Finaloptica.2012

Libro para traducir

Metro Edge Solutions
Similar to traditional taxonomies, the metro edge will continue to represent
a merging between the core interoffice and the client-access spaces. Here it is
becoming increasingly evident that SONET/SDH is not the best unifying layer
[42]. Conversely, since DWDM is bandwidth-inefficient for subgigabit linerates,
advanced electronic multiplexing technologies are needed to aggregate
diverse end-user protocols onto large-granularity optical (DWDM) tributaries
[46]. Dense IC technologies are finding particular favor here, helping collapse
legacy multiplexing hierarchies (i.e., "system-on-a-shelf/card" [141]) and further
blurring traditional access boundaries. Many new metro edge solutions,
broadly termed as optical edge devices (OED) [63], have been proposed, including
DWDM edge rings, next-generation SONET/multiservice provisioning,
and IP routing/packet rings. Some details are now presented.

6.1 DWDM EDGE RINGS
In Section 5.2.1 it was stated that low-cost passive optical rings are generally
well-suited for hubbed traffic patterns and can serve as metro edge solutions,
i.e., metro DWDM access [64, 65]. However, a key issue is mapping client
protocols onto the underlying wavelengths, and several solutions are possible (see
Fig. 8.15). A straightforward, transparent approach is to assign a complete
wavelength to each client signal, regardless of bit rate (termed "protocolper-
lambda"). This works well for low demand/node counts and sufficient
wavelength channels. For example, [87] proposes to back-haul individual subrate
client tributaries (DS1, DS3, OC-3) across a dual-homed DWDM (access)
ring to a CO for aggregation/termination. Nevertheless, given the propensity
of metro-edge DS1/DS3 subrate tributaries, this approach cannot scale in
wavelengths and is very cost/capacity inefficient for rates below the "breakeven"
OC-12/STM-4 value, see Section 5.4. Clearly, edge aggregation must be
coupled with static DWDM rings in order to improve efficiency and reduce
wavelength port requirements, as shown in Fig. 8.4.
A variety of subrate "circuit" multiplexing schemes can be implemented.
For legacy TDM support, dense IC chipsets can reduce many multiplexing
hierarchies onto Hne cards, e.g., 4:1 OC-3 to OC-12, and even OC-12
to OC-48 (Fig. 8.15). These "integrated SONET/DWDM" interfaces [40],
also termed "thin mux" [51], can combine the benefits of both technologies
and further improve cost effectiveness. For example, in [40], integration of
SONET/SDH line termination functionality with DWDM transport is found
to yield significant savings in electronic protection overheads. The availability of
newer software-programmable SONET/SDH transceivers (OC-3/12/48)

further improves flexibility, as line rates can be adjusted per demand, eliminating
the need for constant line card upgrades. In some cases, a complete subrate
DCS switching unit can also be added to aggregate multiple flows. However,
SONET/SDH multiplexing is only amenable for legacy voice/leased-line traffic
and not native packet interfaces (e.g., 10/100 Mb/s, 1 Gb/s Ethernet). The latter
require expensive "telecom adapter" (mapping) interfaces, more than quadrupling
overall interface costs (electronics, labor) and yielding high bandwidth
inefficiencies [46, 133], see also Section 6.2.
More flexible TDM multiplexing techniques can also be used for metro
edge-aggregation. For example, proprietary "asynchronous" multiplexing
can combine multiple subrate circuits onto a higher-bit-rate time-division
carrier. The resultant protocol concurrencies can be much more efficient
and can include a mixture of SONET/SDH and other alternate data tributaries (e.g.,
155 Mb/s OC-3, 200 Mb/s ESCON, 1 Gb/s Ethernet). Recent developments in
digital wrapper standards [12,13] will also facilitate more flexible TDM multiplexing
schemes. Digital wrappers define client-independent
overheads for transport and management of payload bit streams across optical
domains, for example, bytes for management, monitoring, protection
signaling, even FEC (about 6% FEC overhead [145]). These overheads are
processed at "electronic" (opaque) monitoring points, such as boundaries between
access/core rings/domains. Currently, several "wavelength" rates are defined,
namely 2.5, 10, and 40Gb/s, albeit subrate multiplexing hierarchies are not
defined [12, 145]. Conceivably, a full variety of protocols can be multiplexed
into the payload section, unlike rigid SONET hierarchies, although there can
be FEC implications (see [145]). Nevertheless, digital wrappers will inevitably
entail similar overhead processing complexity as SONET/SDH and related
chipset costs will likely relegate this technology to long-haul/regional transport
and/or for larger (metro) interdomain interfacing functionality for the
near/medium term.
Optical frequency division multiplexing (O-FDM) has also been proposed
for edge multiplexing, using a single laser to modulate "subrate" carriers
(i.e., client channels) within a spectral band. Specifically, all signals undergo
quadrature amplitude modulation (QAM) and are subsequently
frequencymultiplexed onto a fiber/wavelength (i.e., O-FDM/DWDM ring
combination).
O-FDM is genuinely bit rate transparent and can improve spectral efficiency
over on-off keying (OOK) modulation schemes used in SONET/SDH or Gigabit
Ethernet encoding (between 20-50%, 20 Gb/s per wavelength possible).
O-FDM transmission is also more dispersion tolerant, and can work well
on older fibers, for example, high-PMD types, unlike 10- or 40-Gb/s TDM
[44]. Additionally, related electronic costs are lower, since speeds need only

match slower subcarrier channels. Studies for moderate demand scenarios
show O-FDM to be more fiber efficient than OC-48 and more capacity efficient
than OC-192 [44]. Nevertheless, DWDM-induced transmission impairments
for O-FDM transmission may be problematic, and this needs proper
characterization. Note that only DWDM technology can transport O-FDM signals
in their native formats.
2 ''NEXT'GENERATION''SONET/MULTISERVICE
PROVISIONING PARADIGMS
Even though legacy TDM technology has many shortcomings (Section 3.4),
it will continue to play a significant role in the convergence of data and optical
networks at the metro edge. Demand for short-haul SONET/SDH gear is
still high and may continue to grow for the next several years [63, 65]. A large
part of this market comprises larger OC-48/STM-16 and OC-192/STM-64 systems,
although smaller OC-3/STM-1 and OC-12/STM-4 systems will also see
increased deployments (see [63]). Moreover, many existing routers/switches
have SONET/SDH interfaces (e.g., POS, AAL5), and recent efforts to define a
broader generic framing protocol (GFP) [14] (for mapping "nonstandard"
data protocols) may further propagate the ubiquity of such framing [158].
In light of this, many proposals have sought to "enhance" SONET/SDH
paradigms to better suit data traffic needs [130, 131, 133-135, 141-144].
Although these proposals have appeared under different names (e.g., "super
SONET" [63], "data-aware SONET" [130]), herein the term "next-generation
SONET" (NGS) is chosen (Fig. 8.4). Overall, all these solutions share two
main features, namely efficient data tributary mappings and integrated higher
layer (two/three) protocol functionalities, as shown in Fig. 8.16. Concurrently,
these solutions also leverage ubiquitous SONET/SDH performance monitoring,
protection switching, and network management capabilities. Some details
are presented.
SONET/SDH mapping of smaller packet interfaces (10, lOOMb/s Ethernet) is
usually done in "coarse" STM-1 increments and the resultant bit-rate
incongruencies usually yield large amounts of stranded bandwidth [129, 144]
(e.g., 10-Mb/s Ethernet allocated a full STS-1, 80% unused capacity). Bursty
data profiles can further exacerbate bandwidth inefficiencies. Nevertheless,
advanced IC technologies are permitting high-density switching fabrics with
much finer TDM granularities, particularly at smaller/fractional VT1.5 levels
[141, 158]. By combining finer tributaries, for example, virtual concatenation [158]
(wideband packet-over-SONET [130]), native packet interface rates can now be
matched much more closely (e.g., 10-Mb/s Ethernet via seven
VT1.5's). Multiple "matched" tributaries can then be more efficiently packed
into existing standardized tributaries, and this will help collapse multiplexing

(equipment) hierarchies. Switching designs can also use multilevel DCS fabrics to
assign capacities in both VT1.5 and larger STM-1 increments to better
scale electronic complexities [129] (Fig. 8.16). Overall, advanced DCS designs
will extend ubiquitous TDM tributary add-drop/switching/protection functions to
cover a full range of combined streams, in addition to interworking
with legacy streams (i.e., from ADM, W-DCS, B-DCS gears). Furthermore,
more advanced renditions are possible that dynamically adjust allocations to
"match" bursty loads on incoming interfaces (albeit layer two/three
buffering/processing and end-to-end signaling are required here). Along these
lines,
there have been notable developments in the link capacity adjustment scheme
(LCAS) [172] mechanism. LCAS defines a control protocol that allows for
"hitlessly" increasing/decreasing the number of "trails" (e.g., STS-1 circuits)
assigned to a connection. Moreover, each circuit trail can be diversely routed
to improve resiliency and failed trails can be removed altogether. Additionally,
connection asymmetry also can be achieved by assigning a different number
of trail counts to a given connection direction. Overall, LCAS defines a very
powerful new capability for exploiting virtual concatenation techniques and
improving capacity utilization (see [172] for more details).
To further improve data efficiency/scalability, NGS designs intend to provide a full
range of higher-layer "non-SONET/SDH" protocol functionalities
(i.e., "data-aware" TDM interfaces). Examples include IP routing, ATM
switching, LAN switching, and even frame-relay aggregation, see [4, 130,
133, 135, 142, 144]. Namely, intelligent layer two/three cell/packet processing
(e.g., buffering, scheduling, switching, routing, Fig. 8.16) capabilities are
used to increase capacity oversubscription ratios (e.g., statistical multiplexing
gains) between multiple customer ports [143], a step beyond "circuit"
aggregation. Many designs also provide direct data (Ethernet) packet interfaces,
eliminating the need for more expensive "telecom adapter" private-line
interfaces at client switches/routers. Additionally, line-termination capabilities
can also be added to extract and process payloads from existing privateline data
interfaces [130]. This essentially "decouples" link interfaces from
their associated data payload/protocols, an important step in extending the
benefits of oversubscription to private-line traffic (Fig. 8.16). Traffic multiplexing
coupled with tributary concatenation achieves aggregation closer
to the edge, leaving more free capacity inside the ring and yielding very
good bandwidth efficiencies. For example, three 10-Mb/s Ethernet streams
averaging 3Mb/s can be edge-buffered and packed into six VT1.5 circuits
versus three OC-1 legacy interfaces, a bandwidth savings of 94%. Note
that edge-multiplexing of multiple packet interfaces also reduces port counts
and the need for complex, costly centralized back-hauling setups [131].

Overall, integrating formerly distinct packet/cell and SONET/SDH protocols onto a
common platform removes multiple subtending aggregation
gears (routers, switches, ADMs) and their associated management systems.
This reduction can yield significant operational cost/provisioning complexity
improvements and reduce footprint space/complex cabling considerably.
Moreover, the emerging generalized MPLS framework (Section 7.1) presents
a comprehensive control setup for NGS systems, i.e., edge equivalence
mappings between circuit and packet labels and more recently, even provisioning
of concatenated SONET/SDH tributaries (see [153]). As an aside,
note that some earlier schemes proposed using ATM as the primary multiservice
SONET/SDH aggregation layer [4, 144]. However, these designs
suffered from high bandwidth inefficiency (about 20% [158]) and hardware
scalability/cost concerns, and have been largely usurped by improving IP
paradigms [65].
Despite its capacity improvements, NGS still reuses rigid, synchronized
electronic payload framing/encapsulation formats. As a result, this solution
is not truly capacity scalable (i.e., electronic bit rate and cost limitations) and
is much better suited to improving time-slot packing on existing rings (OC-3,
OC-12) and/or for areas with limited demand growth or high fiber count [65].
SONET/SDH framing again precludes transparency, making it difficult to
support data protocols such as Fibre Channel, ESCON, or FICON without
proprietary handling [51, 141] (until the formalization of GFP at least [14]).
Moreover, the associated functionalities of such protocols may be too specialized,
still mandating the use of subtending gear. A more ominous concern with
NGS is that its associated packet/cell functionalities/features likely may not
match those of "best-in-class" solutions offered by specialized router/switch
vendors [65]. Here, many operators already have (or plan to deploy) separate
"best-in-class" gear/management systems and will be unwilling to accept
single-vendor solutions. Consequently, more generalized multiservice provisioning
platforms (MSPP) attempt to address these limitations by further
integrating DWDM functionality to boost transparency/scalability. In essence,
MSPP solutions combine NGS with DWDM (ring) technology (Section 6.1),
and have also been more aptly termed as integrated metro DWDM [64,65]. An
overview of an MSPP node is given in Fig. 8.16, where added passive DWDM
transport/ring functions are shown. To lower costs and increase flexibility,
many MSPP designs intend to add "optical" functionalities (multiplexing,
transport, filtering) in a modular fashion via line-card additions. Again, "allin-one"
MSPP solutions may be overly expensive and impractical, especially
if the existing base of legacy TDM and/or "best-in-class" routing/switching
infrastructures is large (see also Section 6.4).

On the subject of SONET/SDH enhancements, recent advances have proposed
increasing SONET/SDH line rates to 40Gb/s (OC-768/STM-256),
further propagating existing TDM-paradigms. Currently, 40-Gb/s transmission is
usually done by optically interleaving [2] four 10-Gb/s streams (channelized), as
commercial availability of electronic SONET/SDH overhead
processing/clock recovery circuitry at direct 40-Gb/s rates is still a ways ofF(i.e.,
OC-768c concatenated interfaces). Regardless of the interface type, dispersion
(slope) effects at these bit rates will restrict transmission distances significantly
(e.g., chromatic dispersion at 40 Gb/s is 16 times larger than at 10 Gb/s, yielding a
dispersion limit of about 25 km [106, 112]). This will hinder applicability
beyond small metro domains, and usually extensive dispersion compensation
and fiber characterization considerations will be necessary (as used in most
studies, see [111]). Moreover, bandwidth scalability at these increased bit rates
still falls well short of those yielded by DWDM. Furthermore, mapping OC-
768/STM-256 tributaries onto wavelengths (e.g., for transmission across core
metro rings) will likely require larger 100-Ghz spacings (and not 50Ghz) due
to interchannel crosstalk limitations. To an extent, this mitigates the gains of
increasing the channel bit rate. Overall, 40-Gb/s OC-768/STM-256 solutions
have yet to be deployed, and related technical and cost concerns will adversely
affect or delay their applicability in the highly cost-sensitive metro edge [88].
When they do emerge, such large TDM interfaces will likely interface with
larger metro core wavelength routing gears. Moreover, it is also conceivable
that cheaper multiplexed 40-Gb/s Ethernet router interfaces will emerge first.
Namely, these interfaces will simply multiplex four 10-Gb/s Ethernet streams
together, thereby avoiding many complexities associated with genuine 40-Gb/s
TDM clock recovery and/or header processing.
6.3 PACKET-BASED SOLUTIONS
Although DWDM rings provide significant improvements over TDM rings,
as discussed previously, they still embody a circuit-switching paradigm. It
is well known that circuit switching is generally less bandwidth efficient than
packet switching [4], and bandwidth utilization on circuit-multiplexed DWDM
rings can be very low for bursty data profiles [60, 166]. Nevertheless, the
emergence of "next-generation" packet-switching devices (Ethernet switches,
IP routers) is helping resolve many of these data inefficiencies. Specifically,
advanced hardware-based packet filtering [164] and switching technologies
[168] can now support line-rate input/output switching at full "wavelength"
tributary rates (OC-48c/192c, 10-Gb/s Ethernet) (e.g., via custom high-speed
ASIC solutions or even generalized network-processor chips). Hence these
nodes can serve as direct (POP) aggregation boxes for metro edge, even core,
DWDM rings, completely collapsing inefficient, legacy "leased-line" data
hierarchies (e.g., "evolutionary delayering," see Fig. 8.18). There have also

been significant improvements in the overall IP routing framework to support
"TDM-style" guarantees (bandwidth, delay, loss, etc.), namely the multiprotocol-
label switching (MPLS) framework and its associated resource reservation
protocol (RSVP). Overall, this emergent framework can support very fine
quality of service (QOS) levels via the integrated services model (Intserv) or more
coarse (scalable) class of service (COS) levels via the differentiated
services (Diffserv) model (see [154, 156] and related references). These
capabilities provide "soft circuit" setups that achieve high statistical multiplexing
gains and can vary bandwidth allocations per any given criterion (e.g., perport,
client group, application, etc.). However, various provisioning concerns
still need to be addressed before "carrier-class" services can be offered (e.g.,
hardware/control scalability, service survivability). In light of these, more
specialized packet-switching schemes are being developed.
Recently, the concept of "packet rings" has been proposed, aiming to
combine the salient features of "TDM-origin" ring topologies (i.e., simple
connectivity, high resiliency) with the advantages of packet switching (statistical
multiplexing, finer QOS), namely resilient packet rings (RPR, IEEE 802.17)
[162, 163]. Specifically, a new Ethernet-layer media access control (MAC)
protocol is defined to statistically multiplex multiple IP packets onto Ethernet
packets (i.e., layer-two). The MAC protocol itself is "media-independent," and
will be capable of running over various underlying networking infrastructures,
including SONET/SDH, DWDM, or dark fiber. RPR designs can provide separate
"layer-two" packet bypass capabilities at coarser granularities, and this
will relieve packet loads at the IP (layer-three) routing level and improve QoS
provisioning [158]. For example, sample RPR node design in Fig. 8.17 shows two
data priorities, or COS categories. Additionally, packet rings will also
provide a rapid "layer-two" protection signaling protocol, designed to match
the 50-ms timescales yielded by SONET/SDH [158]. The current RPR framework
focuses on two (i.e., dual) counterpropagating "rings" that can both
carry working traffic (i.e., no reserved protection bandwidth for bandwidth
efficiency). All control messages are carried "in-stream," making the control
strictly in-band. Additionally, (layer-two) destination stripping is performed
for unicast flows, unlike earlier source-stripping FDDI rings, permitting spatial reuse
of bandwidth (note that multicast and broadcast still require source
stripping, however). Collectively, the above features significantly improve ring
capacity utilization/throughput (i.e., bandwidth multiplication, see [158] for
details). Currently, a spatial reuse protocol (SRP) framework has been tabled
for standardization and is commercially available (amongst others), aiming to
provide all RPR features (e.g., protection switching, topology discovery, bandwidth
fairness, etc.). In particular, the related protection switching protocol,
termed the intelligent protection switching (IPS) protocol, is an architectural

counterpart to the SONET/SDH K1-K2 byte protocol.
Nevertheless, since packet rings have emerged from enterprise LAN requirements,
they clearly cannot support legacy TDM traffic (without proprietary
mappings). Moreover, since RPR nodes must perform "electronic" packet
processing operations, realistically, their scalability to high speeds (lOGb/s and
beyond) and large node counts needs to be proven [158]. As such, they are most
suitable for new IP-based carriers [65], at least initially, providing very low-cost
metro solutions. Most likely, initial deployments will run over smaller-scale
metro edge rings, and here, packet ring/optical ring interworking will become
an important issue (see [75] for early discussions on this topic). Nevertheless,
it is likely that future advances in optical packet switching (Section 8) will be
leveraged to design substantially faster terabit packet rings. Overall, this is an
evolving area, and more work will emerge (i.e., standardization, design, and
performance evaluation).
6.4 MIGRATION STRATEGIES
Metro edge evolution will likely exhibit high variability due to the diversity
of subrate client protocols and available solutions. Ultimately, any chosen
solution will depend very much upon existing infrastructures, economic
considerations, and client/operational needs. Many metro edge networks still run
at lower TDM bit rates (OC-3/STM-1, OC-12/STM-4), and therefore, relatively large
capacity expansions can be cost effectively achieved by simply
upgrading to higher bit rate TDM systems [54]. This is especially true for
moderate demand growth and smaller tributary rates (DS3, OC-3/STM-1)
and/or fiber-rich scenarios. Meanwhile, newer operators with little/no existing
gear and more constrained fiber counts may prefer compact "data-efficient"
NGS/MSPP platforms to rapidly provision a full range of services (TDM and layer
two/three). Furthermore, those incumbents with costly "best-in-class"
routing/switching gear may adopt a more cautious strategy towards NGS,
choosing to deploy full solutions only for highly compelling price/performance
alternatives. Still other incumbents may prefer NGS, given its strong origins
from existing paradigms and finer-capacity allocation capabilities. Meanwhile, for
native Ethernet traffic, clearly packet-ring technology will be
much more cost-effective than solutions using "telecom adapter" interfaces
(SONET/SDH, NGS). Moving forward, this will likely be the solution of
choice for newer data-centric operators without legacy clients. For example,
packet rings will offer very low-cost aggregation between residential (Internet)
cable and DSL hubs.
Although the above alternatives may prevent immediate deployment of
DWDM technology in the metro edge, in the longer term it remains the
most scalable and complementary solution [63, 64]. Namely, DWDM rings

can agnostically support all other solutions (e.g., by reserving different wavelength
sets for SONET/SDH, NGS, and IP routing solutions) and will clearly
decouple operators from continuing fluctuations in technology directions.
More importantly, DWDM rings will allow operators to easily expand service
offerings (e.g., legacy TDM voice/private line to data or vice versa).
Most likely, many larger operators with existing legacy gear and diverse,
specialized "higher-layer" systems (IP routers, ATM switches. Fibre Channel hubs,
telephony switches) will deploy passive DWDM rings with flexible
edge aggregation interfaces to consolidate their architectures [35]. This will
ensure abundant capacities for any future demand "spikes," and also address
the growing "wavelength services" market (gigabits to customer edge [62,
64]). Other operators who choose NGS gear may also move to modularly
add DWDM capabilities in the future (e.g., flexible MSPP solutions). A
key planning/costing activity will be choosing when to cross over to optical
ring architectures (see also Section 5.4). For example, some operators may
move from OC-48/STM-4 rings to DWDM rings rather than upgrade to lessscalable
OC-192/STM-16 systems. Clearly, metro edge evolution requires more
defining studies, see [63-65] for market-related considerations.
7. Network Standards
Interoperable standards are a key factor in ensuring the success and adoption of
next-generation metro optical networking solutions. Standards help
to properly formalize both features and functionality, and will also help insulate
operators from single-vendor solutions. The key components of optical
interoperability are now beginning to emerge. At the physical and link layers,
many interface standards are well-defined, for example, SONET/SDH
concatenated formats, ITU-T wavelength grids, IEEE Ethernet interfaces, OIF
interfaces, etc. Increasingly, higher-layer control and architectural issues are now
being considered, and the ITU-T optical transport network (OTN) architecture
defines three layers of transport (channel, multiplex, transport) [12,
145]. Meanwhile the IETF and OIF are beginning to tackle more detailed network
signaling/protocols issues [157] and the ANSI TlXl is studying optical
ring frameworks. Perhaps the most notable development is the multiprotocol
lambda switching (MPXS) [148, 149, 157] framework, superseded recently
by the more generalized (emerging) multiprotocol label switching (GMPLS)
framework [150, 154]. GMPLS represents a strong push to increase horizontal
control plane integration (data and optical) by extending/reusing existing
data networking concepts/protocols. The overall aim is to replace the features
of multiple protocol layers in traditional multilayered models (e.g., separate
addressing schemes, SONET/SDH protection, ATM traffic engineering) with
a more unified solution, as shown in Fig. 8.18. A brief summary is presented
here (refer to related references for details).

7.1 CHANNEL PROVISIONING
There are several major required components for dynamic channel provisioning
and advanced SLA management in metro optical networks, namely
setup signaling, resource discovery, and constraint-based routing [7]. GMPLS
implements all of these requirements by extending MPLS signaling and resource
discovery protocols and defining multiple link-specific abstractions
of the original MPLS label-swapping paradigm (i.e., "implicit labels" for timeslots,
wavelengths, and fibers), see [148,149,153,154]. These definitions can be
further coupled with hierarchical label-stacking schemes to exploit scalability
(e.g., packet labels into TDM circuit labels into lambda labels). In particular,
this ubiquity make GMPLS an ideal control framework for multiservice metro
edge platforms (Sections 6.1 and 6.2).
First, optical channel setup signaling is accomplished by extensions to
MPLS signaling protocols, namely RSVP-TE (RSVP traffic engineering) and
CR-LDP (constraint-routing label distribution protocol) (see [148, 154] and
references therein). Here, the explicit-routing (ER) capability [149] is used
to indicate the channel route and reserve resources. Meanwhile, actual route
computation (i.e., RWA, Section 5.2.3) is done via constrained routing/path
computation (i.e., constraint-based routing (CBR) [148]). Moreover, CBR can
also incorporate advanced traffic/resource engineering algorithms for dynamic
ring/mesh networks. Finally, route computation requires network
topological/resource information (i.e., self-inventory capability), and this is
propagated
via extensions to pertinent routing protocols, namely open-shortest path first
(OSPF) and intermediate-system to intermediate-system (IS-IS) (see [154] for
full details). Examples include fiber-types, wavelength counts, wavelength
conversion resources, and possibly even analog metrics. More recently, resource
diversity information has also been proposed to explicitly capture risk associations
(physical, logical) [106, 155], and this can help channel-routing
algorithms improve the "disjointedness" between working/protection paths.
A key concern is provisioning architectures, namely centralized or distributed
architectures [7]. Data routing traditionally uses distributed control
(signaling, routing), whereas optical ring/mesh routing is much more amenable
to centralized implementations [7, 106, 155]. For example, many shared protection
schemes (Sections 5.2.2 and 5.3) require advanced optical ring/mesh
RWA algorithms with global per-connection state. Distributing/flooding such
information to all nodes is clearly unscalable. In other cases, if transmission
impairments are incorporated, the resulting computations themselves are less
amenable to distributed renditions. Nevertheless, many distributed shortestpath
heuristic RWA algorithms are still possible, and possible future advances
(components, algorithms) may permit more feasible distributed renditions (see

[1,7,128]). Regardless, the GMPLS framework can be applied for either model
(e.g., appropriate LSA extensions (distributed) and/or policy/route servers
(centralized) [161]).
7.2 PROTECTION SIGNALING
Dynamic optical rings, and likely even hybrid/mesh architectures
(Sections 5.2.2 and 5.3), must provide fast optical protection signaling protocols in
order to match the capabilities of SONET/SDH APS (i.e., 50-ms recovery).
Moreover, these protocols are necessary to implement advanced
service definitions (e.g., multilevel resource sharing. Section 5.2.2). Various
standardization efforts for optical protection are underway [15, 75, 95], but no
signaling standards currently exist. However, early proposals for fast optical
protection signaling in GMPLS have appeared [75, 95]. For example, [75,
94] discusses extending existing MPLS LSP protection signaling or defining an
altogether new optical APS protocol. Initially, APS protocol(s) can
be defined for rings (i.e., leveraging on SONET/SDH concepts), but subsequent
generalizations to hybrid ring-mesh networks can also be considered.
Meanwhile, [95] presents a new "lightweight" restoration signaling protocol in lieu
of RSVP/CR-LDP signaling. In general, until such standards are
defined, metro operators will continue to rely upon SONET/SDH protection,
ultimately delaying the introduction of dynamic optical services provisioning.
Assuming that fast optical protection signaling schemes will emerge, interlayer
protection coordination becomes an issue. Many metro-area protocols
have their own recovery mechanisms, operating across multiple domains (e.g.,
optical channel protection, SONET APS, MPLS LSP protection switching,
IP flow rerouting), and the simultaneous interference of such functionalities
can be very detrimental. Specifically, problems can include reduced resource
utilization, increased recovery times, or routing instabilities [96, 100, 102, 103]
(e.g., prolonged SONET/SDH recovery times. Section 5.2.2). DWDM can
also compromise higher-layer survivability, as the high degree of multiplexing can
lower higher-layer connectedness without proper preplanning [102].
Additionally, replicated (excessive) protection functionality across layers can
be very inefficient [98]. To date, no standards exist for multilayer protection
interworking, and this is largely done via careful preplanning, see [102] for a
detailed study. Moving forward, more formalized mechanisms are needed for
coordinating interlayer recovery actions between the packet/wavelength/fiber
levels, termed escalation strategies [94, 100, 103]. Various escalation strategies
are possible, such as bottom-up/top-down [7, 100] or serial/parallel [103],
and these will require complex interlayer signaling and hold-off timer mechanisms.
In particular, metro edge (NGS, MSPP) platforms handling many
protocols and their associated control/monitoring functions present some very
unique protection coordination possibilities. For example, routing diversity

information can be used to ensure higher-layer working/protection resource
separation. Overall, this is a complex area that requires much more work [98].
73 DOMAIN INTERFACING
Edge clients will need intelligent interfaces in order to automatically
request/release "optical" bandwidth (i.e.,"bandwidth-on-demand" applications.
Here, several interworking models have been defined to propagate
routing/connectivity information between the data (IP) and optical routing
domains, namely overlay, peer, and integrated models [154, 155]. The overlay
model achieves maximum separation using separate routing/signaling protocols in
each domain and defining an intermediate optical user network
interface (O-UNI) [146, 147, 151]. Conversely, the peer model achieves maximum
integration running the same (extended) routing/signaling protocols in
both domains. However, this proves overly complex/cumbersome, requiring
routers (optical nodes) to maintain/process optical (data) routing information. The
integrated model strikes a balance between the above two schemes,
running different instances of the same protocols (e.g., signaling, routing with
extensions) and using gateway protocols for end-point exchange (see [155]
for full details). In the near term, however, the overlay model will see most
favor since related UNI standards are available and proprietary optical control
protocols can be accommodated. Moreover, this model provides better
multiservice support, not just IP, and thus is well-suited for the metro space.
At the core of "optical" channel provisioning is the concept of a service definition,
as extended via an O-UNI or element/network management system
(EMS/NMS) interface. The O-UNI is intended improve vertical integration
between layers by allowing automated service discovery along with bandwidth
signaling functions (e.g., request/release/modify operations). In addition, a set
of generic signaled attributes are defined that can be mapped to subsequent
channel requests (e.g., RSVP/CR-LDP, Section 7.1; framing type; bit rate;
protection type; priority; etc.) [147, 151, 157]. These mappings can cover a
broad range of underlying capabilities (e.g., ring protection, mesh restoration, etc.)
[94]. Signaled attributes will help facilitate multiple service levels
for differing customer requirements, a necessary requirement in metro networks.
Overall, O-UNIs are very germane to metro edge platforms, and even
metro core nodes with direct wavelength interfaces. Several O-UNI definitions have
been tabled for standardization of which both the ODSI interface
[147] and OIF standard [146] have been completed. Meanwhile, interdomain
channel routing and protection coordination between operator networks will
require (optical) network-to-network interface (O-NNI) definitions and early
considerations are also underway here [152, 155].
7.4 NETWORK MANAGEMENT

As metro network elements continue to integrate many more diverse
interfaces/capabilities, especially at the metro edge, integrated network
management is obviously a major requirement [22]. Network management is a very
large focus area in its own right, and here only a brief discussion is provided
due to scope limitations. In general, the well-accepted telecommunication
network management (TMN) framework defines a hierarchical management
model comprising vendor element management systems (EMS) entities interfacing
with multivendor network management systems (NMS). Although
most early metro (DWDM) systems only provided proprietary EMS support
for point-to-point transport nodes [156], more advanced solutions are now being
offered. Optical channel (and link) visibility is of particular concern
here, and is complicated by the fact that there are still no standards for related
parameters. As an interim, detailed SONET/SDH Bl/JO overhead byte monitoring
(or digital wrappers equivalent) can be used at "opaque" points to
measure bit-level performance (e.g., errored/severely-errored seconds, etc.).
These "opaque" points can either be inside opaque nodes and/or at "edge"
interfaces in transparent networks. Note that some vendors are also beginning
to offer various (proprietary) sets of optical monitoring parameters, such as
laser powers/current/temperature, amplifier power, etc. [68].
Meanwhile, with metro networks supporting many more protocols,
advanced NMS solutions will be the key enablers for "end-to-end" services
management operating across multiple vendors' equipment [22]. Ideally, NMS
solutions should provide operators with a full range of functionalities that
are derived across multiple protocol domains, such as remote configuration,
performance monitoring, rapid fault detection/alarm processing, failure isolation,
diagnostics testing, and comprehensive logging/reporting, well-defined
graphical interfaces, etc. [43]. However, genuine multivendor services
management requires widescale adoption of standardized management
frameworks,
and overall this area is still in its infancy. Going forward, the common
approach here will likely be to adapt TMN concepts and develop appropriate
management information models between EMS and NMS systems [156].
8. Future Directions
As data traffic volumes continue to increase, packet-switching paradigms will
gain increasing favor, owing to their inherent statistical efficiencies [7, 165-
168]. Particularly, optical packet switching (OPS) designs are under study,
utilizing optical techniques to perform as much of the packet-routing operations as
possible (e.g., switching, buffering, even processing, i.e., "fourth
generation" optical networks). On the transport level, data packets are
sent directly over wavelength channels (i.e., "packet-over-lightwave" (POL),
Fig. 8.1). OPS nodes intend to achieve ultra-high packet throughputs, in the

multiterabits range, largely surpassing current gigabit router designs. Even
though all packet-switching/routing functions are difficult to perform optically (and
may remain so for the foreseeable future), various multifaceted
opto-electronic designs are being studied, and inevitably this work will lead
to significant improvements in packet-routing performance. In particular, the
three main functionalities pertaining to OPS are buffering, switching, and
header processing (i.e., label lookup) [2, 168]. Some brief details are reviewed,
and readers are referred to related references for more complete treatments.
By and large, OPS nodes have the same architecture as electronic packet
switches (i.e., input buffering, space switching, output buffering [168,169], see
Fig. 8.19). In packet switching, contention can occur if multiple packets are routed
to the same output port [165], and resolution is commonly achieved via
buffering. Most OPS designs use fiber delay line buffers, and recently, the use
of fast tunable lasers/converters has also been proposed to exploit the wavelength
dimension to store multiple (wavelength) packets in a given delay line
[167]. However, fiber delay line buffers constrain packets to multiples of a fixed
length, as there is no means to retrieve packets before minimum buffer delays.
Furthermore, large buffer sizes become costly/bulky (requiring complex sharing
setups), and additionally, fiber attenuation concerns will limit the number
of "circulations," (usually under 100 [165]). Hence, as a tradeoff, a mixture
of electronic memory and delay line buffering can be utilized [166]. Note that
there are also very interesting, early developments in all-optical memories (e.g.,
molecular transistors, see references in [167]). OPS switching fabrics, meanwhile,
can also exploit the wavelength dimension to reduce contention and
boost throughputs by orders of magnitude. However, nanosecond timings are
needed for packet transfers between switch ports, and this can be problematic for
MEMS technology. SOA gate technology has been considered here,
although careful design is necessary to control crosstalk [166]. Another switching
setup couples ultra-fast tunable lasers and wavelength converters with
passive wavelength routing devices (e.g., AWG) [169]. As component tunability
performances improve and integration technologies mature, this approach may
become very feasible. Finally, carefully note that unlike circuit switching,
OPS is still linked to the bit rate of the client signal, at least the header. Specifically,
even though payload flow is optically transparent, electronic header
processing/synchronization (and subsequent control of switching/buffering
resources) must be done electronically, as "all-optical" processing is not currently
feasible (see [168]). Clearly, electronic cost/scalability concerns can arise
for high-wavelength/fiber count systems and/or very small packet sizes, and
this may limit the complexity/range of label processing/packet filtering operations
performed. Consequently, various bit-serial packet-coding techniques

and/or guard-band schemes have been proposed to reduce electronic processing
bottlenecks [169]. Note also that transparent payload sections will suffer
from multi-hop optical degradations (loss, crosstalk), and this may require
all-optical regeneration to maintain transmission distances.
Overall, OPS will provide a good match for limited metro distances and
can reuse much of the existing packet-switching (MPLS, DiflHServ) protocol
suites [156, 166, 167]. Moreover, metro OPS will prove highly complementary to
emerging packet-based PON access solutions. Most likely, future
metro packet switching solutions will evolve towards hybrid opto-electronic
switching architectures (Fig. 8.19). Specifically, electronic switching/buffering
will be utilized for finer-granularity/more complex label-processing "edge"
operations, whereas OPS will implement less complex label-swapping
functionalities, achieving higher throughputs for more "aggregated" packet flows
(e.g., lOTb/s stated in [166]). This delineation potentially lends well to
"optical" packet rings, where more "coarse" stages (layer two CoS) can be
implemented using OPS. Even more germane to the metro arena, optical
packet- and circuit-switching paradigms can be integrated onto a common
platform, as the packet switches are still optically transparent to data payloads
[167]. Namely, lightpaths can be switched over the same OPS switching
fabric by simply "decoupling" the switching state from header-processing control
logic (i.e., bypass control logic and apply zero delay lines. Fig. 8.19). This
optical circuit bypassing will permit transparent legacy support (in exactly the
same manner as current DWDM systems. Section 5), and when applied to
the packet domain, will further improve scalability (i.e., eliminate per-node
processing of large transit/labeled packet flows). Overall, OPS is an exciting
new frontier, and future advances in optical processing/buffering will undoubtedly
yield fundamental conceptual evolutions in the metro domain. Note also
that slotted DWDM rings have also been proposed for access/metro data transport,
utilizing fast tunable transmitters and/or receiver devices and fixed slot
timings [170, 171]. Here, no optical buffering is performed inside the ring, and
instead, advanced multiaccess (MAC) protocols are used to arbitrate DWDM
channel slots in a fair manner between multiple users/traffic classes. These
rings require edge packet buffering but can yield improved efficiencies versus
circuit-provisioned rings, as the wavelengths are shared between multiple
source/destination points. A sample, advanced design using subcarrier sensing
techniques is studied in [171]. However, fixed slot timings are very inefficient for
variable-length IP packets, and proposed protocol amendments (e.g., multiple slot
sizes, backoff schemes [171]) entail further design/protocol complexity.
Moreover, access-control protocol timings may adversely affect scalability over
larger metro core distances.
9. Conclusions

Metropolitan networks occupy a strategic place in the overall network hierarchy,
bridging end-users with abundant long-haul capacities. Traditionally,
hierarchical SONET/SDH architectures have dominated the metro landscape,
with slower speed access rings interconnecting to larger, faster metro core rings.
However, as metro operators look to the future, many foresee surging bandwidth
demands, primarily driven by data traffic, and a plethora of diverse
clients with differing protocols and service requirements. As competition
intensifies, legacy multilayered architectures are proving overly sluggish and
unscalable in meeting complex, stringent service requirements. Clearly, metro
operators are in urgent need of scalable, flexible, multiservice bandwidth
provisioning solutions that allow for achieving a high level of service differentiation.
DWDM technology provides many benefits in the metro arena, including
scalable capacity, transparency, and survivability. Moreover, many
technoeconomic studies have confirmed the cost-effectiveness of DWDM for bit
rates
beyond OC-12/STM-4, bolstered further by falling component pricepoints. As
a result, DWDM technology has gained strong favor as a metro core solution,
and various architectures are possible (ranging from simpler point-to-point
transmission systems to dynamic wavelength-routing ring and mesh networks).
Nevertheless, given the large existing base of (SONET/SDH) fiber rings in the
metro area, network migration is a key issue. Very likely, the first step in this
migration will be a move to point-to-point DWDM "fiber-relief" applications,
and then onwards to more advanced optical ring/hybrid architectures. Meanwhile,
metro edge networks are evolving to represent a merging of the optical
and electronic domains, aggregating many user protocols onto large metro
core wavelength tributaries. Many metro edge solutions have been proposed,
ranging from DWDM edge rings, next-generation SONET/SDH, and multiservice
provisioning platforms, to "IP-based" packet rings. The choice of edge
solution will clearly depend upon an individual operator's needs, but over time,
those incorporating DWDM technology will be most beneficial and hence will
likely gain prominence.

Finaloptica.2012

More Related Content

What's hot (20)

Viewers also liked (8)

Similar to Finaloptica.2012 (20)

Recently uploaded (20)

Finaloptica.2012