabs & University of Alberta © Wayne D. Grover 2002, 2003.

E E 681 - Module 16

Path-oriented Survivable Mesh Networks Path-oriented Survivable Mesh Networks

W.D. Grover

TRLabs & University of Alberta© Wayne D. Grover 2002, 2003

E E 681 - Module 15 © Wayne D. Grover 2002, 2003

2

Three basic concepts about end-to-end path protection/restoration:

Shared Backup Path Protection– each working path has a (single) fully-disjoint backup route pre-determined at path-

provisioning time.– When needed a protection path is cross-connected from spare channels along the

backup route. – like “1+1 APS with a shared backup”– same end-node activated reaction regardless of where failure occurs on working path.

(“True”) Path Restoration– adaptive, failure-specific, response to failure and network state.– for each failure scenario the set of affected end-nodes are simultaneously restored with

an MCMF-like response. • has cognizance of the “mutual capacity” issue and global (or self-organized) coordination.

– Allows reuse of working capacity on surviving portion of failed paths.– Capacity design to assure 100% restorability to all defined scenarios.

GMPLS “mass redial”– completely ad-hoc reliance on mass independent re-provisioning attempts.– no co-ordination, self-organization, or other way to address mutual capacity

considerations – inherently unassured, best-efforts, unpredictable outcomes.


3

Some initial notes on each:

Shared Backup Path Protection– most dominant current paradigm for “survivable routing.”– preferred for router-centric control paradigm where “network is dumb, edges are smart.”– high dependency on conventional software, databases, abd current ideas of Internet like

global state dissemination, etc. – advantage in fully optical networks is that we don’t need rapid fault location.

(“True”) Path Restoration– theoretically most efficient possible scheme.– not currently popular with industry due to perceived complexity. – self-organizing distributed protocol for MCMF-like adaptive performance developed by us.– may return to importance in context of adaptive-second line of defence strategy for ultra-

high availability, or, for maximal recovery from arbitrary-attack failure scenarios (9/11 etc.)

– true self-organization concepts currently too different from conventional software/ messaging paradigms for “distributed interaction.”

GMPLS “mass redial”– industry currently in two camps:

• those that know this is a disaster waiting to happen if it is positioned as the only survivability mechanism needed.

• those that don’t understand the issue of mutual capacity yet, or just want to overprovision as much capacity as needed to have a good chance of restoration

– characterized in recent M.Sc. Thesis by G. Kaigala (and Globecom 2003 paper).


4

Shared Backup Path Protection• “Protection” scheme (pre-determined fixed backup paths)

• End-to-end path protection

• Single protection path for each working path

• Spare capacity can be shared between disjoint working paths

• Restoration of path 1

• Restoration of path 2

Failure scenario 1

Failure scenario 2

Working path

Working path

Spare capacity re-use


5

A slightly more general example of SBPP

Green-blue-yellow protection sharing (x3)

Green-red sharing (x2)


9

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

5.0

0 1 2 3 4 5 6Sharing Limit

To

tal w

ave

len

gth

-dis

tan

ce c

ap

aci

ty

B-37B-40B-43B-46B-50

Cost of Limiting the Maximum Sharing Relationships*

Average total capacity increase for

F =2: 31.4%

Sharing Limit of 3 may be acceptable.


F=1: 150.4%


F =3: 5.1%

Sharing Relationship Limits of 1 and 2 yield total capacity increases that may be unacceptable.

* Results for test network family with varying nodal degree based on a 25 nodes - 50 spans master network


10

Qualitative Appreciation of SBPP vs Span Restoration

Multiple restoration paths

Characteristics(relative to 1+1 APS)

Sp

an R

esto

rat i

on

SB

PP

Effect on capacity efficiency

Effect on availability

Sharing of spare capacity

Localized response

Distributed/adaptive restoration

Sharing of spare capacity

Single backup path

End-to-end response

Non-adaptive restoration

+ +

+ + +

+ ++

+ +

+-

- +

-

--

- - --

no effect

no effect


11

“True” path restoration: what we mean

• The set of working paths severed by a span cut are restored by establishing a set of replacement paths end-to-end, simultaneously, between each O-D pair affected.

• The replacement paths are formed on-demand using only shared spare capacity (and possibly released working capacity (stub release).)

– There is no dedicated reservation of a 1-for-1 backup path for each working path.

• Path restoration is equivalent to abandoning the damaged pre-failure paths entirely and rapidly re-provisioning new paths end-to-end.

• Path restoration distributes the impact of failures and the recovery effort more widely over the network as a whole and therefore generally permits greater efficiency in spare capacity design.

• The capacity design and real-time restoration problems for path restoration are considerably more complex than span-restoration

– the fall-back to each O-D pair creates a capacitated multi-commodity max-flow problem.


12

Comparative illustration of span versus path restoration

Pre-failure

3 service paths


13

Failure occurs

Comparative illustration of span versus path restoration

All 3 service paths are lost

until …


14

Span restoration reaction

First look at a span restoration reaction … (1)

Note: example only, exact routes depend on working and spare capacities

All 3 service paths are lost

until …

failed working links on failed

span are restored by

span restoration


15

A span restoration reaction …(2)

Loopback / backhaul

Loopback / backhaul

This restoration path could stop hereThis restoration path could stop here


16

Now view a path restoration reaction...

Same failure occurs


17

A path restoration reaction …with “stub release” (1)

Path restoration action

Stub release


18

• Stub release is an option / issue which does not exist in span

restoration.

• From a capacity design standpoint it is preferable to have stub-

release.

• From an operational viewpoint stub release complicates things:

– a means of automatic signaling needed to rapidly release the surviving working

“stub” capacities,

• AIS (Alarm inhibit signal) usually serves nicely for this, however

– after physical repair, the reversion process is more complex.

• Ironically, without stub release, a reserve network capacitated to

support span-restoration may not be restorable under path

restoration ! • Class: Can you think why?

Notes about stub release in path restoration


19

Optimal Capacity Design for Optimal Capacity Design for path- restorationpath- restoration

Approach that follows is to first develop a “master formulation” that can model joint / non-joint designs and cases with / without stub release, then discuss modifications for each special case.


20

• The “master formulation” for path-restoration allows for:– modularity– joint optimization of working path routing– stub release or non-stub release

• The master formulation requires as inputs:– point-to-point demand

– a set of eligible distinct working routes for every (O-D) pair r– a set of eligible distinct restoration routes for every (O-D) pair r

for each failure scenario i .

• It solves for:– the amount of working flow on each working route for each O-D pair

(working flows may be split over several routes)– the working, spare, (and module) capacity totals on each span– the composite restoration path-set for all affected demands in each failure scenario

Note: in span restoration this isfor every span,

here it is for every OD pair.

Variations and options within the master formulation for path restoration


21

mCj

rxi

,r qg

0

,si j

, prfi

,,r p

i j

riP

* some variables become pre-computable parameters in the variations that follow

Input data

Intermediate (internal) variables

Design output variables

Cost of mth modulesize on span j.

D

,r qj

S

rd

rQ

Set of all point to pointdemand quantities, indexed

by r

amount of demand on relation r

Set of all spans betweenmesh cross-connection points

Set of eligible working routes for relation r

Encodes routes in= 1 if span j is in qth route

for relation r

rQ

Set of eligible restoration routes for relation r

upon failure i.

= 1 if span j is in pth routefor relation r upon failure i

Stub release quantity on span j

from failure i

Amount of demand loston relation r for

failure i

mnj

jw

js

No. of operatingworking and spare

links (channels)on span j

No. of modules oftype m to install

on span j for min cost

mZCapacity of mth

module size

Working andrestoration

routingsolutions

N.B. “relation” = “OD pair”

Parameters and variables in path-restorable capacity design(in the master formulation)*


22

B

A

fir,p

eligible restoration routes for A-B after the failure of span i,

workingroute for A-B

gr,q

Xir

span i

Qr = 1

Pir = 2

r = A-B

C

D

Orientation to the path restorable design context(variable and parameters)


23

m mj j

m j

Minimize C n

M S

Cost of modules of all sizes placed on all spans

S. t.

;i r S D, ,r q r qi

q

rg xi

rQ

Defines the amount of damaged working flow for each relation under each failure scenario

,

r

r q r

q

g d

Q

r D

, ,r q r qj j

r q

g w

rD Q

j S

All demands must be routed

Working capacity on spans must be adequate

(1)

(2)

(3)

Master formulation for path-restorable capacity design


24

, , ,

0

,0

r q r q r qj i

r q

gsi j

rD Q

With stub release

Without stub release

2( , )i j

i j

S

rP

,

p i

p rrf xii

;i r S D

0

rP

, ,,,

pr i

r p prf s si i i jjj

D

Restorability of working flows for each relation

Spare capacity on spans must be adequate (see note on stub release)

M

m

mmjjj Znws

1

j S Modularity of installed capacity

(4)

(5)

(6)

(7)

Master formulation for path-restorable capacity design (2)


25

O D

Relation r

Route q

Failure span i

Other span j Working flow gr,q

, 1r qj

, 1r qi

Span j enjoys a stub release “credit” of spare capacity = g r,q for any

failure on span i such that: , ,( 1) ( 1)r q r qj i true

, , ,0

,r q r q r qj i

r q

s gi j

rD Q

Understanding how the formulation effects “stub-release”


26

• joint optimizedrouting makes alargedifference in span restoration

Typical result comparing span and path-restorable network designs

4000

4200

4400

4600

4800

5000

1 2 3 4 5 6

To

tal N

etw

ork

Cap

aci

ty (

Lin

ks)

Combined workingand spare capacityoptimization

Spare capacityoptimization only

Design Case

Span restorablePath restorablePath restorablewith stub release

“non joint”

“joint” designs

• joint-span is aboutas efficient as non-joint path

• joint designadds relativelylittle benefit to path restoration


27

• If integer (“ideal”) but non-modular capacity is desired:

– change objective function to cost-weighted sum of spares (and / or working, if joint)

– drop set M (the family of modularities), variables and constraint (7)

• If non-joint design is desired:

– drop (1), (2), (3), and (6)

– pre-compute all and as input parameters based on the pre-defined routing

– pre-compute all stub-release quantities according to (6)

• If stub-release is not desired:

– drop (6), i.e., set all = 0

mnj

rxi jw0,i js

Variations and options within the master formulation

0,i js


28

• For each OD pair relation:– generate the set of “all distinct routes” between O-D with effectively unlimited

hop limit, e.g, H ~ 3/4 |S| and store in a (large) temporary routes file.

– sort the routes by increasing geographical length

– If joint formulation: • take first N routes (a budgeted number) as the set of eligible working routes for the

formulation. (Do not remove from file).

– If non-joint formulation:• take the single shortest route for the working paths between O-D

– For both joint or non-joint generate eligible restoration route-sets:• Repeat (for each active O-D pair):

– step out one span onto the shortest route, i – find first K routes in routes file which do not include span i as eligible restoration routes– remove chosen routes from file– go to next span along shortest route

Until last span in shortest route

• merge all routes found as eligible route-set for restoration of relation r.

A route-generating method for path restorable design formulation


29

• What does this achieve ? – This procedure is effective in mediating the following trade-off in populating the route-

sets:

“All distinct routes”A single budgeted number

of distinct routes

Far too large DAT file sizesfor realistic AMPL / CPLEX runs

Insufficient diversity / uniform disjointness of route-sets found by Depth First Search, infeasibilities arising with even large route-set

budgets

project topic: write program to statistically sample the large eligible route space

Route generating method for path restorable formulation (2)


30

O D

Boundary of a network graph

Outright infeasibility possible if all ‘budgeted’ routes neck

down onto one span in common

The preceding process for selecting routes stays within a budget but avoids this problem by

repeatedly pushing out the grey envelope as itproceeds in the direction across the network between

O-D nodes.

( Root end

of DFS tree )

( Leaf end

of DFS tree )

Observed tendency from taking budgeted number of distinct routes of successive length found by DFS


31

A

B

C D

E

F

G

H

C

E

F

G

H

D

B

A

Relation Lost Capacity

A - D 3

B - D 2

G - D 1

Restoration Path

Unused Spare Capacity

Possible restoration pathset Preferred restoration pathsetRestorability = 6/ 6= 100%Restorability = 5/ 6= 83%

X Failed Span

(a) (b)

• example shows same span-failure affecting 3 OD-pairs

• only spare capacity and restoration paths are shown for simplicity

• ad-hoc set of independent replacement paths

• -incomplete restoration

• collectively co-ordinated set of replacement paths

• -complete restoration

Why ad-hoc replacement-route finding (“mass redial”)is not an assured form of path restoration


32

• (1) A path restorable network inherently provides a response to node-failure and multiple span failures

- 100% restoration not guaranteed

- span restoration needs special extensions to the distributed protocols to respond to these situations as gracefully

• (2) Path restoration also copes more gracefully with the multiple logical span failures arising from nodal “bypass” situations.

terminated flows

express or “bypass” flows

same cable

Other comments on path restoration


33

physical picture =

logical picturefor span restoration

= (A-B)

(A-C)

simultaneous dual logical span failures

A C

B

AC

B

(A-B)

span failure

generic issue / phenomenon: physical to logical layer fault multiplication • for path restoration, however,

same set of end-node OD pairfailures arise in either case

Issue of nodal bypass and fault multiplication in span restoration - not a problem for path restoration


34

• Recent findings indicate that the capacity benefit of path restoration (over span- restoration) may be considerably less than hoped for in low degree networks.....

chain

chain

higher degree mesh component

consider:

- intra chain demands

- demands that cross only one mesh span or chain----- >

can do no better -( spare capacity - wise) than with span restoration )

• for many demand pairs in a low degree network their path restoration solution is no different than span restoration in the mesh that results from collapsing all degree-2 chains.

Other comments on path restoration

E E 681 - Module 16 Path-oriented Survivable Mesh Networks W.D. Grover TRLabs & University of Alberta © Wayne D. Grover 2002, 2003.

Documents

working path spare capacity

true path restoration

path provisioning time

reuse of working capacity

yellow protection

capacity design

mutual capacity issue

issue of mutual capacity