Trends in Dynamic Optical Networks

Introduction

NOVEMBER 05, 2007

Discuss >


Today’s dynamic optical networks use various forms of software-controlled optical switching technology to help enable automation and reconfiguration. One increasingly common approach is to use remotely configurable optical add/drop multiplexers (fortunately, universally known as ROADMs) to route optical wavelengths through a network. The point is to break the inflexible one-to-one association between wavelengths and routes in traditionalWavelength Division Multiplexing (WDM) transport networks. To subsequently route optical services at capacities below that of an entire wavelength, and to create granular-service meshes, other devices such as intelligent optical switches can then come into play.
This increased flexibility, it is claimed, when combined with other technology developments, should hugely simplify network and service planning and provisioning, save carriers a lot of money, and generally make networks fit much better with end-user requirements for packet services in general (and forEthernet in particular). Ethernet matters a lot, because this is the way the telecom wind is blowing, and there are lots of things that carriers could do to improve their service offerings (and hence revenues) if their networks were more Ethernet capable.
Unsurprisingly, the ROADM market has been growing rapidly at the metro and regional levels since that area of application first emerged in earnest in 2005. In its September 2006 report, "ROADM & WDM Worldwide Market Forecast, 2006-2011," Heavy Reading forecast the ROADM market to grow more than twice as fast as the overall metro WDM market to 2011, when it should reach around $1 billion.
While ROADMs have been deployed in core backbones for a few years now, North American Tier 1 carriers have recently started their first rounds of deployment of the technology in metro and regional optical networks, and similar activity is taking place in Asian markets such as Japan. In Europe, ROADM deployment continues to be mainly focused on core networks, but this is expected to extend to the metro and regional levels over time.
So a lot of momentum is building up around this approach to dynamic optical networks as vendors looking for a metro optical growth opportunity pile in. But how do the various bits behind the ROADM story fit together? How strong are the reasons for making optical networks dynamic? Are the standards sorted? What are the alternative ROADM architectures, and how do their requirements and capabilities vary? And what of the underlying optical wavelength-switching components that make ROADMs possible? How are they keeping up with the tough performance requirements?
The market dynamics and drivers of optical dynamism are pretty straightforward, and include:
Network automation to lower costs Ensuring network survivability Deployment of fully meshed optical networks Support for new services
While it has been appreciated for a long time that automating network processes like bandwidth provisioning reduces operational expenditures, optical Ethernet has recently emerged as a key driver for network automation, according to Vinay Rathore, Ciena‘s director of product marketing.
“Ethernet is the fastest-growing service out there, and is becoming the universal choice for services such as voice, TV, business Ethernet, and mobile 2.5G/3G. And the 1- and 10-gig versions are all-optical services,” he says. “If operators are to improve margins and scale the network and improve overall profitability, automation is an absolute necessity.”
Further, with everything migrating to optical networks, their survivability becomes absolutely critical. Although current ring-based networks and 50ms automatic protection switching are widely used, Rathore points out that most networks today can be severely crippled with just two simultaneous fiber cuts. As the potential for future disasters can only increase as networks scale and environmental effects such as global warming intensify, networks need to be able to cope with multiple simultaneous failures.
An intelligent control plane that controls the optical network dynamically in real time provides a means of both automating Ethernet (and other) services, and of offering a range of protection and restoration schemes, including 50ms restoration, through enabling fully meshed optical networks.
“This level of automation allows a network operator to move beyond simple ring configurations towards a fully meshed optical network that not only allows networks to survive multiple failures, but also automates the network for delivering new services,” says Rathore. “Essentially, you can just point and click different ingress and egress points, choose your bandwidth, and provision your services.”
Further, mesh networks are inherently highly compatible with the characteristics of Ethernet, he says, whereas traditional rings are less efficient as they waste bandwidth through the use of preprovisioned restoration paths and cannot, for example, support dual homing. In contrast, in addition to supporting automated path setup/activation, dynamic protection path, automated provisioning, and 50ms restoration, an intelligent Ethernet mesh supports load sharing/ECMP and offers multiple restoration choices.
Additionally, a G.ASON-based control plane enables dynamic inventory management, facilitates Ethernet dual homing and diverse routing, accelerates Ethernet service delivery, and improves Ethernet service availability.
“Not only is the network more efficient and more survivable, but such automation can dramatically reduce the operational expenses, because the network itself is aware of all the inventory and all of the resources that it needs to operate,” says Rathore. “We have seen in business-case analysis that opex can be reduced by 85 percent, and that just by using a mesh-network implementation you can reduce your capex by 65 percent.”
And, of course, there is the potential for new service revenues that comes from point-and-click-style provisioning. A very simple example is the daily redeployment of the same network and transport bandwidth at different times of day between different applications. For example:
Cmmercial services from 8 a.m. to 6 p.m. (Ethernet) Residential video from 6 p.m. to 2 a.m. (MPEG over IP) Data recovery from 3 a.m. to 5 a.m. (Fibre Channel)
Today’s optical networks can be quite varied and complex, both within a single carrier and between different carriers. This diversity extends across both the data and the control planes, and, as networks become more dynamic, the control plane is putting extra pressure on optical networking technology. This is because there is a move away from changing network topology and behavior via the management plane – Element Management System, for example – towards having the network elements (such as routers and switches) respond or recover from changes in the optical network topology or behavior.
Figure 1 shows, at a high level, two different ways of organizing an optical network, depending on whetherITU Telecom ASON GMPLS orInternet Engineering Task Force (IETF) GMPLS interfaces are used in the control plane. So, if Operators 1 and 2 prefer to have a topology and hierarchy whereby they are not exposing information outside their network, they use the ITU interfaces (shown on the red lines). In this case, traffic enters (say) from the left through a UNI boundary that largely hides Operator 1’s network from the end users. This is both for commercial reasons and to protect the network from incorrect end-user behavior. Subsequently, the traffic may pass through an interoperator NNI into Operator 2’s network, where a similar hiding occurs. Finally, the traffic will exit through a UNI, here placed on the right.

Figure 1:High-Level View of Idealized Optical Network & Control-Plane Interfaces


Click thumbnail for full image.


Tier 1 carriers generally prefer this kind of hierarchical model, where intelligence resides in the network, because they protect themselves from end users and from each other. But a flatter alternative can be produced by using the IETF interfaces, where traffic tends to be driven at peer-to-peer levels by edge and user intelligence. This is shown by the black lines in Figure 1, running top to bottom, where traffic enters an operator’s network, moves across multiple operators’ networks and multiple devices, and leaves again into the end user’s network without much, if any, awareness of topology or hierarchy.
Both approaches can use a range of different control-plane protocols and variants, resulting in complex and diverse real-world architectures.
Standards
Generalized MPLS (GMPLS) is a key technology for dynamic optical networks. As its name suggests, GMPLS has a lot in common with the well-establishedMultiprotocol Label Switching (MPLS) used in packet networks (especiallyIP networks), but has been extended to support different sorts of nonpacket label information – for example, a label may be an optical wavelength. Table 1 compares some of the features of MPLS and GMPLS.
Click here to view Table 1.
“There is a great deal of commonality, but GMPLS has been driven by extending MPLS to support optical networks, rather than pure packet networks,” saysTony Downes, principal technologist of Data Connection‘s Network Protocols Division. “For example, both support packets, so you can do packet switching in GMPLS networks. Both are very strong on QOS and traffic engineering. Both have different resilience schemes because you want to recover either the first path or subsequently you want to switch over to a backup path in sub-50ms. In packet networks (MPLS networks) that’s called FRR (Fast Reroute); in optical networks it’s called end-to-end signaling (E-to-E). And there are commonalities in the use of different routing protocols with traffic engineering, but both have different schemes for doing peer versus overlay.”
GMPLS is being progressively defined by two major standards bodies: the IETF and the ITU. A third body, theOptical Internetworking Forum (OIF) , acts as a sort of vendor-sponsored referee on the very practical issues of getting the IETF and ITU standards to interwork, and on defining implementation profiles from the very rich standards material to ensure same-standards interoperability. So the OIF has taken a lead in defining UNI behavior, for example, and NNI behavior between providers, while allowing each provider to choose its own schemes internally. There is a common defined boundary between the various networks at the NNI interface.
The IETF has been developing optical networking GMPLS control plane technology primarily for a single-region network, where everything operates peer-to-peer. However, under pressure from carriers, various IETF drafts have recently emerged or been proposed for multiregional operation. The idea here is to preserve a single routing area, but to subdivide it into regions so that not all the network topology is exposed to the entity doing the routing calculations. Naturally, traffic-engineering calculations may become substantially more complicated with this approach.
“This is work that is ongoing, and the IETF has liaisons with the ITU, but I don’t believe that we will see unified standards in this area. There will continue to be IETF standards and ITU standards,” says Downes.
The ITU is developing the Automatically Switched Optical Networks (ASON) standards, which follow a somewhat more abstract, architectural, and hierarchical approach than the IETF’s implementation-driven one, according to Downes, and reflect the strong carrier influence. Arguably, it is a more realistic view in terms of what service providers are actually doing, and it means that the protocols in some areas are substantially more complex.
So ASON protocols can be very much more difficult to implement, or they encounter much more complex real-world scenarios about multi-area routes and traffic-engineering calculations prior to the setup of Label Switched Paths (LSPs).
Table 2 summarizes some of the characteristics and status of the IETF, ITU, and OIF work.
Click here to view Table 2.
There is a huge amount of other standards work that affects dynamic optical networks, and as Downes points out, the standards situation has become somewhat complex. As well as the IETF and ITU flavors of GMPLS, there are now additional technologies, such asProvider Backbone Transport (PBT), which is an Ethernet-only data-plane technology, but claims to provide a simpler, managed way of doing optical switching, with the benefits of protection switching and scaleable point-to-point connections.
The IETF and ITU are also providing transport versions of MPLS (T-MPLS) through optical networks, and there is activity in the IETF to extend GMPLS to dynamically control PBT.
“It continues to be a very active area, driven by service providers’ desire to have very dynamic networks. But there are multiple ways that people believe they can achieve that,” Downes says.
Theremotely reconfigurable optical add/drop multiplexer (ROADM) has rapidly emerged as the key element in current dynamic optical networks. It allows operators to provision wavelengths flexibly, and to reconfigure the topology of the traffic on networks at the granularity of a single wavelength.
ROADM networks also allow the creation of optically meshed or interconnected networks, thereby replacing the traditional need for optical-to-electrical-to-optical (OEO) conversion to interconnect separate optical network domains. All-optical mesh networks using wavelength-selectable switches (WSSs) can flexibly interconnect multiple optical domains directly.
These, and other, characteristics translate into many potential capex and opex savings for operators, as shown in Table 3.
Click here to view Table 3.
ROADMs, although initially considered mainly for high-capacity backbones, are now seen as being deployed at all levels of the network: long-haul, regional, metro, and access. Such wide potential deployment puts considerable pressure on the optical performance of ROADMs – in particular, the crucial ROADM optical component, the WSS.
Says Jonathan Homa, vice rresident for marketing and PLM at Xtellus, “Two parallel evolutions taking place are the emergence of mesh networking to interconnect multiple optical network segments, and higher-speed optical signals, with 40G beginning to occur, and 100G on the horizon. To meet these evolutionary trends, ROADMs will need to support very strong optical performance to transport higher-speed optical signals through a large number of cascaded nodes across interconnected networks.”
Signals cascading through up to 30 nodes with increasingly faster signals from 10 to 100 Gbit/s will mean an optical performance with:
Wide, flat passbands Low insertion loss (IL), low polarization-dependent loss (PDL), and low dispersion High extinction ratios for blocked channels Low static and dynamic channel crosstalk
Wide, flat passbands are needed to accommodate the spectral bandwidth of higher-speed signals, and to avoid ripple buildup – effectively squashing the signal into the noise layer – or any narrowing of the passband over multiple cascades. Moreover, since WSSs are also used for power equalization, it is essential that these passbands remain wide and flat even under attenuation conditions – to at least 15dB.
Low PDL under attenuation is also mandatory to accommodate the swings in polarization in optical systems. Typically, the loss variation due to PDL should be kept to fractions of a decibel.
“Perhaps as important as either insertion loss or polarization-dependent loss is the need for flat chromatic dispersion within the passband,” says Homa. “There is no means to compensate for the distortion this causes to optical signals. This is perhaps best measured – or perhaps best displayed – by examining the group delay that is mathematically related to dispersion. What is required is flat group delay across the passband.”
Low static and dynamic channel crosstalk are essential in a multiport device such as the WSS, as optical power from one port must have a high degree of crosssport isolation from power in adjacent or other ports – otherwise the WSS cannot function.
There must also be low first-capital costs. ROADMs are infrastructure and, as such, cannot be tied directly to a new revenue or a specific customer in the way an expensive transponder can be. So there will always be a greater than usual pressure to reduce ROADM costs.
Because the WSS is the engine and the most expensive part of a ROADM, Homa argues that this translates into a need for a family of WSSs with different capabilities and associated costs for different ROADM applications.
There are various forms of ROADM architecture, and Figure 2 shows some examples for a degree-2 ROADM, which is likely to be the predominant type of ROADM to be deployed in the coming years.

Figure 2:Examples of Degree-2 ROADM Architectures


Click thumbnail for full image.


The upper-left diagram shows the classic 1x1 WSS or wavelength-blocker architecture that has been widely deployed since the turn of the century. Optical power is split before the 1x1 WSS by using all the wavelengths available for a local drop through a demux. In a similar fashion, wavelengths are added through a mux.
The 1x1 WSS is used to block dropped wavelengths to make wavelength slots available for added wavelengths, and also to equalize the optical power of passthrough wavelengths to improve transmission.
The upper-right diagram shows a variation on this architecture that uses a 2x1 WSS. That is, it has two input ports and one common output port. The advantage here is that the WSS can equalize individually the power levels of wavelengths being added locally.
In the first two architectures, the dropped and added wavelengths have fixed physical associations with the demux and mux ports. The lower diagram shows an architecture for colorless add/drop by using a 1xN WSS. The drop ports on the WSS can dynamically select and can drop any wavelength. On the add side, a scheme to add wavelengths in a colorless way is shown using tunable lasers coupled to the outgoing fiber. This works quite efficiently for a small number of added wavelengths.
In contrast to Figure 2, Figure 3 shows various multi-degree ROADM architectures.

Figure 3:Examples of Multi-Degree ROADM Architectures


Click thumbnail for full image.


The incremental requirement over degree-2 nodes is the ability to send optical traffic to, and accept optical traffic from, other DWDM fibers. For instance, to interconnect two fiber rings, a total of four WSSs are required, making this a degree-4 node.
The left diagram shows a segment of a multidegree node with fixed add/drop. The first splitter routes DWDM fibers to other WSSs in the node. The second splitter drops local traffic as for a degree-2 node. An Nx1 WSS is then used to accept traffic from the other DWDM fibers in the node, as well as local traffic. For a degree-4 node a 4x1 WSS is required, for a degree-8 node an 8x1, and so forth.
The right diagram shows a variation for colorless add/drop. Here, a 1xN WSS is used both to route the DWDM traffic to other WSS in the node, and also to drop individual wavelengths. And an Nx1 WSS performs the converse function of accepting traffic from the other DWDM fibers and adding local wavelengths.
“This architecture has advantages in terms of flexibility and reducing the optical power budget, but, of course, is much more expensive as it requires two large WSSs for each segment of the node. For instance, in a degree-8 node, a total of 16 WSSs are required,” says Xtellus’s Homa.
Figure 4 shows some typical WSS configurations for use in different parts of the network. In Homa’s view, while these may change in the future, the current indication is that, for smaller WSSs, N ranges from 1 to 5 ports, and, for larger WSSs, it can range up to 16 ports – and probably higher in the future. It is also expected that, irrespective of the size of the WSS, there need to be versions available to support both 100GHz as well as 50GHz channel spacing, as both are widely deployed in optical networks.

Figure 4:Typical WSS Configurations


Click thumbnail for full image.


In terms of device costs, Homa sees these rising smoothly as the number of ports increases, because the material costs and the implementation complexity increase with the number of ports. This is true for both the main WSS technologies: liquid crystal and MEMS.
“Our view is that this is not an either/or situation, and that both technologies have different strengths,” Homa says. “For smaller WSSs, we believe that liquid crystal has clear cost advantages. However, for larger WSSs, MEMS can more easily provide the required flexibility to switch among many ports. So we see both technologies being used.”
Pulling all the different ROADM requirements together for applications in different segments of the network, as in Table 4, makes it clear that it is unlikely that a single-size-fits-all approach is viable for the optical switching components.
Click here to view Table 4.
“Towards the top of the network, as you get towards longer-haul applications, you require a performance-oriented ROADM, although obviously cost is important, too,” says Brandon Collings, director of the Optical Networks Research Lab at JDSU. “Towards the lower segment of the networks – the metro and access – performance becomes somewhat secondary as cost becomes absolutely critical. So you are talking about fewer channels and probably more degree-2 nodes.”
Accordingly, families of optical components are appearing to span the range of requirements and are supported by several key technology trends.
According to Collings, one of the ROADM features that is receiving a lot of interest currently is the ability to support colorless add/drop ports. In a colorless add/drop port the wavelength for each port is remotely provisioned, so each port has access to all wavelengths present. This contrasts to traditional colored ports, which have a permanently assigned wavelength associated with each port.
Advantages of colorless operation include:
Simpler faceplate, as 40 or 80 fiber pairs are not required Natural fit with wavelength-tunable transponders and dynamic networks as the operating wavelength can be changed remotely, and transponders can be pre-installed and the wavelength selected when they are put into service
However, colorless operation does drive WSS components to support higher port counts in order to provide sufficient numbers of add/drop ports, so Collings argues that WSS technology must be selected to support this trend.
A further trend, common to many optical systems and components, is functional integration, and the WSS is no exception. A typical example is the integration of an optical channel monitor within a WSS ROADM. This leads to circuit-pack consolidation and lower overall costs.
Wavelength-tunable transponders can now tune across the entire C-band, giving rise to the universal tunable transponder that can replace 40 or 80 dedicated single-wavelength modules. In principle, this greatly simplifies the planning and procurement processes because specific wavelengths don’t need to be identified early in the network design process. Further, with only one type of wavelength pack, technicians cannot install the wrong pack by mistake.
Collings sees further operational advantages of using such universal tunable transponders, as it allows transponder linecards to be pre-deployed in small groups periodically to handle bandwidth growth more efficiently.
“In a single visit you can deploy a handful of transponders and remotely provision the particular operating wavelength for each of those transponders as they are eventually put into service, without having to revisit that site each time,” he says. “Also, while in service, the wavelength and its path through the network can be modified and rerouted to relieve congestion points by doing some load balancing, or to provide service restoration around a fault network.”
Another key element of ROADM networks are theoptical amplifiers, which compensate fiber and component loss. With the flexibility of the ROADM network, the number of channels being amplified by these components is dynamic and subject to frequent change, or transient events, due to the flexibility of the network. So these amplifiers must maintain a stable performance during such changes so that pre-existing or surviving channels do not experience service outages. Transient-suppressed amplifiers are designed to minimize performance deviations during transient events. Without transient suppression, deviations in optical-amplifier gain can cause outages because of insufficient or excessive optical power output.
This concludes our overview of trends in dynamic optical networking. As stated before, this report is based on a Webinar, Deploying Dynamic Optical Networks, moderated by Sterling Perrin, Senior Analyst, Heavy Reading, and sponsored byCiena Corp. (Nasdaq:CIEN -message board),Data Connection Ltd. (DCL) ,JDS Uniphase Corp. (Nasdaq:JDSU -message board; Toronto: JDU), andXtellus Inc. You can view the Webinar for free, anytime, by clickinghere.
http://www.lightreading.com/document.asp?doc_id=133896