Network operator perspectives on optical networks Evolution towards ASON

Transcription

1 Network operator perspectives on optical networks Evolution towards ASON Beatriz Farias Craignou, France Telecom R&D, Issy-les-Moulineaux, France, Roberto Clemente, Telecom Italia Lab, Torino, Italy, Jacques Robadey, Swisscom Corporate Technology, Berne, Switzerland, Laszlo Jereb, Matav Telecom, Budapest, Hungary, Zacharias Ioannidis, OTE, Athens, Greece, Jose Santos, Portugal Telecom Inovaçao, Aveiro, Portugal. Abstract This paper 1 addresses the evolution of Optical Transport Networks considering the possibilities offered by the Automatically Switched Optical Network to network operators. The architectural model and main features of an ASON are presented in some detail and compared with those of a traditional OTN. Transport service connection offerings are listed. The architectural models for the transport of IP traffic over optical networks are briefly discussed. Network operator perspectives based on ASON are evoked. Findings concerning the advantages and drawbacks of ASON compared to an OTN are derived from a number of case studies carried out using first fictive and small test networks and validating the results on networks of real size. The comparison between ASON and OTN is made for static and dynamic traffic cases, commenting on routing aspects, recovery, restoration and protection issues. Numerical results illustrate and complement the presentation. Key words: ASON, Optical Transport Networks 1 This paper is based on the results achieved in the EURESCOM Project FASHION (P112), involving all participants to the Project, whose work is acknowledged, even if they are not formally mentioned. The authors thank, in particular, the contributions of Hisao Nakajima and Serge Bibas (France Telecom R&D), Marco Vitale and Fausto Saluta (Telecom Italia Lab), Daniel Rodellar (Swisscom Corporate Technology), Tivadar Jakab (Matav Telecom) and Catarina Monica (Portugal Telecom Inovaçao). This paper does not necesarily reflects a common technical position of EURESCOM shareholders. The authors gratefully acknowledge the EURESCOM support to accomplish this work.

2 1 Introduction In the last years, transport networks have benefited of the great advances in optics and packet technology. Using several optical technologies, WDM 2 systems allow a very large increase of transport capacity, since a single fibre can carry a number of optical channels (OCh), currently 8, 32,, 16, or more. Optical Transport Networks (OTN), enriched with functionality improving traffic routing and network protection, become more flexible and reliable, with increased control possibilities. Operators are then able to cope with huge capacity demands at access and core levels. They can also offer enhanced transport services through an adequate utilisation of new transmission facilities and the potential capacity of optical fibre networks. The present situation is improving with the development of intelligence in the optical network, that makes easier the instantiation and routing of survivable optical connections in real time. All these possibilities enrich service offerings, including multi-service flows and Internet advanced services. A promising candidate for this kind of transport is the Automatically Switched Optical Network (ASON), proposed by ITU-T Recommendations G.87 [1] and G.88 [2]. Based on the earliest ASON Recommendations, the EURESCOM Project FASHION (P112) studied the implementation of ASON, its applicability and migration issues. Some of the obtained results are the subject of this paper, where a brief description of the architectural models is given and, to enlighten operator s choices, a comparison between ASON and OTN for static and dynamic traffic is discussed. 2 Architectural models of optical networks 2.1 The ASON model An ASON is an optical network based on WDM technology and able of carrying switched optical channels. A switched optical channel is an OCh automatically set-up or torn-down using signalling. 2 Wavelength Division Multiplexing systems can transmit several signals at different wavelengths on a single optical fibre. Each signal is sent in the shape of light pulses at a specific wavelength. Signals are mixed and separated by optical multiplexing and demultiplexing devices acting like a prism that separates colours from white light. A single optical amplifier can simultaneously amplify all the signals of a multiplexer. Optical add/drop multiplexers insert or extract channels without separating the channels present in the fibre. The systems are located at the connection ends. The logical architecture of an ASON is presented in Figure 1. It is comprised of three planes: a Transport Plane (provided with Optical Cross-Connects and optical fibre links), a Control Plane (provided with Optical Connection Controllers and signalling links) and a Management Plane. The Transport Plane carries user information flows, the payload clients want to exchange, from one location to another. It can also provide transfer of some control and network management information. The Control Plane performs the call control and connection control functions. Through signalling, the Control Plane sets up and releases optical connections and may restore a connection in case of failure. It transports and distributes information on neighbour discovery, network topology and resource discovery, which is basic for routing algorithms. The Control Plane and the Management Plane supervise the network behaviour, determining routing of new optical connections, configuring the devices realising each connection, monitoring the state of the connections, collecting and taking care of alarms notifying fault conditions of network elements. CCI EM/NM NMI-A NMI-T NNI OCC UNI UNI CCI Client device IP router ATM switch OCC Control Plane OCC NNI OCC Transport Plane NMI-A Management Plane NMI-T EM / NM Connection Control Interface Element Management / Network Management Network Management Interface for the ASON control plane Network Management Interface for the Transport Plane Network to Network Interface Optical Connection Controller Optical Cross-Connect Physical Interface User to Network Interface Figure 1. Logical architecture of an ASON The Control Plane supports connection set-up and teardown as a result of a user request (switched) and a management request (soft-permanent). It also carries and distributes link state information (like fault and signal quality), which is detected by the Transport Plane. Through the User Network Interface (UNI), clients can directly request the opening of connections in the transport plane, using the direct link between ASON clients and Control Plane. The function of the UNI is to pass signalling messages to the network control plane entity. In case of provisioning, connection set up is the responsibility of the network operator, whilst in the signalled case, connec-

3 tion set-up may also be the responsibility of the end user. Information flows through the UNI support basically call control, connection control and selection, as well as resource discovery. The ASON Control Plane is sufficiently generic to support different technologies, different business needs and different distribution of functions (or components) by equipment suppliers. Therefore, ASON is client independent and several types of clients (IP, ATM, SDH and GbE) can use its functionality. As said, both the Control Plane and the Management Plane can implement functionality for network supervision and control. ITU-T Recommendation G.88 defines the ASON architecture, stating that the ASON control plane facilitates fast and efficient configuration of connections within a transport layer network to support both switched and soft permanent connections, reconfigures or modifies connections that support calls that have previously been set-up, and performs restoration functions. The different components of the ASON control plane provide specific functions including routing and signalling. Note that the G.88 draft provides a functional description of the ASON control plane but does not precise the implementations of this functionality. So, for example, the routing component of an ASON Control Plane can be simply a table in which the routing map is stored (being reckoned by the Network Management System (NMS), for instance), or a clever component that implement a distributed routing algorithm. That means that the network intelligence could be, to a certain extend, centralised or distributed. Nevertheless, the common interpretation used by standardisation bodies and in most scientific environments refers to the ASON with a distributed intelligence, which is the interpretation of an ASON used in this paper. Within Recommendation G.88, the ASON Control Plane supports both SDH transport networks (G.83) and OTN transport networks (G.872). Instead, Recommendation G.87 specifies the network level requirements for the control plane of ASON based transport networks (ASTN). This work, based on the earliest ITU-T Recommendations of ASON, deals with the optical (WDM) layer only. 2.2 The OTN model In order to fix the ideas and compare ASON with an OTN, the reference architectural model of a traditional OTN is illustrated in Figure 2. Its features are similar to those of ASON, except that the OTN does not have a separate control plane and OChs cannot be automatically switched. In an OTN, both control and intelligence are centralised and OChs are provisioned via the management plane through configuration of each that is involved in the routing. The OTN layer offers a transport service of static provisioned bandwidth. Client device IP router ATM switch EM/NM NMI-T Transport Plane Management Plane NMI-T EM / NM Element Management / Network Management Network Management Interface for the Transport Plane Optical Cross-Connect Physical Interface Figure 2. Logical architecture of an OTN 2.3 ASON service offerings and applications The control of connectivity is essential to the operation of a transport network. Connections can be established by provisioning, by signalling or by a combination of both methods. ASON transport service offerings include - (Hard) Permanent OCh connections (end-toend, similar to enhanced leased lines), configured by means of a management system either manually. - Automatically OCh switched connections, established on demand by signalling. They are of flexible provisioning, activated or torndown by the customer itself, via the UNI signalling interface, and completed by means of E-NNI or I-NNI signalling. - Soft-permanent OCh connections (involving Management + NNI signalling), established via network generated signalling and routing protocols. The transport network provides a permanent connection at the edge and utilises a switched connection to provide end-to-end connection between the permanent circuits at network edges. The establishment of such connections is dependent upon the definition of an NNI. There is no defined UNI. From the perspective of the end points, a soft permanent circuit appears no different than a provisioned, management controlled, permanent circuit. Figures 3 and 4 illustrate the set-up of a switched connection and a soft-permanent connection, respectively.

4 Connection request UNI Control Plane Set-up request Transport Plane Switched connection Figure 3. Example of a switched connection set-up Connection request Permanent connection Management Plane Control Plane Switched connection Soft-permanent connection Set-up request Transport Plane Permanent connection Figure 4. A soft-permanent connection set-up, as a hybrid case of the switched and permanent types Sets of these connections allow Optical Virtual Private Network and Lambda Trunking offerings. - For O-VPN the client has the direct control and visibility of the resources dedicated to this service. Customers (a closed user group) are free to reconfigure s and use the resources as switched channels. The O-VPN is an optical layer service on the top of the switched connection service for some applications with high privacy requirements. - For Lambda Trunking, optical channels can have particular routing and performance constraints. It is an offer of a bundle of permanent or soft-permanent channels between two end points, with the same transmission performances, routed along the same path and possibly on the same fibre pair. All these offerings are under Service Level Agreement (SLA), which is the contract between a Network Operator and a Customer defining the global responsibilities between them. The SLA also contains the service level specifications, like availability, grade of service, service guarantee and potential penalties in case of service degradation. Several applications of these connections interest operators and ISPs. ASON applications concern mainly meshed networks, and, considering the granularity of connections, the core networks. Currently, this granularity is supposed to be 2.5 Gbit/s or 1 Gbit/s. Edge nodes can house IP routers or ATM switches, serving clients with traffic demands of at least 2.5 Gbit/s. Smaller demands could be multiplexed at edge nodes, to be more efficiently carried over big capacity OChs across the core network. Operators have to consider the issues of grooming, leading to resource optimisation and significant cost savings. 2.4 Architectural models for IP traffic transport Admitting that IP traffic will be dominant in future networks, transport infrastructures should logically be optimised mainly for the transport of this kind of traffic. Basically, the OTN layer can offer a transport service of fixed bandwidth, statically provisioned. To ease the provisioning of more dynamic services, like bandwidth on demand and O-VPN, two main layered models were initially proposed for the Control Plane (by the Internet Engineering Task Force and the Optical Interworking Forum), which differ on how and how much information, signalling and routing are exchanged between routers and s as well as on the disclosure of this information to the customers. They are the overlay and the peer models. In the overlay model (similar to a client/server model), the client and optical networks are completely separate, with independent control planes, with no exchange of routing or topology information. GMPLS (Generalised Multi-Protocol Label Switching) signalling is used to request circuits to the optical layer via the UNI interface. This model, already implemented by some equipment suppliers, is applicable to different administrative domains. In the peer model (similar to the unified model), regarding the control plane, IP and optical networks are a single network, with routers and s acting as peers. There is a total exchange of information (i.e. internal topology) between routers and s. Having the same routing instance for IP and optical networks, IP clients have a complete knowledge of the optical topology. There is no longer distinction between UNI and NNI, and there is only one link state topology database. GMPLS is used to establish end-to-end Label Switched Paths (LSPs). This model is applicable to a single administrative domain. It is intended for a rather future implementation.

5 As a compromise between the overlay and the peer models there is the augmented model, which is being constructed implementing as far as possible advantageous features of the other models and avoiding their drawbacks. For instance, there is automatic discovery of reachable destinations, optical channels can be triggered by IP LSP, and so on. This model supports the common scenario in which the client and the optical networks are administered by different entities and represents an intermediate option of practical value for service providers. The IETF standardisation work on this model is in progress. ASON is based on the overlay model. The evolution path for IP traffic transport over optical networks seems to be initially deployed on the overlay model. In a later phase, the peer model can reach enough maturity and more sophisticated routing interaction between IP and optical networks could be implemented by equipment manufacturers. But most probably, the augmented model benefiting of the introduction of reachability information exchange between the two domains will be preferred by operators. The administrative domain organisation, business models and service offerings are significant drivers of this evolution. 3 ASON perspectives for operators Network operators seek new network solutions allowing either the matching of service requirements or network functionality provisioning in a cost-effective way. ASON allows operators offering to customers or client networks optical connectivity on demand, in a fast and resilient way. It also contributes to improve fundamental network functionality, like provisioning, resilience and differentiation of connection attributes. These features leading to a better resource utilisation and enhancement of service offerings should result in a satisfactory business plan. Although the ability of offering switched OCh connections seems to be the most attractive (and natural) feature of an ASON, its perspectives appear still fuzzy in a near future. In a short-term scenario, ASON can be foreseen for the transport of permanent OCh connections, as an alternative to OTN, as well as a response to the moderate present requirements of bandwidth on demand, while a real exploitation of the switching functionality to transport dynamic optical traffic will happen further in the future. An ASON offers new recovery options, facilitates connection attribute differentiation and the integration of different levels of quality of service and classes of service, features operators appreciate very much. 4 Comparison of ASON and OTN Within the EURESCOM Project FASHION, case studies were carried out to derive potential pros and cons of an ASON compared to an OTN. To tackle problems, fictive test networks with few nodes and relatively small traffic were analysed, for different routing schemes and resilience strategies. The options were evaluated both for networks realised using opaque nodes and when transparent 3 nodes were implemented. The same exercise was performed for real size networks, provided by members of the Project. In this latter case, the findings were not so conclusive. Case study specifications lead to a better understanding of fundamental questions and to provide answers based on simple modelling and information. In order to handle network modelling and give a ground to simulations and planning evaluations, it was indispensable to choose, among several options, some routing algorithms, traffic models, recovery schemes, reconfiguration possibilities, and reduce quite a lot the scope of case studies. The validity of results for real size networks was verified. 4.1 ASON vs. OTN with permanent (static) traffic A traditional OTN and an ASON were compared assuming that the studied networks have the same topology (nodes, links) and transport the same permanent optical connections. A test network with seven nodes and three demand matrices were considered. Nodes were assumed to be opaque and 4 the maximum number of optical channels per fibre Routing aspects Transporting the same demands, over the same network, a priori only different routing algorithms can justify a difference of resource requirements between an OTN and an ASON. Routing algorithms based on link state protocols like OSPF (Open Shortest Path First) or IS-IS (Intermediate System to Intermediate System) employ a distributed database model. These protocols allow traffic engineering and converge more rapidly than distance vector protocols (for instance, the Routing Information Protocol). Link state protocols are also faster at calculating and updating routing tables. To carry out the comparison, it can be assumed that simple hop-based and/or length-based routing algo- 3 Optical transparency means that the optical signal carried over the OCh is not subject to Optical/Electronic conversion in any of the traversed network elements. Opaque s use O/E converters, without changing the signal bit rate or protocol; they can be of OEO type (optical incoming and outgoing ports, electronic switching matrix). A transparent is of type OOO, all optical, no use of converters.

6 rithms (like OSPF) are suitable for ASON. In fact, several manufactures have proprietary routing algorithms implemented in s. Some of these algorithms are variants of OSPF. First generation ASONs are supposed to follow a simple shortest path routing algorithm, like OSPF, or better Dijkstra s, minimising the number of switches crossed by the path (hops), or the length of the path. A simple routing simplifies signalling between OCCs, however more complex and better performing routing algorithms can be used, keeping the control plane complexity at a low level (cf. [6]). OSPF is also widely adopted in IP networks and is a fundamental algorithm for MPLS, where more sophisticated constraintbased routing schemes can also be implemented. As an OTN has a centralised view of the optical network, sophisticated routing algorithms can be used, trying, for instance, to minimise the overall network deployment cost (carrying a maximum load over the available resources, adapting routing, and so on). However, optimal resource evaluations are valid during a certain time with given demands and resources. But, when demands or network resources evolve, reconfigurations are necessary in order to keep optimal the network load, requiring at least partial rerouting of the already set-up optical connections. Since reconfigurations are seldom efficiently managed, OTNs often adopt non-adaptive routing, like Dijkstra's. In fact, no matter if the network control is centralised or distributed, an adaptive routing algorithm applied to a static traffic optimises the resource utilisation in a fixed point in time, not along the network evolution. Then, for static traffic and plain routing algorithms, no real difference of performance or resource requirements can be expected between an OTN and a first generation ASON. As soon as ASON will reach a good maturity, more sophisticated routing algorithms could allow improving performances in terms of traffic load for given network resources Connection reconfigurability For static traffic, a significant ASON advantage is the ability to efficiently perform OCh reconfiguration. This pro is related to the fast rerouting feature (fast enough to avoid service disruption). Connection reconfiguration, in conjunction with adaptive routing, allows maximising the load of an existing meshed network along its evolution. In an ASON, this can be done in an automatic and distributed way and, most important, with a continuous and asynchronous process at a low frequency (e.g. weekly, monthly, connection by connection) and without service disruption (thanks to the fast rerouting) Recovery and protection issues In case of failure, when applying a recovery scheme to an ASON or an OTN, different resource requirements could be expected, even selecting the same routing algorithm. ITU-T Recommendation G.85 [5] describes the enhancement of techniques for transport network availability. Protection, the replacement of a failed resource with a pre-assigned standby, and restoration, the replacement of a failed resource by rerouting using spare capacity, are used to classify these techniques. In general, protection actions complete in tenths of millisecond range, while restoration actions normally complete in times ranging from hundreds of milliseconds to up to a few minutes. Although both preventive (protection) and reactive (restoration) rerouting mechanisms can be applied to both OTN and ASON, automatic protection is better suited to OTNs, while restoration appears more appropriate for ASONs. The reason is that a centralisedcontrolled restoration mechanism (i.e. a restoration managed by the NMS) reveals itself rather slow in the transmission network (e.g. SDH), with typical Timeto-Restore (TtR) of the order of tenths of seconds up to some minutes. In case such a slow TtR is not acceptable, restoration cannot be applied, and a protection mechanism should be chosen. The TtR of protection mechanisms is of tenths of milliseconds (5 ms of switching + a few ms to detect the failure and trigger the protection switch). Unfortunately, protection schemes generally lead to less efficient resource utilisation. The distributed control allows ASON implementing distributed, dynamic and fast restoration schemes, whose target TtR is short enough to avoid service disruption. It is a matter of discussion whether a TtR of 5 ms is the real threshold to guarantee no impact of a network failure on the carried applications. Several experts think that a TtR equal to 2 ms could be a good enough performance: so, a TtR of 2 ms becomes the goal for distributed restoration in an ASON. Six types of survivability mechanisms were analysed: dedicated protection, - n:m shared protection with different levels of priority, - Fast restoration with alternative routes previously calculated, - Restoration with alternative routes calculated after fault detection, - No protection (the channel is simple ejected in case of failure), - Best effort (non protected and carried on protection resources with pre-emption of the channel to restore a traffic of higher priority). Protection and restoration mechanisms, used by the network, can correspond to different Classes of Service. They may complement the differentiation of connection attributes related to a CoS.

7 The ASON control plane allows operators offering to users calls with selectable CoS, depending on the selection of the survivability mechanism (protection, restoration or none) for a particular connection supporting a call, the policy of the network operator, the topology of the network and the capability of deployed equipment Numerical results Providing network reliability against single failures, an ASON allows significant resources savings compared with an OTN, thanks to the efficient spare resource utilisation of restoration mechanisms. This result is clearly illustrated in figures 5 and 6 below OSPF +164% +94% OSPF-based path restoration +36% OSPF-based 1+1 protection Figure 5. Comparing overall wavelength per link required by path restoration and 1+1 protection for a 7-node network and a uniform traffic demand OSPF +158 % +13 % +27% OSPF-based path OSPF-based 1+1 restoration protection Figure 6. Comparing overall wavelength per link required by path restoration and 1+1 protection for a 7-node network and a dual hubbed traffic demand For a fictive meshed network of 7 nodes, considering a uniform traffic demand matrix of soft-permanent OCh connections and a dual hubbed traffic demand matrix, if 1+1 protection requires about 2.6 times the resources needed for routing working paths according OSPF 4, this increasing is reduced to about 2. adopting path restoration. Furthermore, it is interesting to notice in Figure 7 that the different TtR of protection and restoration does practically not affect the availability figures of the OChs in the fictive 7-node meshed network considered above, on condition that restoration is faster than a few seconds. The graph shows the iso-availability curves of the considered network, when the MTBF and TtR can change and the MTTR 5 is 8 hours. Considering that the life cycle of an could be of 8 years, it can be seen that, with a typical MTBF of 4 years for each piece of equipment, a TtR shorter than a few seconds does not change the availability figures. This behaviour means that a multiple failure provides a higher contribution to the unavailability than recovery time after a single failure. However, as already said, while a TtR as long as a few seconds does not affect availability, it could have an impact on carried applications, so that a target value for restoration TtR should be in the order of 2 milliseconds. equipment MTBF 4 years iso-availability curves OTN (1+1 protection) ASON (path restoration) increasing availability MTTR = 8h 1ms 1ms 1s 1s 1s TtR (log scale) Figure 7. Iso-availability curves for the 7-node network 4.2 ASON vs. OTN with the dynamic traffic case A few network design case studies were performed to identify the advantages of the ASON dynamic provisioning, for network cost saving purposes in terms of planned resources. Their main objective was the evaluation of the statistical gain achieved with respect to the permanent allocation (or full resource allocation) when adopting a statistical resource allocation strategy (using statistical traffic models). 4 It should be noted that the OSPF routing algorithm used in this work selects one and only one route between two nodes even if more than one route does exist. A variant of OSPF selects the shortest node disjoint routes between the nodes and routes the demands according to load balance or similar criteria. 5 MTBF Mean Time Between Failures MTTR Mean Time To Restore/Repair

8 Intending a fair comparison between permanent and statistical planning, input traffic matrices were interpreted as an indication of the number of users or potential connections. Specifically, each entry of the matrix T ij indicates that T ij users connected to node i may communicate with users connected with node j using one mono-directional OCh each. Users are not continuously transmitting data, being active only for a certain percentage of the time (called user activity) according to an on/off process. In an OTN, there is no choice: one connection must be provisioned per user. In an ASON, users require connections when they are ready to use them, and a statistical resource allocation strategy can be adopted allowing a certain connection blocking probability. The selection of a traffic model for the statistical planning of an ASON is a cumbersome matter, because it is difficult to foresee the behaviour of the ASON customers. Three classical traffic models, applicable to connection oriented switched traffic, have been considered in our evaluations, namely the Poisson model (infinite number of statistically independent users), the Engset model (finite number of statistically independent users) and the Fredericks model (infinite number of statistically dependent users). The following results refer to the Engset model. It must be noted that, while results obtained using the Poisson and the Engset models are not so far from each other, as soon as the number of users is high, the Fredericks model reduces the statistical gain and, also, the higher is the dependency among users the higher is this reduction. Two sets of results concerning the same case study are presented here. The first set aims at evaluating the statistical gain achieved for a given user activity as a function of the routing/protection strategy, the second set aims at enlightening the dependence of the statistical gain from the user activity. Figure 8 compares permanent and statistical planning with user activity equal to.5 (5% of the time in the on state and 5% in the off state), considering two routing strategies without recovery (OSPF and load balancing over the two shortest paths) and three recovery schemes (path restoration, link restoration and 1+1 protection, all of them based on OSPF routing). It shows a statistical gain of about 1% to 15% without recovery and higher savings with recovery. Figure 9 provides results in function of user activity, for a selected set of routing and protection/restoration strategies and shows that a statistical gain is almost always possible as long as the user activity is substantially low (let us say less than.6). With user activity greater than 6%, the obtained gain in capacity savings is less than 1% and rapidly becomes negligible. Total number of per link Engset model Permanent OSPF Load Balancing Figure 8. Total number of allocated wavelengths with the Engset model, user activity=.5 and permanent allocation, for different routing/protection strategies relative savings Figure 9. Relative amount of wavelength savings in respect to the full allocation case, as a function of user activity, with OSPF, path restoration and 1+1 protection Both Figure 8 and Figure 9 show higher statistical savings when recovery schemes are applied. The reason for that is that a recovery scheme requires additional resources on network links, to be used by dynamic traffic in case of failure. This can be viewed as an additional demand offered to the network: the higher the dynamic demand, the higher the statistical gain. That also implies that, given a certain user activity, the higher the number of customers, the higher the ASON advantages compared to an OTN. 4.3 Real size ASON Erlang/user Path restoration Link restoration 1+1 APS OSPF Path rest. 1+1 APS permanent To verify the validity of results in real size networks, ASON architectural options were examined on two test networks of 18 nodes, one with a squared shape, another forming a long rectangular shaped network. Path selection was made on a shortest path, in kilometre length or in number of hops. Resilience options were dedicated 1+1 protection, link restoration and path restoration. The assumptions that OTN and ASON carry the same permanent OCh and have the same topology (nodes, links) were still adopted.

9 Most of the results obtained with the fictive 7-node network have been confirmed with these realistic networks as well (and are not reported in this paper). Total number of WLs per link nodes, directional demand OSPF Path Rest. Disj. Link Rest. APS 1+1 Figure 1. Total number of allocated wavelengths for different routing/protection strategies in an 18-node real size network However, performing a planning exercise with a demand pattern in which the optical demands are highly concentrated between a few couples of nodes (this case is referred to as a directional demand pattern), unexpected results were obtained : 1+1 protection requires less resources than link restoration and the savings of path restoration are rather low too (cf. figure 1). The same behaviour has been obtained for both the squared and the rectangular networks. This is due to the traffic distribution: few strong traffic directions concentrate all the working and (especially) spare resources on few links with few chances of sharing them. Let us consider the main OCh demands between the couples of network nodes, in particular all demands greater than 15 OChs. These demands are shown on the left of Figure 11. In the centre and right side of the figure, it is also shown how these OChs are routed using the OSPF protocol. Given these routes, two resilience mechanisms, link restoration and 1+1 protection, are considered. To implement these mechanisms, the needed resources in terms of number of wavelengths per link are shown in the centre of the figure for link restoration, and on the right for 1+1 protection. In this situation, it is clear that the sharing property of the link restoration mechanism is under used: all the spare resources that are computed are not properly utilised. That leads to two additional considerations: First, the restoration performances in terms of number of resources are strongly dependent on the pattern of the considered traffic and on the level of meshness of the network. The statement that restoration always allows significant savings compared with protection can be considered a common misunderstanding. Second, in order to overcome this problem, more sophisticated routing schemes should be used. Following this intuition, a possible approach can be the realisation of a load balancing of the traffic. In this sense, two (or more) directions are found and the traffic is routed not only on the shortest path between the source and the destination, but also in different directions, such that the spare resources can be better used. Km 5Km 1Km Km 5Km 1Km Km 5Km 1Km Resource sharing OChs 17 OSPF1 Link restoration 17 OSPF1 1+1 Figure 11. Required resources for OSPF (centre and right), Link Restoration (centre) and 1+1 Protection (right), considering the main OCh demands (left).

10 5 Conclusions This paper provides an architectural description of an ASON, comparing it with a logical view of an OTN architecture. The main difference resides in the ASON control plane that facilitates fast and efficient configuration of connections within the optical transport layer network, to support both switched and soft permanent connections, reconfigures or modifies connections and performs restoration functions. The ASON intelligence can be totally or partially centralised or distributed. For permanent connections, ASON can provide resilience in a fast and cost-effective way through fast restoration. Although path restoration is always better than 1+1 protection (using the same routing schemes), the actual resource savings of restoration strongly depend on the network meshness (better if high) and on the traffic distribution (better if uniform). A restoration time lower than 2 ms can be considered as a target value for an ASON. In spite of the fact that this value is 4 times higher than the 5 ms of TtR of a 1+1 protection, we have demonstrated that it practically does not affect the OCh availability. An ASON could use adaptive routing also in case of permanent connections, thanks to its connection reconfiguration ability. In fact, an ASON can reconfigure OChs in an automatic and distributed way, within a continuous process at a low frequency, connection by connection, without service disruption. The usage of adaptive routings could improve the resource savings of restoration schemes too. An ASON shows advantages over OTN in respect to the dynamic channel provisioning under demand as well. Some traffic models have been introduced for an ASON and the Engset model has been analysed in the paper. The statistical gain of an ASON is strongly dependent on the customer activity: when the activity factor (expressed in Erlang/user) is higher that.5, the statistical gain is really limited and an ASON does not allow significant resource savings. Network operators seek new network solutions allowing either the matching of service requirements or network functionality provisioning in a costeffective way. ASON offers network operators and service providers improved functionality for fast and reliable provisioning of optical connections and differentiate offerings. Although the ability of offering switched OCh connections seems to be the most attractive and natural feature of an ASON, its perspectives and demands appear still fuzzy in a near future. In the short term, ASON can be foreseen for the transport of permanent OCh connections and to cope with the emerging bandwidth on demand requests, while a real exploitation of the switching functionality to transport dynamic optical traffic will happen further in the future. The new recovery options as well as the possibility of connection attribute differentiation are features of ASON that operators greatly appreciate. Considering migration aspects, the ASON overlay model can be considered as the least complex to implement from the current model. The migration speed will depend on target traffic demands and envisaged customers. equipment availability and appropriate functionality implementation are crucial factors. Issues on transparency are under study. A network provided with transparent nodes will save regenerators and costs, but a number of complex reconfiguration problems have to be solved. Possibly, a hybrid network, with opaque and transparent nodes could be a good solution to cope with the 2.5 Gbit/s (or more, perhaps 1 Gbit/s) connection requests. Concerning the architectural model for IP traffic over optical transport networks, the overlay and augmented models have better chances of rapid implementation by equipment suppliers, opening opportunities for business models, being more satisfactory for service providers and carriers. References [1] ITU-T G.87 draft V1., Requirements for the Automatic Switched Transport Network (ASTN), Caracas, May. [2] ITU-T G.88/Y.134, Architecture for the Automatically Switched Optical Network (ASON), Geneva, October. [3] ITU-T G.83 Architecture of transport networks based on the Synchronous Digital Hierarchy (SDH), Geneva, 2. [4] ITU-T G.872, Architecture of Optical Transport Networks, Geneva, [5] ITU-T G.85 Generic functional architecture of Transport Networks, Geneva, 2. [6] José Miguel Santos, Teresa Almeida and Paula Fonseca, The Use of Distributed Restoration on WDM Networks, DRCN 98 proceedings, Bruges, [7] R. Clemente, G. Ferraris, ASON: New Perspectives for Resilience in Optical Networks, Roadmap towards next generation optical networks, Workshop proceedings, Torino, May. [8] series/p112/default.asp