Detail Study of IP/ Reconfiguable Optical Network Architectures

maidtweetNetworking and Communications

Oct 29, 2013 (3 years and 10 months ago)

103 views

Abstract

IP over reconfigurable optical network ar
chtectures have been extensively discussed within the research
literature over the


This word document was downloaded from
http://www.worddocx.com/

please remain this
link
information when

you

reproduce , cop
y
, or

use

it.

<a href='http://www.wordwendang.com/en'>word documents</a>


past few years. However, although reconfiguable optical networks have been deployed and signaling protocols between IP router
s and
optical networks have been standardized, large IP backbo
ne over reconfiguable optical network deployments have not yet been widely
accepted. One of the most important criteria in determining whether an IP backbone should be carried over a reconfigurable op
tical
network is economic viability
-

which necessitates

a detailed, accurate economic study of IP backbone over reconfiguable optical
network architectures. In this paper, we analyze and explore four IP over optical network architectures for a typical large I
SP
backbone, which have been proposed before. In con
trast with other published claims, our results suggest that an IP over opaque re
-
configurable optical network architecture is not economically attractive with current equipment and IP backbone network desig
n
requirements. However, for ISPs also carrying la
rge volumes of transport network pri vate line services, our proposed integrated IP
over re
-
configurable optical network architecture may provide an attracti ve alternative for providing rapid, cost effective failure
recovery.

A.

I
NTRODUCTION

Large IP backbone
s are today typically deployed directly over point
-
to
-
point DWDM systems. However, with the
growing deployment of Opaque Reconfigurable Optical Networks (RON), there is interest in using a RON to provide
the connectivity (links) between IP routers. Here, w
e define an
opaque

reconfiguable optical network as a network of
optical cross
-
connects (OXCs)


with either optical or electrical fabrics surrounded by optical transponders that
provide opto
-
electronic regeneration and wavelength conversion. A RON offers
rapid provisioning that can be used
to respond to changes and growth in IP traffic demands along with fast, cost efficient, restoration of IP link failures.
As a result, there has been a great deal of research in many facets of IP over RON design, includin
g experimental
investigation, routing algorithms, restoration schemes, network management and control protocols, and signaling
protocols that enable dynamic connection establishment between IP routers over the RON. However, large IP
backbone over reconfigu
able optical network deployments have not yet been widely accepted. One of the most
important criteria in determining whether an IP backbone should be carried over a reconfigurable optical network is
economic viability
-

which necessitates a detailed, accu
rate economic study of IP backbone over reconfiguable
optical network architectures.

Large ISPs typically use a hierarchical IP network design:
single or dual large backbone routers (BRs), located in major
Points
-
of
-
Presence (PoPs), aggregate traffic fro
m access




Guangzhi Li, Dongmei Wang, Jennifer Yates, Robert Doverspike, Charles Kalmanek

AT&T Labs Research, 180 Park Avenue, Florham Park, NJ 07932

Detail Study of IP/ Reconfiguable Optical Network Architectures


Figure
1
: Dual
-
Router PoP Architectur
e

routers before forwarding it to other cities. Access routers (ARs) are either co
-
located with the backbone routers,
supporting traffic originating and terminating in that city, or are remotely located, aggregating traffic from smaller
cities. Traf
fic from remote access routers may be connected over protected facilities, but are often dual
-
homed over
diversely routed unprotected SONET/SDH
-
over
-
DWDM circuits. The use of dual routers in major PoPs plus
diversely routed dual
-
homed AR
-
BR links enable IP

routers to restore traffic under a variety of failures including
fiber cuts, router line card failures, and router failures. This architecture is known as the
Dual
-
Router Architecture

(see Figure 1). If each AR is instead connected to only a single BR, as

in the
Single Router Architecture
, then once
the BR fails, all traffic terminating in that PoP will be isolated from the network. Thus, such an architecture requires
an ultra
-
reliable router. In this paper, we only consider the dual
-
router architecture si
nce router reliability is still a
serious concern for most larger service providers.

Preliminary studies into the economic viability of IP over an optical network have already been presented within
the literature [1,2,3,10]. In [1], the authors examined bo
th single and dual router architectures and demonstrated that
an IP over RON design is only cost effective in the single router architecture where the backbone is not designed to
handle BR failures. In contrast, [2] and [3] claim that IP over RON architect
ures are cost effective even for dual
routers. However, they make this claim by comparing different IP virtual topologies for the IP over RON and the IP
over DWDM architectures. When routing IP links directly over DWDM, the authors assume that all of the I
P links
passing through an office terminate on the router, whereas in the IP over RON architecture, they use optical cross
-
connects in offices to allow IP links (referred as “express” links) to bypass the router to reduce the number of router
line cards re
quired. Such a comparison may not be justified since the “express” links enabled by the optical cross
-
connect can (and are today) also deployed over point
-
to
-
point DWDM directly. Secondly, the authors in both papers
assume that the RON topology is identica
l to the IP topology. In reality, the RON topology and the IP backbone
topology could be totally different and the RON network is usually larger (i.e., more nodes and links) than the IP
backbone network. Thus the access links from access router to backbone

router could be built on the RON network
for reliability. Furthermore, [2] and [3] both ignored the costs of some optical layer devices in their economic
analysis, including the DWDMs, optical transponders, and optical amplifiers. In [10], we proposed two

integrated

IP/optical architectures, that use spare router line cards connected to a RON and tight integration between the two
layers to provide cost effective failure recovery. We denoted these as the
Hybrid

and
Modified Protected
architectures, and comp
ared them with
the
Baseline

architecture (IP directly over point
-
to
-
point DWDM systems)
and the
Protected

architecture (each IP link is routed over the RON, with optical network restoration)

and our
preliminary results show that the integrated IP/optical a
rchitectures are attractive alternatives to the Baseline
architecture. To further explore the integrated IP/optical architectures, we believe that a more comprehensive analysis
and modeling are necessary.

This paper extends the economic analyses in [1] an
d builds the results from architectures proposed in [10]. In this
paper, four architectures (i.e.,
Baseline, Protected, Hybrid
, and
Modified Protected
) are analyzed and explored in
much more detail than in [10]. We evaluate the capital cost to implement ea
ch of these architectures by dimensioning
both the IP and optical networks for given traffic matrices and then costing the required components and simulate the
traffic impact on capital cost of each architecture. The networks are dimensioned with sufficien
t capacity to ensure
recovery from any single failure event, including any fiberspan and router failure.

The paper is organized as follows: Section B describes the four IP over optical network architectures in detail.
Section C discusses network dimensioni
ng procedures and modeling assumptions. Section D presents our cost
analysis models, whilst section E is our numerical results and analysis evaluating the cost effectiveness of each
architecture.

B.

IP

OVER
O
PTICAL
N
ETWORK
A
RCHITECTURES

The Structure of IP a
nd Optical Layers

Figure 2 shows a typical structure relationship between the IP
and optical layers, where the optical layer consists of the fiber
media layer, DWDM transmission layer, and the optical
switching layer. The DWDM layer could consist of poin
t
-
to
-
point systems or ultra
-
long haul systems with optical add
-
drop
multiplexers. Then, under such a layer decomposition, in general
there are three alternatives to design an IP network over an
optical layer: (1) route IP links over hard
-
wired point
-
to
-
poi
nt
DWDM systems and rely on the IP layer to provide all
restoration; (2) route IP links over the optical switch layer and
utilize both layers for restoration; and (3) route IP links over Ultra
-
Long Haul (ULH) DWDM systems with
restoration capabilities, usi
ng restoration in both the IP and transmission layers. The first and third alternatives utilize
only three layers, while the second alternative requires a total of four layers. In this paper we only focus on the first
two alternatives, more specifically, t
he integrated forms of the first two architectures, and leave the third alternative
for future study. We now describe the four IP over optical network architectures we studied in more detail.

1)

Baseline Architecture

Our
Baseline

Architecture corresponds to t
he first alternative above, i.e., each IP link is permanently routed (“hard
-
wired”) over one or more point
-
to
-
point DWDM systems. Thus, the baseline architecture does not use any form of
optical/electrical cross
-
connect platform and restoration is not prov
ided at the optical layer.

In such an archtuecture, the IP layer is responsible for all failure recovery. We assume here that the IP layer uses
routing protocols ,such as OSPF or IS
-
IS, that “reconverge” after failure to reroute traffic via alternative pat
hs. IP
routing protocols have traditionally been slow to recover from failures


taking from tens seconds to minutes to
recover from a single failure, which may be unacceptable for delay
-
sensitive applications such as voice over IP or
real
-
time Internet ap
plications. However, there is significant ongoing work focusing on reducing routing protocol
convergence times to enable fast failure recovery [4]. Additionally, mechanisms such as MPLS fast re
-
route can also
be introduced to provide fast failure recovery


although this introduces additional complexity [5].

Although architecturally simple, the baseline architecture introduces significant capacity planning and network
maintenance complexities as well as coordination between IP layer and transport layer. As

the optical layer links are
unprotected, the IP planning process must dimension that sufficient capacity in the IP layer to recover from both IP
Figure
2
: IP over Optical Layer

and transport layer failures. In particular, IP network planners must be aware of the physical layer routing o
f each
unprotected IP link, so that links routed over common fiberspans can be identified and sufficient capacity can be
deployed to recover from any fiberspan failure. This complicates the planning process. In addition, as transport layer
planned maintena
nce events impact the IP traffic routing, planned maintenance must be coordinated between IP and
transport to ensure that sufficient capacity is available to carry traffic during such events (i.e., that maintenance
should not be simultaneously occuring on
both fiberspans and routers, otherwise that can result in severe network
performance degradations) and that work on fiber facilities does not result in traffic continually being re
-
routed as
maintenance workers continually pull fibres in and out (i.e., avo
id link flapping in the routing protocols). This
requires carefully designed procedures and coordination between different work centres within a large ISP.

2)

Protected Architecture

Our
Protected

architecture corresponds to the second alternative discussed a
bove, i.e., a RON lies between the IP
layer links and the DWDM layer. The RON provides restoration from optical
-
layer failures (e.g., fiberspan cuts, and
optical component failures) for the links connecting the backbone routers. Generally, the RON can rest
ore faster than
IP routing protocols, thereby reducing the failure recovery times for link failures. However, the RON cannot recover
from IP layer failures, such as router failures
-

these must instead be restored within the IP layer. Thus, in addition to

the spare capacity deployed in the RON to recover from IP link failures, further spare capacity must also be deployed
in the IP layer to recover from router
-
related failures. Thus, this architecture duplicates spare capacity in both the IP
and optical lay
ers.

In addition to reducing failure recovery times, using protected IP links also reduces the operational complexities
associated with planned maintenance activities and planning restoration network capacity discussed above. The
protected architecture sep
arates the planning and maintenance activities between the IP and optical layers, thereby
eliminating the need for careful coordination between the two layers. This simplifies procedures, for example,
between the IP and transport work centres for maintenan
ce activities, eliminating the need for additional processes
and tools to manage the coordination.

3)

Hybrid Architecture

Rather than carry all IP links connecting BRs over the
RON, an intermediate approach referred to here as the
Hybrid architecture,

uses

the RON only to provide extra
link capacity during restoration. As in the Baseline
architecture, IP layer links are carried directly over
unprotected point
-
to
-
point DWDM systems. But, in contrast
with the Baseline architecture, the IP layer is only
dimens
ioned with sufficient capacity to carry the service
traffic: no spare capacity is provided for failure recovery in
IP layer. However, as proposed in [6], spare router line
cards are installed in each BR and connected to a co
-
located
OXC in the RON. During
normal operation, these router
Figure 3: Hybrid Architecture

spare line cards are idle and no connections are established between IP routers over the RON. Once a failure occurs,
the spare router to OXC links are used to establish new connections via the RON to increase the IP network c
apacity
(“bandwidth on demand”). This approach shifts IP layer restoration to the optical layer and provides an opportunity
to more broadly share existing RON restoration capacity between IP layer services and other services, such as Private
Line services.

In particular, router failures and fiberspan failures are generally non
-
simultaneous events and, as such,
restoration capacity can be shared, which means that the restoration capacity reserved for fiberspan failures can also
be used for IP router failure.


Figure 3 illustrates the hybrid architecture. There are two network clouds: the upper one is a reconfigurable optical
network with shared mesh restoration whilst the lower one is the pure point
-
to
-
point DWDM systems. IP links for
service bandwidth are ha
rdwired through the DWDM systems without restoration. However, each router will be
installed with one or more spare router line cards which connect to the OXCs. We assume that the optical network
also provides transport services to customers with shared me
sh restoration [7]. Most likely, the DWDM layer and the
RON will share the same physical fiberspan topology. Then a single fiberspan failure would impact services in both
the point
-
to
-
point DWDM systems and the RON. Thus, additional capacity is required in

the RON to handle
restoration for IP layer link failures. However, a single backbone router or router line card failure will only impact
the IP network and thus the restoration resources used in the RON for link failures (both IP and other services) could

also be used to restore IP traffic from such a failure. Thus, there must be sufficient additional capacity in the optical
network to handle IP network fiberspan failures, but IP layer failures can effectively be recovered for almost “free”.
In this archit
ecture, we add more spare router line cards, which increases the total cost. But we eliminate IP layer
restoration and reduce the required router line cards and IP link capcacity, which decreases the total cost.

Recovery from both IP and optical layer

failures in the hybrid architecture requires IP routing protocol
reconvergence, followed by the establishment of a new optical layer connection to provide the necessary restoration
capacity. New IP links could be established between routers rapidly if fas
t signaling is used in the RON and link
layer protocols are tuned appropriately [8]. However, as this scheme requires IP routing convergence, it is clearly
going to be slower than the recovery times achieved in the baseline architecture. Thus, failure reco
very times could
be a significant issue for this architecture. Note that this architecture also inherits the complexities of the baseline
architecture in terms of managing planning and maintenance across layers.

4)

Modified Protected Architecture

In light of
the relatively slow recovery times for the hybrid architecture, we propose the
Modified Protected

architecture, which is similar to the hybrid architcture, but routes the IP service links over the RON instead of
directly over the unprotected DWDM layer. Th
e IP layer is dimensioned with sufficient capacity to handle only the
service traffic, and each of these links is carried over the RON. Optical layer restoration is then used to handle optical
failures such as fiberspan cuts, thereby reducing the recovery
times compared with the hybrid and baseline
architectures. However, like the hybrid architecture, IP layer failures are recovered using spare router line cards
connected to the optical network and dynamic connection establishment over the RON. This approac
h again shifts IP
layer restoration capacity to the optical layer. In the event of a router failure, the IP layer routing protocols re
-
route
the traffic, and then the router dynamically requests new connections using the spare router line cards and the RON

to establish new IP links between the spare router line cards to provide the restoration capacity. Although this
approach is still relatively slow, as routers become increasingly reliable, as is the trend today, router failures are
expected to become less

common and their restoration time becomes less critical. Router line card failures can be
recovered using rapid 1:
n

local router line card failure recovery, in which a single spare router line card is used to
recover from the failure on any one of
n

worki
ng router line cards [8].

C.

N
ETWORK
D
IMENSIONING AND
M
ODELING ASSUMPTIONS

To evaluate the capital required to implement each architecture, we need to dimension each network and then cost
the required components. We do this in each case by starting with a co
mmon, pre
-
defined, fiber topology with
deployed DWDM systems, over which our IP and RONs are carried. We assume that the network topologies for both
IP and RON, the traffic demands for both IP and private line services, and how each IP and RON link is rout
ed over
the fiberspan topology are all known.
This latter information is used to design and dimension the network under
particular fiberspan failure scenarios.

To determine the capacity required in both the IP layer and the RON, we start by routing the tr
affic over the
network to dimension the service requirements. We assume that IP routing is based on Open Shortest Path First
(OSPF) and that connections in the optical network are routed based on a minimum number of hops. We then
specify
the set of failure

scenarios to be considered and enumerate each failure scenario in turn, routing the traffic in
response to the failure. The capacity dimensioned (i.e., link channels) between each pair of adjacent routers and on
each RON link is the maximum required acros
s all single failure scenarios.

Although multiple simultaneous failures
can occur in networks, they are typically very rare. We thus assume here, as is common practice in network design,
that we need to dimension sufficient capacity to survive only a singl
e failure: either a single router or fibrespan
failure.
To simplify our analysis, we only focus on the backbone network


i.e., backbone routers and backbone links.
ARs and remote RARs are aggregated into a single virtual AR so that each PoP is modelled us
ing three routers: two
BRs and one AR. The single AR is dual
-
homed to the two BRs and the two BRs are connected to other BRs via
backbone links.

We now consider specific details on network dimensioning for each of our different architectures.

1)

Baseline Arch
itecture

To dimension the baseline network, we fix the IP topology (router locations, allowable BR
-
to
-
BR links, and the
routing of those links over the point
-
to
-
point DWDM systems), OSPF weights and the AR
-
to
-
AR traffic matrix.
Following standard IP networ
k design procedures, each IP link has two maximum traffic utilization constraints: one
for the network in a non
-
failure state and the other during failure. The maximum acceptable link utilization during
failure is usually the same or slightly higher than t
he maximum acceptable utilization during non
-
failure conditions.
A higher maximum acceptable link utilization during failure is often viewed as an acceptable compromise between
the additional cost of the spare restoration capacity and performance during th
e relatively rare failures. We then route
the traffic and compute, under the utilization constraints, the resulting capacity required between each pair of
adjacent routers to meet the utilisation constraints. We repeat this for each failure scenario, inclu
ding all router, and
fiberspan failures, dimensioning each link to the maximum capacity required over all failure events. After that, we
determine the network cost by adding the contributions from router line cards and optical transport systems
(composed o
f optical transponders, DWDM terminals, and optical amplifiers). We assume that the cost of router
chassis is amortized into the cost of router line cards.

2)

Protected Architecture

To model the Protected architecture, we use the same IP topology, OSPF weight
s and traffic matrix as were used to
dimension the Baseline architecture. We then dimension sufficient capacity at the IP layer capacity to ensure
recovery from any single router failure. These IP links are routed over the RON, we thus also calculate the R
ON
restoration requirements to restore from any fiberspan failure. Note that each IP link may carry multiple IP OC48 or
OC192 channels.

We assume that the optical network uses shared mesh restoration. In shared mesh restoration, each optical
connection has

two paths: a primary and a physically disjoint restoration path [6]. The restoration path is pre
-
calculated to speed up the restoration time. During normal optical network operation, the IP link is established along
the service path only, with resources r
eserved along the restoration path. The resouces reserved on each optical link
along a restoration path are shared across multiple restoration paths whose service paths are not expected to fail
simultaneously, i.e., the service paths that do not share a co
mmon fiberspan. This is feasible since the restoration path
is only established when the service path fails. The resouces reserved on the restoration paths are dimensioned so that
there is sufficient capacity to recover all affected IP links and optical ne
twork services in the event of any single
fiberspan failure.

3)

Hybrid Architecture

To dimension the network for the
hybrid architecture
, we first dimension the IP layer and the corresponding
DWDM layer resources required to support the IP service traffic. We

then determine the additional IP spare router
line cards and optical network capacity required to recover from any single router or fiberspan failure. We achieve
this by again enumerating each failure scenario, and determine the additional bandwidth requi
rements as a result of
each failure. Since the spare router line cards can be shared across independent failure events, the number of spare
router line cards required at each router is computed as the maximum over all failure scenarios of the total extra l
ink
channels required to adjacent routers. The additional RON restoration capacity on each link is the maximum over all
router failures. We calculate this by determining the additional capacity (i.e., channels) required between routers to
recover from the
failure, whilst for fiberspan failures, the additional capacity is that required to recover both the IP
layer connectivity and the private line services impacted in the RON. The capacity efficiency of this architecture
derives from the flexibility of a sin
gle spare router line cards to cover multiple, non
-
simultaneous failures and the
ability to share RON restoration capacity between IP and other services.

4)

Modified Protected Architecture

To dimension the network for this

architecture
, we first dimension the

IP layer resources to support the IP service
traffic. This is achieved via routing IP traffic on IP topology using shortest path algorithm. We then determine the
additional IP spare router line cards and optical network capacity required to recover from a
ll set of failures. We
achieve this by again enumerating each failure scenario, and determining the additional bandwidth requirements in
both the IP layer and the RON as a result of each failure. Since fiberspan failures are restored at the optical layer,
no
spare router line card is required for fiberspan failure scenarios. The spare router line cards are used for router and
router line card failures. In the event of a router failure, the router line cards at the remote ends of IP links connected
to the fa
iled router can be released and used for establishing new IP links to provide the required restoration capacity.
The number of spare router line cards required at a router upon another router failure is calculated as the total extra IP
link capacities requ
ired to adjacent routers minus the total released router interface cards due to the failure. The
minimum number of spare router line cards at each router is then computed as the maximum of the number of spare
router line cards required upon another router
failure over all router failures. We assume that at least one spare router
line card is available on each router to recover from any single router line card failures protection (1:
n

interface
protection, where
n

is large!). The additional RON capacity on e
ach optical link will come from both IP link capacity
and IP layer restoration capacity. The additional RON capacity for IP layer restoration is the maximum over each
router failure scenario. We calculate this by determining the additional capacity require
d between routers to recover
from the failure based on existing restoration capacity in RON. For fiberspan failures, the additional capacity is that
required to recover both the IP layer connectivity and the private line services impacted in the RON. The c
apacity
efficiency of this architecture again derives from the flexibility of a single spare router line cards to cover multiple,
non
-
simultaneous failures and the ability to share RON restoration capacity between IP and other services.

D.

C
OST ANALYSIS MODEL
S

We describe our cost analysis models on how to calculate the costs of above four IP over optical network
archiectures in this section. The inputs to our model include the optical and IP network topologies and corresponding
fiberspan information, the uni
-
directional IP network traffic matrix, the bi
-
directional optical network traffic matrix
for private line services and the equipment parameters and costs. The IP nodes (PoPs) are a subset of the optical
network nodes, and the DWDM systems topology is the s
ame as that of the RON.

1)

Notation

Let

represent the IP network, where
N

is the set of IP nodes,
L

is the set of IP links, and
F

is the set of
fiberspans. A function
f: L


F

represents the relationship between
L

and
F
. At each IP nod
e, there are four network
elements: an access router (AR), two backbone routers (BR), and an optical cross
-
connect (OXC). Let

be
the underlying optical network, where
O

N

(i.e., the IP router locations are a subset of the OXC locati
ons),
E

is the
set of optical links riding on DWDM systems, and
g:E


F

represents the relationship between
E

and
F
.

The IP uni
-
directional traffic matrix denotes the traffic between access routers. As the IP links are bi
-
directional,
we calculate the capa
city on each IP link as the maximum required to satisfy the uni
-
directional traffic demands in
both directions on the given link. On each IP link
i
in direction
k
(
k=1

for direction of small router id to big router id
while
k=2

for direction of big router
id to small router id
)
, we denote
S
k
[
i
]

to be the minimal service capacity and
R
k
[
i
]

as the minimal restoraton capacity. The total required capacity on link
i, T
[
i
]
,

is then
max
k
(
S
k
[
i
]
+R
k
[
i
])

whilst the
total required service capacity on link
i,

S
[
i
]
,

is
m
ax
(
S
1
[
i
],
S
2
[
i
])
. We also denote
failneed
k
[
i,j
]

to be the restoration
capacity required on link
i

in direction
k

if failure
j

occurs, where failure
j

represents either a fiberspan or backbone
router failure. Similarly,
failrelease
k
[
i,j
]

denotes the capacit
y released on link
i

in direction
k

after the service traffic
is rerouted upon failure
j
, and
failreroute
k
[
i,j
]

is the traffic rerouted onto link
i

in direction
k

when failure
j

occurs.
Then
failneed
k
[
i,j
]
=max
(
0,failreroute
k
[
i,j
]
-
failrelease
k
[
i,j
])

and
R
k
[
i
]

= max
j

failneed
k
[
i,j
]

over the different failure
scenarios.

Furthermore, we define
P
[
i,j
]

to be the number of restoration channels required on link
i

upon failure
j
.Then
P[i,j]
=

(

max
k
(
failneed
k
[
i,j
]
+ S
k
[
i
]
))/U/

2





(
max
k
S
k
[
i
]
)/U/

1


)
, where

k=1,2
for the two directions per link
, U
is the
router line card capacity
,
and

1,

2

are the allowable link utilization under normal and failure conditions respectively.
Let
Q
[
r,j
]

be the minimum router restoration line cards required on router
r

upon failure
j
,

then
Q
[
r,j
]

=

i
P
[
i,j
]

where
i

covers all backbone links adjacent to backbone router
r
. Then
H
[
r
]

= max
j

Q
[
r,j
]

over all failures is the
minimal number of spare router line cards required on backbone router
r.
To simplify the modelling, we assume here
tha
t the backbone router uses homogeneous granularity line cards, such as OC48 or OC192.
1

For the
modified
protected architecture
, if router
r

releases
q
[
r,j
]

router line cards upon failure
j
,
Q
[
r,j
]

must be revised as
Q
[
r,j
]
=
max
(
1,

i
P
[
i,j
]
-
q
[
r,j
])

to inc
lude the service router line cards released on router
r

in response to the failure of router
j

and at least one spare router line card for the router line card failure
1:n

protection.

An optical transport system (OTS) is a pair of WDM
devices connected w
ith two fibers and multiple intermediate
optical amplifers (OAs). Between two neighboring OTSs,
one wavelength requires one optical regenerator (OR). An
optical transponder is also required at each wavelength
between each WDM device and OXC. Figure 4 illus
trates an
example link interconnecting two BRs, depicting the BRs,
OXCs, WDMs, OAs, ORs and OTs. We denote the maximal
OTS length as
D,

the maximal distance between two OAs
along an OTS as
d
, and the number of wavelengths per OTS
as


.

To estimate the cos
t of the different network archtectures, we further define the OA cost to be
C
1
, the WDM
cost to be

C
2
, the optical regenerator cost to be

C
3
, the router line card cost to be
C
4
, the optical transponder cost to
be
C
5
, and the OXC port cost to be
C
6
. Simil
ar to the router chasis cost assumption, the cost of OXC chasis is also
amortized into the OXC port cost.

2)

Cost Analysis

The capital cost of an IP over optical network can be divided into two components: the cost of the IP layer and the
cost of the optica
l layer. The IP layer cost is calculated as the sum of the costs of the IP router line cards deployed
within the network whilst the cost of the optical layer includes the cost of the OXC ports, WDMs, optical amplifiers,
optical regenerators, and the optic
al transponders on each optical link in the network. Physical fibers are assumed to
be pre
-
deployed and are thus not included in this cost analysis.

We start by calculating the capital costs of the IP layer for the four architectures. The total number of s
ervice
channels required on IP link
i

will be:

S
[
i
]
/U/

1


while the total number of channels is

T
[
i
]
/U/

2


. For both the
baseline and protected architectures, the total IP layer cost would then be
2*C4*

i


T
[
i
]
/U/

2

,

where the summation
Figure 4: Example of equipment model

(
i
)

is over al
l IP links. The value
T
[
i
]

will be smaller for the protected architecture than the baseline, as the protected
architecture dimensions only sufficient capacity in the IP layer for router failures, whilst the baseline architecture has
to ensure sufficient c
apacity for both router and fiberspan failures. For the hybrid and modified protected
architectures, we only need to consider capacity for the router line cards required to carry the service traffic and the
additional router line cards required in the IP l
ayer for restoration (the ones connected to the OXCs). The total IP
layer cost would then be calculated as:
2*C4*

i

S
[
i
]
/U/

2


+C4*

r
H
[
r
]
,
where the first term is summed over all IP
links,
i
,

and the second term is summed over all backbone routers,
r.

To calculate the capital cost of the optical layer, we need
to consider the IP service demands and, in the protected,
hybrid, and modified protected architectures, we also need to
consider the other optical network service demands (e.g.,
private line deman
ds) so that we can account for the sharing
of restoration capacity across the multiple services.

In the
baseline architecture, there is no optical layer restoration
capacity and OXC port costs to consider. However, for each
IP link channel, which is the ch
annel between two router line
cards in an IP link, we need two optical transponders (an
optical transponder is requied per channel between router
line card and WDM terminal), a list of optical transport
systems along with an optical regenerator between eac
h pair of adjacent optical transport systems. Given IP links
routing on the DWDM system, we can calculate each IP link channel cost in the optical layer, denoted as
Y
[
i
]
. Thus
the optical layer cost for baseline architecture is:

i


T
[
i
]
/U/

2

*
Y
[
i
], the su
mmation of (
i
) is over all IP links
.

For
protected architecture, the capital cost of optical layer requires detail analysis. Figure 5 depicts a procedure flow
-
chart depicting how to calculate the additional optical network capacity for IP restoration in th
e RON. The RON is
initially dimensioned to support the non
-
IP service demands (e.g., private line) by determining the capacity required
to support both service paths and restoration. The RON is then dimensioned including the IP requirements. This
increases

the RON capacity requirements
-

the difference is the extra capacity required for the IP over RON design.
Assume that
T
1[
j
]

and

T
2[
j
]

represent the total number of optical channels required on optical link
j
before and after
applying the IP over RON desig
n. We can then calculate the cost of the RON associated with protected architectures
as:

j

(T2[j]
-
T1[j])*C[j] + 2*C6*

i


T[i]/U/

2

.
Here
C[j]

is the channel cost along optical link
j
, which is
calculated as
((

l[j]/d

-

l[j]/D

)*C1+

l[j]/D

*C2)/


+ (

l[j]
/D

-
1)*C3+2*C5+2*C6
,
where
l[j]

is the length of
optical link
j
. The other contribution of optical cost comes from the OXC ports for IP link because each IP link needs
two extra OXC ports in our considering RON network at the edge of RON, where
T[i]

is the

total IP traffic capacity
required on IP link
i
. Next, for the hybrid architecture, the optical layer cost consists of two portions: the cost of the
DWDM system and the cost of the RON elements. The cost in DWDM system is modelled as
I=

i


S
[
i
]
/U/

1

*
Y
[
i
]







1

In practice, IP links may not always have the same granul
arity.

Figure 5: IP/RON capacity flow
-
chart

and the cost of the RON elements is modelled as

j
(T2[j]
-
T1[j])*C[j]+C6*

r
H[r],
where
i

sums over all IP links,
j

sums over all RON links, and
r

sums over all backbone routers
.

Then, the optical layer cost for hybrid architecture is:
I+II
.

Similary the o
ptical layer cost for modified protected architecture is the same as hybrid

j

(T2[j]
-
T1[j])*C[j] +
C6*

r
H[r],
except that the values of
T1, T2, H

will be different.

E.

R
ESULTS AND ANALYSIS

1)

Network Topologies

We use realistic ISP backbone, optical network to
pologies, and traffic forecasts to calculate the capital costs of
the four architectures described above. The IP network consists of 18 backbone locations, where each location is
modelled as two BRs and a single AR as we described above, 32 backbone links,

and about 1000 fiberspans. The
AR
-
to
-
AR traffic matrix represents the aggregate demand between the backbone locations. In addition, we have a
traffic matrix that represents the private line forecast. The underlying RON consists of about 100 OXCs, 200 opti
cal
links, and 1500 fiberspans. Each link of the IP and RON is routed over multiple fiberspans and a single fiberspan
may support multiple links. Thus, a fiberspan failure may cause multiple link failures. We use component prices and
traffic matrices based

on established, commercial networks (note that we assume that a router line card costs three
times that of a cross
-
connect port). The networks were all designed to survive a single router or fiberspan failure.
Note that the network topologies considered h
ere assume significantly more OXC locations than BR locations, which
is contrast with the analyses in [2,3]. This increases the number of OXC ports along each IP link carried over the
RON, but also reduces the restoration capacity required in the RON.

2)

Cost

Analysis

Using the formulas discussed above and the input data,
we calcuate the total cost of the four architectures and
separate the contributions of the different equipment
categories, e.g., router line cards, OXC ports, WDM
terminals, optical amplif
iers, optical regenerators, and
optical transponders. Figure 6 depicts the numerical results
normalized relative to the total cost of the Baseline
Architecture (the equipment price and other parameter
values used in our model are all for OC48 channels onl
y). In
contrast with the conclusions stated in [2,3], it is clear that
the capital cost associated with the Protected Architecture is
significantly higher than the Baseline architecture (in the
case studied here, it is more than 60% higher). The reason is
that RON only protects optical layer failure and IP layer
failure such as router failure has to be protected in IP layer. The protection capacities in both layers can not be shared
which could lead to some restoration capacity duplication in IP and optical

layers.

The Hybrid architecture demonstrates some small savings relative to the baseline architecture, coming from the
reduction in the number of router line cards required


enough to offset the additional cost of the RON for
restoration. However, the lo
nger restoration times and enhanced enigneering complexity are, in our opinion, hard to
Figure 6: Cost Breakdown of Architectures















be justified by the small savings. In contrast, the Modified Protected Architecture is marginally more expensive than
the baseline architecture. As router reliability i
mproves, we expect that the majority of the network failures will be IP
link failures, resulting from failed optical components or fiberspan failures. In this case, the Modified Protected
Architecture could be attractive since it reduces the restoration ti
me for link and fiberspan failure with only a modest
increase in cost. Some of the key mechanisms required for implementing the Hybrid and Modified Protected
Architectures are demonstrated in [8].

3)

Impact of IP Traffic

Figure 6 compares the capital cost

of the four architectures
for a fixed IP forecast traffic matrix and RON direct service
traffic. In order to verify our observations regarding the four
IP over optical architectures, we simulate the impact of IP
traffic to the total cost for each architec
ture. The network
topologies and the RON demands remain fixed at the original
values, whilst the IP traffic load is increased in powers of 2,
i.e., 100%, 200%, 400%, and 800% of original values to
reflect the assumption that the internet traffic almost dou
ble
per year [9]. Figure 7 shows the results after normalizing by
the cost of the Baseline architecture of 100% IP traffic. We
find that the costs of the four architectures increase as IP
traffic load increases with almost the same ratio. The cost of prote
cted architecture is much higher than the other
three architectures. The hybrid architecture is slightly lower while the modified protected architecture is marginally
higher in cost than baseline architecture.

4)

Impact of Optical Traffic

Currently, most o
f optical network direct services are
relatively small volume with small granularity. In our previous
simulations, we used STS
-
1 granularity of optical services,
which reserve STS
-
1 granularity restoration capacity in optical
network links. However, we ass
ume the IP link channels are
OC48 granularity and require OC48 channel restoration. Thus,
small granularity and small amount of optical network
services may not be able to provide enough restoration
capacity to share with IP service OC48 restoration. To
si
mulate the impact of optical traffic to the total cost of each
architecture, we fix the network topologies, IP traffic matrix,
and change the optical network demands by increasing the
connection rate from STS
-
1, to STS
-
3, STS
-
12, and STS
-
48,
and fixing the

number of connections between each
Figure 7: Normalized cost of IP traffic

Figure 8: Normalized cost of RON traffic

<source,destination> pair. Figure 8 shows the results after normalizing by the cost of the Baseline architecture of
STS
-
1 granularity optical network traffic. With higher rate of optical network demands, the costs of thr
ee
architectures except baseline decrease due to the increasement of restoration capacity sharing in optical network as
we expected. Thus, in our simulated network, if there are enough STS
-
12 and STS
-
48 RON service demands, our
proposed modified protected
architecture can achieve the same or less cost as the baseline architecture.

F.

C
ONCLUSIONS

Our results illustrate that using an opaque re
-
configurable optical network to carry ISP backbone links cannot be
cost
-
justified based on current technologies and IP

backbone requirements, which constrasts claims of other
published papers. Although the Hybrid and Modified Protected Architectures can be implemented at similar capital
cost as the Baseline, network engineerings are more complicated. However, if router re
liability improves and the
larger optical network demands increases, the Modified Protected Architecture could potentially provide improved
restoration times with less capital cost, thereby justifying the additional network engineering complexity.

Referen
ces

[1]

S. Philips, R. Doverspike, and N. Reingold, “Network studies in IP/optical layer resstoration,”, OFC 2002.

[2]

S. Sengupta, V.Kumar, and D. Saha, “Switched Optical Backbone for Cost
-
effective Scalable Core IP Networks,” IEEE Comm. Mag. 2003, June.

[3]

J. Lab
ourdette, E. Bouillet, S. Chaudhuri, “Role of optical network and spare router strategy in resilient IP backbone architecture
,” DRCN 2003.

[4]

A. Basu and J. Riecke, “Stability issues in OSPF routing”, Sigcomm 2001.

[5]

P. Pan
et al (Ed),
“Fast reroute extensions
to RSVP
-
TE for LSP Tunnels,” Internet draft,
draft
-
ietf
-
mpls
-
rsvp
-
lsp
-
fastreroute
-
04.txt, 2004.

[6]

P. Pongpaibool,
et al
., “H
andling IP Traffic Surges via Optical Layer Reconfiguration
,” O
FC 2002.

[7]

G. Li, et al, “Efficient distributed restoration path selection in shared mesh restoration”, INFOCOM 2002, New York.

P. Sebos, J. Yates, and G. Li, "Ultra
-
fast IP link and interface provisioning with applications to IP restoration," OFC 2003.


Thi
s word document was downloaded from
http://www.worddocx.com/

please remain this
link
information when

you

reproduce , cop
y
, or use

it.

<a href='http://www.wordwendang.com/en'>word documents</a>


[8]


[9]

K. Coffman and A.
Odlyzko, “Growth of the Internet,” Optical Fiber Telecommunication: Systems and Impairments, Academic Press, 2002.

[10]

G. Li
et al
, “
Economic Analysis of IP/Optical Network Architectures” , OFC 2004.