The internet traffic explosion has placed enormous capacity demands on service providers, challenged to supply the necessary capacity and flexibility while generating cash and properly managing assets to meet shareholder demands.
In the optical networking domain, dense wavelength division multiplexing (DWDM) has met the joint engineering/financial challenge by cost-effectively increasing the capacity of fibre assets.
Still, bottlenecks exist at optical-electrical-optical (OEO) switching points for high-bit-rate, high-channel-count systems.
In addition, high-speed, long-haul transmitter and receiver costs remain high. Many innovative all-optical technologies have been proposed to eliminate these switching bottlenecks.
However, little attention has been paid to how these technologies affect what matters most to investors - return on assets, cost of ownership and customer churn - if reliability is poor and provisioning takes too long.
In the long run, small, scalable, reliable all-optical switches may be the right choice to meet the engineering, reliability and financial requirements of the new optical network.
Some aggressive market research reports have projected exponential increases in budgets for optical switching equipment, but reality is far different.
Capital budgets are not increasing exponentially. Past spending indicates that perhaps 20% of these capital budgets would be spent on optical switching equipment.
The future for these budgets is more disquieting. Services such as voice are being priced down to zero margin, and profitable business customers are being lured away by new competitors.
The future top-line revenue stream will have difficulty supporting massive increases in capital spending.
Core Optical Cross-Connects
Optical cross-connects (OXCs) will be used to route wavelengths between inputs and outputs while adding and dropping local traffic.
Many carriers expect to need nodes with several thousand input and output wavelengths within a few years.
Consider: 25 fibre times 160 wavelengths per fibre implies a 4000 x 4000 OXC. The question is, how should these nodes be architected?
From an engineer's standpoint, the most flexible architecture is the opaque, wavelength interchange cross-connect (WIXC).
In the WIXC, all wavelengths are received and retransmitted at the cross-connect by transponders at each port.
In this architecture, any wavelength can be switched to any other wavelength on any fibre through transponder wavelength conversion.
However, the long-haul transmitter and receiver costs at these nodes for 10 Gbps and 40 Gbps data rates may be considerable. According to market researchers, the 2000 price for a 10 Gbps (OC-192) long-haul transmitter/receiver pair is about $5000.
In a 4000 x 4000 node, 4000 ports at $5000 means $20 million in component-level transmitter/receiver costs alone.
In the network equipment manufacturer value chain, the final price of a network element to the service provider contains other electronic control components, manufacturing and optical component costs, as well as the profit margin.
The final price can be estimated at five times the optical component cost. Therefore the service provider's price for this unprotected node would be $100 million, excluding the optical switch fabric, intra-office transceivers, amplifiers, mux/dmux and other components.
For the largest service providers, which may need 100 very large nodes, this adds up to $10 billion.
A similar calculation using 40 Gbps transmitter/receiver costs results in costs about 2.2 times higher.
These stretch the total capital expenditure for switching and transmission budgets and do not consider hundreds of optical add/drop multiplexers (OADMs), smaller regional cross-connects, services and future upgrades.
Of course, all these nodes would not be purchased in a single year, and cost, markup and cross-connect port count models may vary slightly.
However, even with the most optimistic numbers, the conclusion remains that long-haul service providers cannot afford to construct WIXCs with all those transponders around every port in all the large nodes in their networks.
Single Big Fabric or Multiple Smaller Fabrics?
Ultralong-haul transmission systems will enable unregenerated transmission over very long distances, and thus the possibility of transparent cross-connects that eliminate expensive OEO conversions.
Eliminating transponders does away with wavelength conversion. There have been advances in all-optical wavelength converters, but even if these devices materialise, they will still be expensive.
Network management, especially wavelength assignment and restoration, in a mesh network or subnetwork is challenging, but these challenges are reduced in highly connected networks with many wavelengths.
The previous financial analysis shows that there are significant incentives to overcoming the remaining challenges.
The most attractive transparent node architecture is the wavelength selective cross-connect (WSXC), which can be constructed out of a single fabric or several smaller fabrics.
The WSXC operates by switching all the 'green' wavelengths between fibres on one plane, the 'blue' wavelengths on another plane, and so on.
A WSXC node with 25 fibres and 160 wavelengths per fibre requires 160 25 x 25 switches or a single 4000 x 4000 switch.
Slightly larger switches allow dropping and adding of local traffic, allow some channels to be regenerated, and allow some channels to move between wavelength planes.
If one allows 'k' channels (where k represents the number of extra ports) for these purposes, either 160 [25+k] x [25+k] switches or a single (4000+k) x (4000+k) switch is required.
The WSXC architecture using small building blocks is more reliable, more serviceable, and has a lower cost of ownership than a single, large switch fabric.
Pay as You Grow
A 'pay-as-you-grow' business model using architectures of small fabrics that match capital outlays (adding new switches) with revenue streams (turning up new channels) is best at meeting shareholder expectations.
Using smaller switches, new fabrics are added when new wavelengths are turned up.
With large single fabrics, the entire capital cost of a node has to be paid on day one, making the architecture financially unattractive.
For example, compare a single $100 million investment vs five consecutive years of $20 million investments.
Using a conservative cost of capital of 10%, the net present value of the pay-as-you-grow approach is $83 million vs $100 million for the large fabric purchase - a significant penalty.
The only thing that service providers dread more than not meeting investor numbers is explaining to a government body why the failure of a very large node caused a very large outage.
A figure of merit for service providers is system availability, which can be derived from system downtime (derived from component failure rates), system diagnosis and isolation of a failure, and mean time to repair.
For an architecture where all traffic is switched in a single large fabric, an entire spare fabric is required, and a failure disrupts all channels.
But the fabric failure rate is likely to go up as the size and complexity of the single switch fabric increases.
In addition, it is difficult to see the problem diagnosis and failure isolation capabilities of proposed large fabrics.
Finally, managing 4000 x 4000 fibres for a failed, large, single fabric is likely to have a tremendous mean time to repair.
If an entire small fabric went down in the previous 160-fibre fabric example, at most 25 channels would be lost. The small fabric is easy to hot-swap with another small fabric because there are many fewer fibres.
System availability would be better, because it's easy to diagnose and contain faults and easy to repair. In addition, it is possible to implement a shared protection scheme with smaller fabric building blocks.
Bottom line: Trusting large nodes to a single, large fabric is a difficult proposition.
Opaque WIXC architectures will be required in future networks. These nodes will provide vendor interoperability, connect different networks, and connect 'islands of transparency'.
Creating these nodes with all-optical switch fabrics will benefit providers by creating protocol- and bit-rate-independent cores, eliminating forklift upgrades.
Smaller, scalable fabrics can create these nodes using architectures such as Clos.
Constructing these nodes from smaller fabrics will enable better fault isolation and pay-as-you-grow capabilities.
Implications for OADMs
The business case for all-optical OADMs is clear. Since only 25% of traffic at a typical node is dropped, eliminating all the OEO conversions from express channels (traffic not being dropped) cuts the large cost of long-haul transmitters and receivers.
One might think that WIXC nodes in the core can provide enough intelligence to route the right wavelength to the right fibres to be dropped, enabling some fixed or quasiflexible OADMs.
However, consider the case of a city inserting traffic on a node and dropping it later in that span at another OADM before arriving at another OXC.
In this case, the OADMs must be able to drop any channel to any port. Therefore future OADMs must be fully flexible, further enabling core WSXCs.
The technology exists today to architect completely rearrangeable and reconfigurable OADMs. OADMs that use 2x2 optical switches or 1x2 switches will not meet the fully flexible needs of OADMs in the future because they can only add or drop a channel to a specific port.
WSXCs need to be large enough to switch wavelengths between fibres with the required extra add/drop ports (ie, [25+k] x [25+k]. WIXCs need a building block that can scale to the right size but is small enough to provide the necessary fault isolation capabilities.
OADM switches must be large enough to meet complete rearrangement requirements.
Given the above necessities, a 32 x 32 all-optical switch is an excellent size building block to architect this optical network.
Certainly, service providers want to have the most flexible, 100% wavelength interchanging ports at all core cross-connect nodes that use existing management and restoration algorithms.
Unfortunately, the sheer cost of this proposition with all the transponders required, makes this difficult or impossible under current financial and competitive circumstances.
A cost-effective network could be constructed of large, transparent WSXC nodes in the core of the network. Some opaque nodes would be used to interface with other networks, connect islands of transparency and provide vendor interoperability.
Fully rearrangeable and reconfigurable OADMs would line the spans of the network. Switch fabrics for these elements would be constructed of smaller, scalable, all-optical switch fabrics that eliminate transponders and forklift upgrades.
These switches, in combination with new ultralong-haul transmission equipment and new wavelength-layer software, will enable service providers to provide capacity for exploding demand at the right cost and reliability.
