Transcript telephony operator

Lecture 18. SDH, DWDM, IP over SDH
D. Moltchanov, TUT, Spring 2015
Outline
SDH capacity and protection design
SDH ring bandwidth design
DWDM restoration design
IP over SDH two-layers design
SDH: capacity design
Telephony: routing of calls again
Synchronous digital hierarchy (SDH)
Replacement for PDH (E1/E3)
Strictly synchronized (by atomic clocks)
Allow for country wide networks
Three types of nodes
Terminal multiplexers (TM)
Digital cross-connects (DCS)
Add-drop multiplexers (ADM)
Possible topologies
Mesh: TM and DCS
Ring: ADM nodes
Telephony: topology example
Telephony: routing of calls again
Isn’t STM-1 155Mbps too huge?
Subrates for carried demands
Virtual containers (VC) in SDH
Classic E1/E3 can also be connected
Things are getting complex…
Telephony: routing of calls again
SONET/SDH are transport networks for
Trunks from digital telephony networks (including cellular)
Leased-line services
Originally no support for IP networks
IP: via Packet over SONET/SDH (PoS) is very important nowadays
Specifics designing SDH/SONET
Analogous to those discussed previously
More complicated: because of particular rates available
Demand: usually subrate VC-12/VC-3/VC-4
Should be connected to a capable node
Common assumptions
Nodes are DSC capable
Can switch as any rate: VC-12/VC-3/VC-4 for SDH
Telephony: routing of calls again
Example 1: SDH as a transport for telephony
1 PDH E1 (2.048Mbps) into 1 SDH VC-12 (2.304Mbps)
Demands between ingress/egress DSCs in terms of VC-12s
A trunk group of 30 trunks (E1) is a single demand for SDH
1 DVU = 1 VC12
Available link rates
STM-n: 1 – 155 – 622 Mbps,16 – 2.4Gbps, 64 – 9.9Gbps
STM-1: carries 63 VC-12 containers
Capacity of a link measured in STM-1 (SDH)
1 LCU = 63 DVU (1 DVU – 1 E1 converted into 1 VC-12, 30 trunks)
1 STM-1: 63 times 30 = 1890 calls…
Should be a classic problem with modular design!
Telephony: routing of calls again
The whole problem
E
Minimize F   e1e ye
Subject to
Pd
x
dp
p 1
D Pd
 hd , d  1, 2,

d 1 p 1
,D
x  Mye , e  1, 2,
edp dp
,E
where xdp , ye are integers
Note the following
M = 63 and  e is the cost of one STM-1 unit on link e
We optimize using flat cost for STM-1
We do not differentiate between costs of STM-1 within STM-n
A problem can be extended…
Telephony: routing of calls again
Let the cost of one STM-n on link e be  en
The whole problem
Minimize F 
Subject to
Pd
x
dp
p 1
D Pd
 y
N
e1
n1 en en
 hd , d  1, 2,

d 1 p 1
 
E
,D
N
x   M n yen , e  1, 2,
edp dp
n 1
where xdp , yen are integers
,E
SDH: restoration design
Telephony: SDH restoration
What are the consequences of failure of a link?
Loosing all the flows xdp assigned to the path traversing link e
i.e. to all the paths Pdp with coincidence coefficient  edp  1
surviving capacity of demand d in case of failure situation s
Pd
hds  hd    edp xdp , d  1, 2,
,D
p 1
simultaneous partial failure of several circuit groups d
recall, demand volumes are in 30 trunks units
What to do? How to minimize the effect?
The capacities of circuit groups d should be split!
Between as many non-zero flows as feasible
This process is called path diversity
Not commonly used in transport networks such as SDH… Why????
Telephony: SDH restoration
One possible way
Split each trunk group d to at least two flows
How? Add a constraint xdp  hd 2 , d  1,2,
,D
What do we get applying this?
Single link failures: availability is around 0.5 for all demand volumes
Even demand volumes:  ds  0.5
Odd demand volumes: at most slightly less than 0.5
Path diversity in SDH
Can be really costly
Recall restoration example we considered! (245 instead of 85!)
Here bundles are really large…
Restores capacity of individual flows!
Path diversity is also called path protection
Another way: link protection
Restores capacity of a link instead of individual flows
Telephony: SDH restoration
Link protection
Fixed link capacities ce , e  1, 2, , E assigned in normal state
Minimizes the cost of additional link capacities ye , e  1, 2, , E
For each link e there is a list of restoration paths q  1, 2, , Qe
Used when the link fails
Capacity is restored using flows zeq , e  1, 2, , E , q  1, 2, , Qe
 feq  1 if link f belongs to path q restoring link e
The whole problem
E
Minimize F   e ye
e1
Subject to
Qe
z
eq
q 1
E Qe
 ce , e  1, 2,
 
e1 q 1
z  yf ,
feq eq
,E
f  1, 2,
where zeq , ye are integers
, E , e  1, 2,
, E, f  e
SDH: ring bandwidth design
Telephony: ring bandwidth design
SONET/SDH two topologies
Mesh: terminal multiplexers (TM) and digital cross-connects (DCS)
Ring: add-drop multiplexers (ADM) nodes
Why ring?
Restoration mechanism
Mesh: restoration via signaling between DCSs
Ring: restoration is built-in…
Restoration in rings: < 50ms. for a single link failure
ADM nodes
Capable of inserting/extracting containers
For example: we may insert or take out VC-12 at any ADM
Telephony: ring bandwidth design
Bi-directional line-switched ring (BLSR)
Four optical fibers (called BLSR/4)
Two pairs of fibers: one basic and one protection pair
This is link protection
Assumptions we take
Service: telephony
Based on STM-1
63 VC-12 sub-containers
Telephony: ring bandwidth design
Problem
Given the inherent SDH circular routing in ring topology and
the demand volume, how do we determine what is the number
and type of parallel rings needed?
Example
Demand d=1 between ADMs 1 and 2 has volume h1  3 VC-12s
Three numbers out of {1,2,…,63} are reserved between 1 and 2
VC-12s between them are used to realize 90 trunks
If between ADMs 1 and 3 then reservation is on both segments
When one segment fails then inner fiber is used
Whole STM-1 is reswitched
Telephony: ring bandwidth design
Allocation of flows in BLSR/4
Not trivial
Why? Demand is split into two parts, not equal
Part is realized over one of the two complementary parts of the ring
Analyzing the problem
Nodes are numbered from 1 to V clockwise
Segments from 1 to E, but E=V – ring
e=1 connects (1,2), e=2 connects (2,3)
Segment e=V connects V and 1
Demand volume between ADMs v and w hvw
Clockwise v to w: flow uvw
Counterclockwise w to v: flow zvw
Telephony: ring bandwidth design
Problem
Minimize F  r
Subject to demand constraints
uvw  zvw  hvw , v, w  1,2,
,V , v  w
and capacity constraints
 evwuvw  (1   evw ) zvw  Mr , e  1,2, , E
where uvw , zvw , r are non-negative integers
Link-incidence is now segment-demand incidence
 evw
1, v  e  w

0, otherwise
1 if segment e belongs to clockwise path from v to w
we did for M=63 (STM-1), can do for M=252 (STM-4)
If segment is cut: inner fiber is used (protection switching)
Another SDH ring: unidirectional path switched ring (UPRS)
Restoration in UPSR: path protection
WDM: restoration
Telephony: WDM restoration
Wavelength division multiplexing (WDM)
What? Expanding capacity of fibers without installing new ones
What? Bidirectional communication over a fiber
How? Mx-ing number of optical carriers into a single fiber
How? Using different wavelengths (AKA colors or lambdas)
Development of WDM
First published in 1978
Lab demonstration 1980
First WDM: two signals
Now: up to 160, 10Gbps -> 1.6Tbps
Type of fibers
Most WDM: single mode fiber 9mkm
Sometimes multimode fiber 50-62.5mkm
Telephony: WDM restoration
WDM systems
Conventional WDM (CWDM): 8 channels in C band, 1550nm
Dense WDM (DWDM): denser spacing
40 channels at 100GHz spacing or 80 at 50GHz
Ultra-dense WDM (UWDM): 12.5 GHz spacing (320 channels)
Units for dimensioning
DVU: 10Gbps (one wavelength, ~64 STM-1 modules)
LCU: ~100 DVUs
Telephony: WDM restoration
Types of nodes in WDM
Optical cross-connects (OXC)
With/without wavelength conversion
Without conversion: switching in optical domain (colors the same)
With conversion: can switch colors (wavelengths as well)
Types of networks depends on OXS
Network without conversion: color of the light-path is the same
Network with conversion: color of the light-path may change
Telephony: WDM restoration
Problem
Wavelength assignment problem: assign wavelength in
optimal way assuming WDM network with path diversity,
routing, and without wavelength conversion.
Specifiyng the problem
Demands to be realized hd , d  1, 2, , D
Candidate path for demand d: p  1,2, , Pd
Links e  1, 2, , E (composed of sets of fibers)
Colors (wavelength) available on links are labeled c  1, 2,
We include failures
Failure situations s  0,1, , E
State s  0 implies no links are broken
At most one link at a time
,C
Telephony: WDM restoration
Analyzing the problem
Volume of demand d realized in failure situation s
hds , d  1,2,
, D, s  0,1,
,E
number of light paths
Cost of one LCU (optical fiber) on link e: e , e  1,2, , E
Link failure coefficients es {0,1}
0: if link e is failed in situation s
Link-path incidence:  edp {0,1}
1: if link e belongs to path p realizing demand d
Path failure coefficients  dps {0,1}
0: if path p of demand d failed in situation s
We have
 dps 


{e:
edp 1}
aes
Telephony: WDM restoration
Number of light paths realizing d in color c of path p: xdpc
Number of times color c is used on link e: zce
Capacity of link e: ye , e  1, 2, , E
The problem (OXCs without conversion)
E
Minimize: F   e1e ye
Subject to:
Pd
C
  x
p 1
dps
c 1
dpc
 hds , d  1, 2,
, D, s  0,1,
,E
in each state s surviving flows realize DVUs for all demands d
D
Pd

d 1 p 1
x
edp dpc
 zce , c  1, 2,
, C , e  1, 2,
,E
number of times color c is used on e
ye  zce , c  1,2,
, C , e  1,2,
,E
fibers sufficient to realize demand for each color on all links
where xdpc , zce , ye are non-negative integers
Telephony: WDM restoration
Notes about the problem
Integer programming: a lot of integer variables
Complex to solve for large networks
LP relaxation
Make all variables continuous
Allows to get approximate solutions
What if OXCs with wavelength conversions?
Becomes way simpler
We get a “normal” switching network
Identical to what we considered for SDH previously…
Can be extended to include link opening costs
IP over SONET (PoS)
Telephony: WDM restoration
Recall the following
All networks are multi-layer ones
Basics: traffic network over transport network
IP over SDH/SONET is one frequently used
SDH/SONET: no built-in support
PoS: packet over SDH/SONET
We already considered
IP with OSPF protocol
SDH/SONET with DSCs
Now let’s take a look at
IP links over SONET links
Two-layered architecture
Implemented using PoS
Telephony: WDM restoration
Problem statement
Given an IP intra-network domain and the fact that IP links are
realized as transmission paths over a capacitated SONET, how
do we determine capacity required for the IP links and the
routing of these links in SONET to minimize IP network cost.
Want to work with SDH? Just replace capacities…
When SONET and IP are owned by the same entity
When owned by different entities (we one needs from SONET)
Specifying the IP layer
SONET network is given (nodes/capacities, etc.)
We are given IP traffic demands d=1,2,…,D
We are using OC-3 line cards (155.52Mbps)
IP links are then modular multipliers of 155.52Mbps
If demand < 155.52 then IP link module value is M=155.52
Telephony: WDM restoration
Specifying the SONET layer
IP links capacity is demand for SONET
1 DVU at SONET layer: 1 OC-3 (155.52Mbps)
OC-3 demands are routed using OC-48 (2.4G) or OC-192 (9.9G)
1 LCU at SONET layer: N = 16M (16 OC-3 in OC-48)
Path in SONET: several DCSs
End IP routers are mapped to some DSCs
Example of demands
1 Mbps between two nodes in IP
Becomes 1 LCU in IP that is equal to M=155.52Mbps (OC-3)
1 LCU in IP becomes 1 DVU in SONET (OC-3)
1 LCU in SONET OC-48: N=2.4Gbps
1 Mbps -> 155.52Mbps -> 2.4Gbps 
Telephony: WDM restoration
SONET
IP
Formalization of the problem
IP network with V nodes and E links
IP network traffic demands: hd , d  1, 2, , D
Allowable paths for demand d: p  1,2, , Pd
Link weights system: we , e  1,2, , E
Flows induced by the weight system: xdp ( w)
IP link incidence coefficients:  edp  1 if path p of demand d uses e
let modular capacity of link e in IP be ye (unknown, in modules of M)
link capacities ye are demands to SONET network
links in SONET: g  1, 2, , G
Link capacities in SONET: q  1, 2, , Qe
candidate path in SONET for IP link (demand) e:
SONET link-incidence :  geq  1 if path q of demand e uses link g
Notes on routing:
Routing IP: dynamic, Routing SONET: semi-permanent
Telephony: WDM restoration
Variables for optimization
Link weights: w  ( w1 , w2 , , wE )
IP link modular capacities: y  ( y1 , y2 ,
SONET modular capacities: zeq , e  1, 2,
Problem
Minimize F 
Subject to:

Pd

E
 ye   e1  q 1eq zeq
E
Qe
e 1 e
x ( w)  hd , d  1, 2,
p 1 dp
 
, yE )
, E , q  1, 2,
,D
IP link termination cost
(cost of OC-3 cards at the
end of link e, often linkindependent,e  2 )
 x ( w)   Mye , e  1, 2, , E
D
Pd
d 1
p 1 edp dp
  z
M 
Qe
C
q 1
c 1 eq
E
Qe
e 1
q 1
 ye , e  1, 2,
,E
 geq zeq  Ncg , g  1, 2, , G
we , ye , zeq non-negative integers
, Qe
Cost of routing in
SONET
Utilization to
minimize congestion
Telephony: WDM restoration
Cost of routing in SONET eq ?
Setting eq  1
Maximizing the spare capacity of SONET links
Setting eq  e , q  1,2, , Qe
Cost rate (per month, per year) of one LCU of IP link e
To be paid by IP provider to SONET operator
The problem is a huge headache…
LP relaxation?...