3.2.2.1  Switching of wavebands

The switching of wavebands is particularly interesting in the following cases:

• To reduce the number of switching elements inside systems. This is the case for a large number of optical cross-connects or switches.
• At the network level, when the traffic is aggregated enough to tolerate a waveband switching with a good bandwidth utilization. This is the case for pipes bridging networks and where the traffic matrix is quite stable.

The waveband switching really exploits the potential of optics because it reduces the complexity of the switching process with respect to electronic techniques. This technique has to be exploited as much as possible to make a system or a network concept really competitive to electronic techniques but only when the traffic matrix is stable enough not to penalize the average load of the waveband.

3.2.2.2  Switching of wavelengths

The switching of wavelengths is particularly interesting as follows:

• To switch at the line bit rate without time demultiplexing inside optical systems, which is an advantage with respect to electronic techniques, especially when the bit rate is high (≥10 Gbit s⁻¹). Generally, the wavelength switching is used to relax the power budget to offer higher switching capacity. In fact, with one important parameter being the optical signal-to-noise ratio, it is fundamental to preserve it to be able to address the higher throughputs. This is generally imposed when input powers launched into optical amplifiers cannot exceed a certain value. By this way, as the optical signal-to-noise ratio is always a function of the channel input power, if the total input power is the channel power, then we reach the maximum optical signal-to-noise ratio value in the architecture. This is the case for a major part of the optical cross-connects or switches.
• At the network level, when the traffic is aggregated enough and stable enough not to cope with traffic transient effects. This technique is still efficient when it is possible. The switching is done at the line bit rate. It is more efficient for backbones than for metro networks, for example, simply because of the traffic matrix characteristic.

3.2.2.3  Switching of optical packets

The switching of optical packets can be processed at the wavelength level or at the waveband level. The only difference with respect to wavelength or waveband switching is that the ON state is relatively short, on the order of a few hundred nanoseconds or microseconds. The switching of optical packets is particularly interesting as follows:

• To create a datagram connection or a virtual circuit connection inside optical switching systems. The switching can be done at the wavelength level (classical optical packet switching) or at the waveband level (we define then a WDM packet) for a short time. Like for the previous cases, the WDM dimension is preferred where possible to reduce the number of active components in the architecture considered.
• At the network level, when the traffic profile is sporadic, we have to cope with time constants that cannot fit with seconds or minutes, and where the packet connection is the only one realistic to exploit efficiently the available bandwidth. This is the case for metro networks, in particular, but also for backbones if the application bit rate is growing at bit rate.

3.2.3.1  Why optical packet switching?

When analyzing the traffic profile at the output of a local area networks (LAN), the sporadic behavior of the traffic, often modeled with self-similar functions, clearly points to the problem. We then need to adapt the network concepts to the traffic nature coming from the access. And how do we adapt such a variability of the traffic profile with circuit connections while having a good efficiency? The answer lies in the high aggregation level that requires a grouping of different LANs, and this is not always possible. Another solution is to cut the circuit into pieces (packets) trying to follow the traffic evolution at a scale comparable to the scale of the incoming traffic. Even if the technique is more complex to manage, it is undoubtedly the most efficient way to optimize the bandwidth utilization, more in the time domain today, than in the volume domain in the recent past.

3.2.3.2  What kind of packets: Fixed packet or variable packets?

There is still a debate on the choice of the packet size: for main arguments in favor of the variable packet, the optical packet size always follows the incoming packet size, whereas for the fixed packet, the optical packet format contributes to a better management of the performance. In both cases, arguments are acceptable, but the reality is more complex.

It is evident to say that where the contention can be managed easily (a case of small or simple topologies), the variable packet has to be envisaged. But in the case of a large topology (meshed or other), the resolution of the contention is then local in each node, and the control of the traffic profile inside the network becomes fundamental. In that case, the fixed packet format is the only reasonable format. Another alternative, probably the best, is the adoption of a concatenated packet. For best effort, the concatenation is created to really follow the incoming packet profile. For high priority traffic where the delay is fundamental, small packets will always experience the smallest delay in the network. In this way, the technique can be adapted to a multiservice environment (Figure 3.1).

Figure 3.1   Different optical packet types that can be adopted.

Figure 3.2   Comparison: variable optical packets versus fixed optical packets versus concatenated optical slots.

Figure 3.2 gives a comparison among fixed, variable, and concatenated packets.

3.3.2.1  In metropolitan networks

In the case of optical rings, the WDM dimension can be exploited first to provide an upgradability of the network in terms of allocated resources, simply by attributing progressively bands of wavelengths. The advantage of the band is mainly to relax the filtering constraints during the cascade of several nodes. This advantage was raised many times in circuit switching platforms. If we want to exploit the same WDM infrastructure while introducing packet techniques, the notion of band can then be advantageously exploited to reduce the latency in the transmitting parts. In fact, we can exploit the statistical time multiplexing of packets over a group of wavelengths to increase the chance to insert a packet in the line when we can have access to a group of wavelengths instead of one.

In summary, we can see that for optical rings, the WDM dimension can relax physical constraints and, in addition, improve the performance in packet rings in terms of latency.

To illustrate the benefit of the WDM dimension, the IST DAVID project is proposing a multiring optical packet MAN exploiting the WDM dimension for these two main aspects.

3.3.2.2  In backbone networks

In backbone networks where the topology is generally meshed, the WDM dimension is exploited for four main reasons:

• Cost and compactness reasons: In optical cross-connect/packet switching architectures, the WDM dimension can be exploited to reduce the number of switching components to make the architecture compact and low cost. The objective is to switch in the WDM dimension as much as possible in the limit of the required power budget. In cross-connect architectures, the dynamic management of the wavelength dimension can, in addition, contribute to a drastic reduction of the number of interconnected fibers.
• Power consumption reasons: By exploiting the WDM dimension, the processing is done at the WDM dimension. In electronics, a double demultiplexing is required: at the wavelength level and at the bit rate level. This is, for example, the case for coarse synchronization stages where the WDM dimension can be exploited efficiently at the WDM level to reduce the complexity of many electronic structures.
• Physical reasons: The number of channels switched can also vary inside the architecture to preserve the optical signal-to-noise ratio at the output. For example, in the first stage of an optical architecture, the WDM switching can be done on a large number of channels (8 or 16), and then progressively reduced to a lower number, converging then to one channel switched.
• Performance reasons: In packet architectures, the WDM dimension can be exploited to reduce the packet loss rate and the latency at the same time. When the packets can be addressed onto different wavelengths, the statistical multiplexing on these wavelengths let a freedom degree for the choice of the wavelength at the output of an optical packet switch. By this way, the contention can be avoided simply by reaffecting a new wavelength to a packet instead of putting it in a queue. The benefit is double because the packet loss rate can be improved, with the same amount of digital memory, and the latency is preserved because the packet will not experience a queue.

3.4.1.1  What about the WDM dimension?

The WDM dimension is expensive in the metro, but there are also some arguments in favor of the introduction of the WDM in this part of the network.

Due to the current traffic profile, not constraining the bandwidth utilization too much (because the native bit rate is still quite low), a circuit platform is interesting. To reduce the cost of the WDM dimension while having enough resources to cope with the traffic volume and profile, probably the waveband approach is the best one. It guarantees upgradability (sub-band per sub-band), physical performance (relaxing filtering constraints), and simplicity (circuit switching) with an existing infrastructure (fibers already installed). But if the native bit rate increases together with the variance, there will be a need to go into a lower granularity to have a better utilization of the optical bandwidth. Then the optical packet technique could be easily introduced making use of an existing infrastructure. The gap becomes natural.

Thus, in the following presentation, we will introduce the circuit switching technique as a first step with a progressive migration toward optical packet switching techniques, which lead to a really efficient platform.

3.4.1.2  Short-term introduction: Optical circuit switching technology

Due to the current traffic profile, optical circuit switching could be rapidly introduced in the market for many reasons:

• The optical technology is mature enough and commercially available to envisage its utilization in optical platforms. It requires active devices such as integrated laser modulators (ILMs), receivers, and basic passive devices (demultiplexers, couplers, filters, etc.).
• It is a very simple technique that can be implemented rapidly.
• The optical technology can exploit advantageously an existing WDM infrastructure. Currently, all the fibers installed are not exploited.
• The management of such a network is mature enough to propose products. Management studies have been carried out leading to clear information models, protection scenario, and well-defined monitoring techniques.
• The traffic profile is not yet so critical to envisage using such a technique in the metro. The bit rate at the output of personal computer (PC) is sufficiently low to have enough aggregation level at the output of the LANs.
• This kind of platform can be easily upgraded with optical packet techniques to be fully compatible with the future traffic profile driven by a powerful PC and fiber to the home (FTTH). This is also a strong argument for operators who want to invest with the possibility to upgrade their platform according to traffic constraints. Examples of implementation are ring topologies, star topologies, and mesh topologies. In ring topologies, we can find the single ring or multiring approach. Figure 3.3a and b illustrates these three interesting topologies.

The particularity of this approach is in the optical add/drop multiplexing structure. We can list mainly as follows:

• Passive optical add/drop multiplexers
• Dynamic optical add/drop multiplexers

The passive structure can be used when the traffic matrix is very stable, whereas the dynamic structure can allow some adaptation of the allocated resources to follow at least the envelope of the traffic matrix. This last type of the active structure is particularly interesting when the time constants are in the range of a few hours.

Figure 3.4a and b describes the two main structures of add/drop multiplexers.

In the metro part, the WDM granularity for the upgradation of the network capacity will depend on the cost of the intervention. And in some cases, the upgradation of one wavelength is not the most cost-saving solution. For that purpose, it is important to identify the minimum WDM granularity for the upgradation. This minimum granularity, which can be on the order of few wavelengths (2 or 4), can impact dramatically the network infrastructure. If we take into account the physical limitations in the cascade of filters, it is clear that the sub-band approach is an interesting approach. Figure 3.5 gives an overview of a ring MAN adopting a sub-band strategy.

3.4.1.3  Medium-term introduction: Optical packet switching technology

The optical packet switching technology could be introduced as a required second step simply to face the traffic profile evolution. In that case, a packet technique will be required on the basis of the infrastructure already installed. The upgradation is made simply by changing the optical ring access node. Two new sub-blocks are then mandatory: the opto-electronic part and the electronic interface compliant with different classes of services.

The more pragmatic approach is the adoption of a commercially available technology. Several concepts are proposed, all based on the adoption of ILMs and synchronous optical NETwork (SONET)-like receivers. The resilient packet ring (RPR) concept is probably the most representative one.

Figure 3.3   (a) Conventional ring topology. (b) Multiring topology.

3.4.1.4  Medium-term introduction: Full flexible optical packet switching technology

In the case of long-term approaches, the adoption of an advanced technology is then mandatory.

In particular, two important components are a fast tunable source and a fast wavelength selector. These two components have already been demonstrated feasible (e.g., agility for the tunable sources, nippon telegraph and telephone (NTT) for the wavelength selectors).

These components are:

• At the transmission side (tunable laser): limitation of the latency in the output queue by exploiting the WDM dimension
• At the receiving side (wavelength selector): the guarantee of the wavelength transparency to receive any packet from the optical bandwidth
  

  Figure 3.4   (a) Fixed optical add/drop multiplexer. (b) Dynamic optical add/drop multiplexer: an optical switching element guarantees the dynamic behavior.

  

  Figure 3.5   Sub-band introduction in an optical ring.

• At the transit side (wavelength selector): the guarantee to introduce some fairness mechanisms when a high priority traffic with a minimum of latency is required
• The structure of the optical add/drop multiplexer is depicted in Figure 3.6a and b.

3.4.2.1  Commercially available technology

There are two types of devices: passive devices and active devices

• For passive devices, we need couplers, attenuators, isolators, fixed filters, circulators, interconnection fibers, connectors, (de)multiplexers, etc. For filters and (de)multiplexers, we can have different specifications if we are addressing channel filtering or band of channel filtering.
• In all the cases, these components are products and today fit all the required specifications for system applications.
• For active devices, we will need lasers, ILMs, slow tunable lasers, optical amplifiers, photodiodes, etc. Particular interest must be devoted to optical amplifiers because we can distinguish two types of optical amplifiers: fiber-based amplifiers and semiconductor amplifiers. For fiber-based amplifiers, largely introduced in point-to-point transmission systems, the main preoccupation is to find the best way to make these devices very cost-effective. For SOAs, even if they are on the market today, they currently suffer from a small market. However, the generic potential of such a component for different functions, such as optical gating or optical conversion, makes this component a promising one for future applications.

3.4.2.2  Advanced technology and feasibility issues

This is the most promising technology to really propose something new and disruptive with respect to what exists today.

Figure 3.6   (a) Structure of the ring access node for an upgrade of the circuit switching platform into a packet switching platform. (b) Structure of the optical add/drop multiplexer including the required functionality to fully manage the transit traffic.

In the following, we will illustrate the four main components:

• The SOA for gating functions
• The tuneable source
• The wavelength selector, a strategic fast tuneable filter
• The packet mode receiver

3.4.2.2.4  Tunable switching sources based on a sampled grating—Distributed Bragg reflector structure

Another solution to build a fast tunable source is to use an Simple Grating Distributed Bragg Reflector (SG-DBR) laser while integrating a SOA section and an electro-absorption section. The following schematic illustrates the structure of the source. As for the hybrid tunable source, the SOA sees only a continuous wave that again prevents cross-gain modulation. The structure is currently studied by agility (Figure 3.8).

Figure 3.7   (a) Structure of a hybrid tunable source using a SOA gate array. (b) Photo of a four gate-array, key building block of the hybrid tunable source.

Figure 3.8   Schematic of an integrated tunable source integrating a SOA section for amplification of the signal or optical gating.

3.4.2.2.5  Wavelength selector

The wavelength selector is probably one of the other key devices because it can be comparable to a fast tunable filter. The principle of operation is very simple. A first demultiplexer demultiplexes the wavelengths, then each wavelength is selected or not, depending on the orders coming from the control part, and finally an output multiplexer regroups the wavelengths selected and contributes to reject the wideband amplified spontaneous emission coming from the SOAs. In principle, only one wavelength is selected among a group in a normal scenario. However, in the case of the third network scenario, the number of output wavelengths can vary from 1 to N (N being the total number of wavelengths at the input of the device).

Figure 3.9a shows the basic structure of a wavelength selector.

Figure 3.9   Structure of a wavelength selector.

3.5.1.1  Where smart routers and where high-throughput routers?

Smart routers are required for the dorsal network and where the aggregation is forced by the poor connectivity of the nodes and by the huge amount of traffic that must be transported. This type of product is particularly interesting when the traffic matrix is stable enough to make semipermanent connections realistic and efficient. This is particularly the case in the United States when connections have to be established between states. The router is mandatory to collect the traffic coming from regional networks or national networks.

High-throughput routers are required when it is not possible to process a part of the traffic with semi-permanent connections through optical cross-connects. It can be the case for metro core networks, where the dynamicity of the traffic could require a transfer mode at the packet level to increase the network efficiency while optimizing the resource cost.

3.5.1.1.1  Smart router

Figure 3.11 shows the global structure of a smart router. The router collects the traffic at the packet granularity. Packets are put into queues and are sent on a specific wavelength. The optical cross-connect has to manage high throughputs. The structure can be based on a MEM technology. The approach is very interesting when the traffic is strong enough to open large pipes, thus enabling the establishment of a waveband. The cross-connection at a waveband level is the best guarantee of simplicity and reliability without creating breakthrough between the transmission system and the switching cross-connect.

Figure 3.11   Smart router schematic.

3.5.1.2  Medium-term approach: Network concept based on O/E/O nodes

In the previous cases, solutions are based on a traffic profile assumption, which enables the circuit switching technique for long-haul networks or enables the packet technique for small-scale backbones. However, the problem will occur when the application bit rate is increased together with the variance. This can rapidly happen if merging low bandwidth connections coming from mobile phones and high bandwidth coming from more and more powerful PCs reinforced by an optical connection giving access to very high bit rates per user. In this particular case, the variance can be increased leading to huge problems for efficient aggregation and forcing the telecommunications companies to think differently toward an optical packet platform for the backbone.

This means that, due to the traffic profile evolution, the packet will be used extensively. And the key question could be as follows: how to realize an efficient network capable of handling the required capacity while providing a mandatory flexibility?

3.5.1.2.1  The generic approach

In the network concept, we mainly exploit the edges to prepare the traffic in such a way that the traffic constraints inside the core of the network are relaxed. This means that all the complex functions are located in the edges such as aggregation, switching per destination, classification, packetization, traffic shaping, load balancing, and admission control. The traffic profile, having a better shape, is then sent to the network. The core nodes will be responsible for the synchronization, the contention resolution, and the switching. In this case, the packet being created in the edges only simplifies the structure of the core router.

Figure 3.12   (a) High-throughput router schematic. A fast optical switching matrix could be adopted in the center of the architecture. (b) Photo of a 640 Gbit s−1 throughput optical matrix. (c) BER performance.

3.5.1.2.2  What type of packet format?

As a first introduction, and if possible, it can be envisaged to introduce an existing packet format. The G709 framing is currently being investigated to identify the potential of the concept, but other packet formats could be considered.

For a second introduction, it seems clear that a more smaller packet size is required to relax the problems of aggregation. This also imposes a standard on a packet that does not exist today.

It must be noted that some universities and laboratories are currently studying the possibility of managing variable packets called bursts. The advantage of this approach is that the edge part is simplified in its functionality, but the core nodes are more complex to control, and the overall performance is affected by a highly sporadic traffic profile.

Figure 3.13   Schematic of a core router in the concept of a packet network.

3.5.1.3  Long-term solution: All-optical network

In the previous scenario, we still needed a lot of costly O/E conversion. One key question is as follows: can we efficiently reduce the number of O/E conversion stages in the core routers?

For this, we need to solve three key problems:

• The synchronization (to realign the packets before the switching)
• The regeneration (to enable the cascade of several optical core nodes).
• The contention resolution (to be able to offer the required packet loss rate and delay with respect to the Class of Services requirements)

3.5.2.1  Advanced technology

To realize compact systems, there is first a need for an integrated technology.

To realize the key sub-blocks presented previously, we need the following:

• For the fast optical switching matrix: compact tunable lasers, SOA gate arrays, and integrated wavelength selectors
• For the optical regenerators: integrated Mach–Zehnder, self-pulsating lasers, etc.
• For optical synchronization: optical gates

3.5.3.1  Optical matrix

In the case of opto-electronic interfaces, the fast optical switching matrix (introduction scenario for the short and medium terms), the constraints are quite relaxed because the sensitivity of the receivers allows the design of switches with small output powers. Typically, reception powers as low as −10 dBm can be considered at the output of the switching fabrics. This also means that the amplification is limited in the core of the switch, leading to very compact and less power-consuming architectures.

One typical matrix is the SOA-based matrix, requiring simply an amplification stage before the splitting stage is the broadcast-and-select architecture. Another one is using tunable lasers and a wavelength router in the center. Both are represented in Figure 3.15.

The first architecture (Figure 3.15a) takes advantage of broadcasting functionalities and exploits robust devices such as ILMS or SOAs. However, it is limited in capacity, mainly due to large losses that have to be compensated by amplification. The optical signal to noise ratio (OSNR) affected mainly limits the capacity.

The second architecture (Figure 3.15b) has a priori a larger potential in terms of capacity because the architecture simply includes a tunable source and a passive wavelength router. However, this architecture is not adapted to the broadcast of the packets, and the fast tunable laser is probably the most challenging switching element.

In the perspective of the long-term scenario, with optical interfaces, the constraint comes from the output power that must be high enough to be compatible with optical interfaces. In addition, the polarization is responsible for problems in the optical regenerative structures, it is then fundamental to transform a switched packet stream into a packet stream in a transmission-like configuration. This is the reason why an optical conversion is mandatory in the switching matrix.

Figure 3.15   (a) Optical matrix-based on SOAs. (b) Optical matrix-based on tunable lasers.

Figure 3.16   SOA-based matrix compatible for optical interfaces.

Figure 3.16 shows an optical matrix based on a SOA technology but including a new element: the wavelength selection/conversion stage, as it is studied in the frame of the 1ST DAVID project.

Figure 3.17   A 32 SOA gate-array module.

3.5.3.2  SOA technology feasibility

The SOAs have been used in different system applications for amplification but also for wavelength conversions or for optical gating. To realize large systems as described previously, there is a need for a large amount of components. In this case, the integration is then mandatory to make such a matrix very compact. OPTO+ has designed and realized 32 SOA gate array modules. The module shown in Figure 3.17 includes 32 SOAs and their respective drivers. It has been used to realize a 640 Gbit s⁻¹ switching matrix.

3.5.3.3  Tunable source feasibility

The tunable source is a key component for many system applications.

In the case of slow switching, we can identify the tunable wavelength conversion to provide the required flexibility to achieve a best utilization of the wavelengths in a network. Another evident application is the replacement of ILMs with tunable sources. The advantage is mainly in the spare cost: instead of duplicating all the sources, the objective is to have only one source capable of emitting at any wavelength of the comb exploited in the WDM system. Finally, another application is the monitoring of optical switching systems. In this particular case, we need a compact structure capable of testing the different wavelengths and paths of a switching system. To be compatible with the system constraints, the requirements are switching times in the range of milliseconds or more (for monitoring or for sources), large tunability, high output power, and good extinction ratio.

In the case of fast switching, the main applications are for the metro and the backbone. The tunability is fundamental in providing the required flexibility to exploit the WDM dimension in optical packet switching network concepts. The requirements are fast switching time in the range of a few nanoseconds, small tunability (four or eight channels), high output power, good extinction ratio, and high ON/OFF to guarantee no impact of the cross talk on the signal quality.

For slow structures, a DBR laser has been tested by different laboratories, and feasibility is not an issue.

For fast structures, the main problem is the stability of the wavelength. DBR can be considered if the tunability is small. These components have been demonstrated to be feasible, with switching times in the range of a few tens of nanoseconds. Another alternative is the selective source. Based on the cascade of a laser array, a SOA gate array, a phasar, and an integrated modulator, this structure has been demonstrated to be feasible.

3.5.3.4  Optical synchronization

The optical synchronization is probably the most challenging function. The objective is to process the signal, if possible, in the WDM regime or at the wavelength level. The second important point is probably the lack of digital memory that forces designers to think differently. In that context, the synchronization cannot be done at the bit level. We assume that the synchronization can be efficient in a resolution of a few nanoseconds. When we have said that the other point is to identify the source of desynchronization with respect to a reference clock.

The first source of loss of synchronization is the thermal effect in the transmission fiber. With a value between 40 and 200 ps km⁻¹, depending on the mechanical protection scheme adopted for the fiber, thermal effects can dramatically affect the phase of the packet streams. The WDM dimension can be advantageously exploited to make the synchronization stage compact and low cost.

The second source is the loss of synchronization in switching fabrics due to a nonideal path equalization. This occurs at the packet level, imposing a synchronization at the wavelength level.

The structure adopted is shown in Figure 3.18.

3.5.3.5  Optical regeneration

The optical regeneration is one of the fundamental functions to make the approach realistic. To build all-optical networks while having optical switches capable of handling terabits/second throughputs, the optical regeneration is then mandatory at the periphery of switching architectures. The main functions are the total reshaping of the pulses in the amplitude and in the time domains. To achieve this reshaping, several techniques can be adopted. We will retain one, particularly adapted to the characteristic of switching fabrics creating strong impairments between pulses or between the groups of pulses. The technique adopted is a total reshuffle of the pulses adopting nonlinear elements such as Mach–Zehnder or Michelson structures.

Figure 3.18   Optical synchronizer as proposed in the frame of the 1ST DAVID project.

The main distortions identified are the following: bits affected at the periphery of packets due to the switching regime, nonlinear effects such as cross-gain modulation and four-wave mixing, cross talk (in-band and out-of-band), patterning effects by crossing active devices, noise accumulation, and jitter accumulation.

To overcome these effects, a structure has been proposed in the frame of the DAVID project. This structure, presented in Figure 3.19, has the following characteristics: by using a cascade of two nonlinear elements, the convoluted function creates a much more nonlinear transfer function, thus limiting the noise transferred in the first cascade. This has an important impact on the OSNR specification, which can be close to the back-to-back value (before the first stage) even in the case of a large number of cascades.

Figure 3.19   Structure of the Re-amplification–Reshaping–Retiming (3R) regenerator as it is studied in the frame of the DAVID project.

To really reshuffle the pulses, different techniques will be adopted. We can retain an amplitude and a phase modulation creating an amplitude modulation in interferometric structures to really enhance the extinction ratio and remove the noise. The second technique adopted is a sampling technique of each pulse with a clock to remove the jitter. The wavelength conversion technique will then be preferred to reallocate the wavelength in the correct wavelength comb of the new system.

3.5.3.6  Feasibility of network concepts

The feasibility of the approach was demonstrated for the first time in 1998, at the end of the ACTS KEOPS project.

In this project, we have cascaded 40 network sessions error free at 10 Gbit s⁻¹ per wavelength demonstrating for the first time the possibility of building an all-optical network at a backbone scale.

Figure 3.20 gives the network session tested and put in a loop to demonstrate the concept.

3.5.4.1  Environment and specificity of the backbone

If, in the metro, the capacity is limited, in the backbone this is the major characteristic. To provide the capacity with a technology limited today to 10 Gbit s⁻¹, the only solution is in the exploitation of the WDM dimension.

Figure 3.20   (a) Network session put in a loop to test the feasibility of an all-optical network. (b) Photo of the demonstrator. (c) BER curves giving the physical performance.

Therefore, the WDM dimension will be fully exploited to provide the required capacity but also to avoid collisions due to the natural statistical multiplexing of packets on the wavelengths.

The second particularity is the aggregation. Depending on the traffic profile, circuit switching or packet switching will be preferred.

3.5.4.2  Circuit switching techniques for an immediate introduction

Circuit switching techniques can be envisaged in the first scenario as a transport layer to provide the capacity of transport.

To be compatible with a DATA traffic, the coupling of a cross-connect with a packet router is, even today, the more pragmatic approach. This is a subject of strong interest for products that are used at present.

However, this solution is not really cost-effective because only two alternatives can be adopted:

• All the wavelengths are connected to the packet router, and in this case, the number of TX/RX dramatically increases the cost per port.
• Only a part of the wavelength is connected to the packet router, and in this case, the traffic matrix needs to be very stable. The deterministic approach for the number of connections becomes nonrealistic when the traffic profile becomes highly statistical.

3.5.4.3  Optical packet switching techniques to cope with a traffic profile evolution

In this case, what could be the benefit of the concept proposed?

• The packetization at the edge level with an optimized size can reduce the latency in the creation of the packets. A second technique to accelerate the filling ratio is in the upgrade of the best effort in a premium class of service. Therefore, the advantage of this solution is that latency can be controlled to reduce the latency in the rest of the network. Calculations show that there is a global benefit in terms of end-to-end latency.
• The exploitation of the WDM dimension once again reduces the latency. The packets cross the architecture, and they see only a transmission path, even if it is switched. No buffers are crossed so the resultant latency is minimum.

Figure 3.21 shows a table summarizing the performance in terms of packet loss rate established in the frame of the Reseau Optique Multiservice (ROM) project. It appears that on the three class of services considered, the end-to-end performance can be obtained. From dimensioning issues, it appears that for the WAN, a sum of 30% for Class of Service 1 (CoS 1) and CoS 2, and a best effort (BE) lower than 80% is tolerated.

Figure 3.21   Performance of the all-optical packet switching network concept.

This demonstrates the viability of an all-optical concept and as a consequence the viability of the opto-electronic scenario.

3.6.2.1  Metro part

If the average load of a wavelength is high enough, due to an efficient aggregation process, then circuit switching is probably viable. However, if the load is low, below 20%, even if the cost of switches are more expensive, the gain in statistical multiplexing creates a real opportunity for packet techniques making them less expensive than circuit switching techniques.

The main reason for this gain is probably the high cost of the wavelength due to expensive infrastructure costs, pushing all telecommunications companies to prefer an increase of the bit rate rather than an exploitation of the WDM dimension.

Therefore, the tendency is probably packet techniques to decorrelate the bit rate from the granularity of switching and high-bit rates to adopt the most cheap technology while providing the required capacity.

Figure 3.22 shows the areas where optical packet switching is better than circuit switching.

From the figure, the numbers 2, 4, 6, and 8 indicate the ratio in terms of cost per port (wavelength) between an optical packet switch and a cross-connect targeting the same size (256 × 256) and the same technology.

The load of a wavelength is the average load.

Figure 3.22   Need to exploit packet techniques in the near future.

The ratio on the horizontal axis is a ratio between the wavelength transmission cost (including the installation costs) and the cross-connect port.

For example, if the ratio between the cost of a wavelength in the transmission system and the cost of a cross-connect port is equal to 1 (red bar), optical packet switching techniques are interesting:

• 2: always, whatever the load is
• 4: if the average load of a wavelength in circuit switching is lower than 33%
• 6: if the average load of a wavelength in circuit switching is lower than 20%
• 8: if the average load of a wavelength in circuit switching is lower than 13%

Therefore, the tendency is the following: If the cost of a wavelength in a transmission system is high (the case of the metro where the installation cost is not negligible), or if the aggregation is not efficient enough forcing an average load very low (this is a serious tendency with the increase of the application bit rate and the sporadicity of the traffic profile), packet switching techniques always exhibit a better performance than circuit switching.

3.6.2.2  Backbone part

As an example, we have computed, for two levels of aggregation, the load of a network with respect to the distance of the network (for the access to the backbone). It appears that in major cases, if the aggregation process is not enough, even in the backbone, the packet switching technique is the cost-effective solution.

Figure 3.23 shows the importance of packet switching techniques, also for the backbone. This is one of several curves that could be drawn. However, it once again shows a tendency.

The grooming or the aggregation efficiency depends on the traffic profile in large part. Therefore, we plot two indicative curves:

• One curve exhibits an efficient aggregation (a realistic case is when the constant bit rate (CBR) is higher than VBR or when the number of connection points is high to facilitate the grooming process).
• The other curve exhibits a less efficient aggregation (a realistic case is when the variable bit rate (VBR) becomes dominant). In that case, we cannot have a stable traffic matrix, and we are addressing a sporadic traffic profile (a realistic case if we have a bit rate convergence from the access to the backbone).

Figure 3.23   Curve showing strong interest to introduce packet switching techniques even in the backbone.

The vertical axis indicates that the required average load of a wavelength is cost-effective. The horizontal axis indicates the average distance of a transmission system with respect to an average network session representative of the network considered. Therefore, the WAN starts for transmissions higher than 100 km. The calculations show the importance of the time multiplexing. If the distance is long, there are a large number of clients sharing the same network infrastructure. Therefore, the cost per client is reduced. In addition, the cost of the installation of a wavelength is considered lower than that of the metro. The reason is that in the WAN, natural infrastructures are exploited to reduce the installation costs (such as highways or railways). If we are under the curve in bold, there is an advantage of introducing packet switching techniques. The grooming tendency gives values for the required load in circuit switching.

For example, in the case of a good grooming efficiency, if the average distance of propagation of a representative network session is lower than 300 km, you will have a cost gain by exploiting packet techniques. In the case of a low grooming efficiency, packet switching techniques are always more efficient.