Traffic Engineering in next generation multi-layer networks based on the GMPLS paradigm Roberto Sabella and Paola Iovanna Ericsson Lab Italy – Via Anagnina 203, 00040 Roma (Italy) e-mail:
[email protected],
[email protected] Abstract After a review of the motivating factors that claim for traffic engineering in next generation infrastructures based on Internet Protocol (IP), this paper briefly reviews the main issues relating to MPLS technology and the possibility to realize a unique control plane in a multi-layer network based on the extended version of MPLS: the GMPLS. In particular, the aspects relating to routing and control data flows in the network while guaranteeing the desired levels of QoS are discussed. Then the paper mention the advances in optical network technology that have led to the possibility to realize a real optical network layer in future infrastructure. Some recent developments in the literature on multi-layer traffic engineering is also reported at the end.
1. Introduction Traffic Engineering (TE) is being considered as one of the hottest topics in the framework of new generation network infrastructures. The definition of TE given by IETF [1] states that the goal of TE is to improve the efficiency and reliability of network operations while optimizing network resource utilization and traffic performance. In recent years the amount of traffic due to Internet-based services has become more and more evident. Thus, future infrastructure must also have multiservice capability, in order to support different types of services, with different requirements in terms of QoS. The unpredictability and the instability of Internet traffic, pose challenging requirements for NGNs: flexibility and the ability to promptly react to traffic changes. The approach based on the over-provisioning of resources, used in nowadays telecom networks, and is not a cost effective solution for NGN due to a huge presence of Internet traffic. Moreover the migration of all services over IP, including the real time ones, requires guaranteeing QoS for a sub-set of services that should be comparable to that one provided by the telecom based networks nowadays. Besides the requirements of flexibility and multi-service capabilities that lead to different levels of QoS requirements, there is another key aspect that needs to be taken into account: cost effectiveness. In fact, a dilemma emerges for carriers and network operators: the cost to upgrade the infrastructure as it is nowadays for fixed and mobile telephone networks, is too high to be supported by revenues coming from Internet services. Actually, revenue coming from voice-based services is usually much higher compared to that derived by current Internet services. Therefore, to obtain cost effectiveness it is necessary to design networks that make an effective use of bandwidth or, in a broader sense, of network resources. The considerations above motivate the adoption of TE in next generation infrastructure. In fact, TE enables the fulfillment of all those requirements, since it allows network resources to be used when necessary, where necessary, and for the desired amount of time. A network with TE capability can dynamically control traffic flows in order to prevent congestions and to optimize the availability of resources. More specifically, TE allows a network to choose routes for traffic flows while taking into account traffic loads and network state, to move traffic flows towards less congested paths, and to react to traffic changes or failures timely. In order to adopt TE solutions, it is necessary to create an intelligent control plane that is able to adequately handle network resources. Such an intelligent control plane will require a paradigm shift in the design of network architecture: some features of traditional SDH/SONET or ATM networks have to be imported in the IP world, and suitably adapted. A relevant network paradigm considered worldwide is based on the Multi-Protocol-Label-
-T13/ 1 -
Switching (MPLS) technique and its generalization, namely GMPLS. In practice, this paradigm consents the re-introduction of the virtual connection into IP based networks that are intrinsically connectionless. The GMPLS control plane allows a harmonization among the Internet world based on packet-switching with the optical world which is intrinsically circuit-switched. Several network architectures and deployment scenarios have been proposed in the literature. Constraint based routing algorithms are also a key component for realizing TE strategies. As regards transport technology, it is widely recognized that WDM optical networks will play a significant role in the realization of the next generation transport infrastructure, which will have to support both traditional and Internet-based services. Besides the general components, TE in optical networks also include the way routing of lightpaths is achieved, the way data flows coming from either a circuit or a packet bearer network layer are groomed and routed onto the lightpaths, and the way lightpath recovery is performed. In all, the key issue in next generation infrastructures is to apply TE in order to control traffic flows in a multi-layer network environment, aiming at optimizing resource utilization and network performance [2,3]. This practically means to choose routes of data flows and the way they are groomed onto optical pipes, taking into account traffic load, network state, and user requirements such as QoS and bandwidth, and to move the traffic from more congested paths to less congested ones. In order to achieve TE in an Internet network context, the Internet Engineering Task Force (IETF) introduced multi-protocol label switching (MPLS) [4], constraint based routing [5], and enhanced link state interior gateway protocols (IGP) [6,7] as key ingredients. Actually, it is widely known that an MPLS control plane together with proper constraint based routing solutions provide the means for achieving TE, so allowing the provisioning of new services based on the “bandwidthon-demand” concept, such as flexible VPNs.
2. The role of MPLS MPLS architecture is based on the separation between data plane and control plane, reusing and extending existing IP protocols for signaling and routing functions, while reintroducing a connection-oriented model in an Internet-based context, [8]. It is based on the encapsulation of IP packets into labeled packets that are forwarded in a MPLS domain along a virtual connection named Label Switch Path (LSP). The key elements of MPLS networks are the label switch router (LSRs), which actually performs label switching, and the edge routers, namely the E-LSRs, which operates at the ingress and at the egress of an MPLS domain. Each virtual connection, i.e. each LSP, can be set up at the ingress LSR by means of an ordered control, before packets forwarding. That LSP can be forced to follow a route that is calculated a priori thanks to the explicit routing function. A crucial feature allowed by the MPLS model relates to the possibility to reserve network resources on a specific path by means of suitable signaling protocols (e.g. RSVP-TE, CR-LDP). Thus, the LSP represents a virtual connection in the MPLS network as the virtual circuits and virtual paths are in the ATM world. In particular, each LSP can be set up, torn down, re-routed if needed, and modified by means of the variation of some of its attributes, including the bandwidth [3]. In fact, the bandwidth of an LSP can be modified dynamically, just for the desired increment [9], according to a specific request at the ingress LSR preserving all the other attributes. Furthermore, pre-emption mechanisms on LSPs can also be used in order to favor higher priority data flows at the expenses of lower priority ones, while avoiding congestion in the network. Another important feature of MPLS relates to the possibility of stacking labels that provides the means to introduce different hierarchical levels instead of the two ones provided by ATM [8]. This feature favors VPN services support and, as it will be clarified in the Section 2.3, allows extending the MPLS control to other technologies.
T13/2
3. Constraint based routing: the key ingredient for traffic engineering The combination of explicit routing function, resources reservation mechanisms and constraint based routing (CBR) in the MPLS network represents the key ingredients to perform efficient traffic engineering strategy [5]. In particular, the criteria utilized to choose routes in a network and, possibly, to re-route traffic flows towards alternative paths, are crucial for applying TE strategies. Such criteria necessarily take into account more parameters than simply network topology. A simple sketch of the constraint based routing operations is shown in fig. 1. In fact, when calculating the route for a requested path (LSP in the case of MPLS based networks), CBR has to take into consideration both network and user constraints. The former regards the link state, resource availability besides network topology, while the latter relates to bandwidth requirements, administrative groups, priority etc. When an explicit route has been computed, the resource reservation procedure is started by means of signaling protocols such as RSVP. In this way, CBR may find longer but less congested paths instead of heavily loaded shortest paths. Thus, network traffic is distributed more uniformly and congestions are prevented.
Network Constraints (link state, resource available, network topology...)
User Constraints
Constraint Shortest Path First (CSPF)
bandwidth requirements administrative groups (colours), priority
Explicit Route
Resource Reservation (e.g. RSVP-TE, CR-LDP)
Figure 1 –Principle of Constraint Based Routing
Basically, two main approaches can be considered for calculating routes: off-line and on-line. Basically, the off-line approach refers to a pre-determined route computation, usually accomplished by an external network optimization tool (e.g. an external server), while the on-line approach refers to an “on demand” route computation, automatically achieved by means of signaling protocols or by an external tool. The off-line approach is adequate for achieving a global path optimization on the basis of a traffic matrix that represents the foreseen connection requests for any pair of network nodes. Such a traffic matrix is usually derived by a statistical expectation of traffic demands. Logically, this method is quite appropriate when traffic demand is reasonably stable: i.e. traffic changes are not so relevant to require a re-design of the routes for the different data flows. This is the case of traditional voice traffic that is quite predicable and reasonably stable: therefore the traffic matrix is quite consistent. Unfortunately, Internet traffic is neither predicable nor stable. Therefore, a pure off-line approach could result inadequate since it could lead on one hand to waste network resources (the transmission pipes are not filled) or, on the other hand, to experience congestion because the amount of traffic is increased and the assigned resources are not enough. To promptly react to Internet traffic changes an on-line approach could be more satisfactory. In particular, the on-line routing method consists in evaluating the route “on demand”, when needed, i.e. when there is a new request or a change of a previous request. Thus, it is suitable to perform a single LSP accommodation at a time. The main problem in those cases is to preserve the stability. In fact, instability can occur when the time necessary to route a new data flow is of the order of the period of time in which the requests are originated. Clearly, the on-line approach is also inadequate to perform a global path accommodation. Moreover, on-line routing may lead to higher resources
T13/3
consumption and is not scalable. As a result, a hybrid approach could be the best solution in order to exploit advantages of both the methods. From the above considerations, it emerges that CBR in real networks is a crucial and complex issue. In this paper we propose a pragmatic traffic engineering system, which utilizes an innovative hybrid routing approach. More specifically, the TE system invokes an off-line procedure for achieving a global optimization of paths calculation, according to an expected traffic matrix, while invokes an online routing procedure to dynamically accommodate, in a sequential way, actual traffic requests, so allowing a prompt reaction to traffic changes. As it will be described with more details in Section 3, the original contribution of the proposed hybrid routing solution consists in the integration of the two routing functions. Such functions can be realized in different ways, without affecting the applicability of the solution. Clearly, the ways the two routing functions are achieved have an impact on system performance: e.g. in terms of accommodated traffic amount.
4. The GMPLS paradigm for new generation networks To extend the features of the MPLS technique, the generalized version of it (namely GMPLS) consents a gradual and future-proof approach towards new generation networks [10,11,12]. In practice, the GMPLS control plane can manage heterogeneous network elements (e.g. IP/MPLS routers, SDH/SONET elements, ATM switches, or even optical elements) using a suitably extended version of well-known IP protocol suite. This makes possible the realization of a single control plane able to handle a whole multi-layer network. In particular, GMPLS extends the MPLS concepts even to non-packet switched technology by means of LSP forwarding hierarchy [13]. This is sketched in Fig. 2.
Fiber LSP
LSP interface hierarchy:
λLSP
»Packet Switch Capable (PSC)
TDM LSP
»Layer 2 Switch capable (L2SC)
Layer 2 LSP
»Time Division Multipl . Capable (TDM) »Lambda Switch Capable (LSC)
Packet LSP
»Fiber Switch Capable (FSC)
Figure 2 – LSP hierarchy in GMPLS
The GMPLS forwarding hierarchy is based on the multiplexing capabilities of the label switch routers (LSRs) interfaces. At the top of such a hierarchy (external LSP in the figure) are nodes that have fiber switch capable interfaces (i.e. fiber cross-connects), at the second stage (λLSP in the figure) are nodes with wavelength switching capabilities (i.e. optical cross-connects), at the third stage (TDM LSP) there are nodes with TDM switching capabilities (e.g. SDH cross-connects), at the fourth stage (layer 2 LSP) are the nodes with layer 2 switching capabilities (e.g. real MPLS routers or ATM switches), at the last stage (packet LSP) are nodes with packet switching capabilities (e.g. IP routers). Any stage can be associated with a network cloud. The outer cloud represents the packet LSP domain. The layer 2 LSP domain is nested inside the packet cloud and so on up to the inner cloud representing the fiber LSP domain. It is to be highlighted that each LSP should be generated T13/4
and terminated on homogeneous devices (i.e. belong to the same network cloud). On the other hand, a packet-switch capable LSP can be nested and tunneled into an already existing higher order LSP. GMPLS can support different network scenarios, where heterogeneous layers can cooperate in several ways at the convenience of manufacturers and operators. Without loosing in generality, we consider a two-layers network as a reference scenario, consisting in an IP/MPLS layer, whose network elements are basically LSRs, and a WDM transport layer, whose nodes are optical crossconnects (OXC) as depicted in figure 3. Specifically, just packet LSP and λLSP are considered. The latter ones represent end-to-end optical connections: that are lightpaths. The inter-working between the IP/MPLS layer and the optical layer is another key issue. Particularly, the packet-based structure of the IP/MPLS layer and the circuit-based construction of the optical layer have to be harmonized. This means that any lightpath bundles several LSPs, characterized by different bandwidth attributes. The bandwidth attribute of each LSP belonging to the IP/MPLS layer varies in a continuous range, while the lightpath bandwidth is fixed to the wavelength channel bit rate. MPLS Network
LSR 2
LSR 3 LSR 4
LSR 1
LSR 6
LSR 5 E
Optical Network OXC
A
D
OXC
I
OXC
G OXC
OXC OXC
OXC
B MPLS link
H OXC
Lightpath
C
Optical link
Figure 3 – Multi-layer reference network scenario
Different deployment scenarios can be envisaged for optical networks based on GMPLS concepts, such as overlay and peer as extreme cases [12]. Each of them defines a different level of interworking between the IP/MPLS layer and the optical one. The overlay model is based on a clientserver approach. In this context, the optical layer acts as a server of the IP/MPLS layer. The control planes are separated in this case and communicate, one to each other, by means of a standard user network interface (UNI) [14]. In this case the IP/MPLS network asks for a connection and the optical network manages its resources in order to set up that according to the service level agreement (SLA). In the peer model a single control plane manage the whole network. In this way all the nodes, both the IP/MPLS and the optical ones, act as peer sharing the same complete topological view. This allows a network operator having a single domain composed of different network elements providing a greater flexibility. The price for that is the amount of information that has to be handled by any network element. The deployment scenario has an impact on routing strategy. In particular, two main strategies can be adopted in GMPLS-based network: single-layer and multi-layer. In a single layer approach, the LSPs are aggregated by edge LSRs into lightpaths. At this point the connection requests are expressed in terms of number of wavelengths that are requested per each pair of optical nodes. The optical layer is then responsible to find the routes for the optical LSPs and assign the wavelengths: that is solving the routing and wavelength assignment (RWA) problem [15,16]. Instead, in a multi-layer approach the aggregation and routing are jointly performed, allowing a LSP to be routed on a concatenation of lightpaths in a single routing instance and leading to an efficient use of network resources [17,18]. It is logical that the awareness of the status
T13/5
of all the network elements and the possibility to manage the whole set of network resources allows to perform more efficient routing functions. Regarding QoS handling, GMPLS can reserve bandwidth for individual LSP at any hierarchical level. The potentiality of a GMPLS control plane in terms of advanced traffic engineering capabilities, provided by a cooperative inter-working among layers, are remarkable, but the feasibility of a simple and effective TE solution is still a challenging task. In effect, different technological and architectural aspects have an impact on the practical implementation of TE strategies: a.
Complexity of CBR function – the realization of such a function taking into consideration simultaneously all network and user constraints, in a network made of many and heterogeneous network elements is very complex. Thus, a simple and practical approach is needed for routing.
b.
QoS handling – managing different QoS requirements for several classes of services in the network is another complex task. Specifically, it deals with the ways to achieve traffic segregation, routing according to different priority levels, and pre-emption.
c.
Signaling – in order to be efficient, the CBR has to know the updated link state of the whole network, and possibly the map of all the LSPs. This means a huge information flood through the network. Therefore, it is necessary to find a reasonable trade off between routing efficiency and amount of information to be flooded throughout the network.
d.
Prompt reaction to traffic changes – in principles, the network should be able to react to traffic changes promptly. This requires the possibility of realizing a dynamic routing of the data flows according to such requests. This could also lead to stringent technological requirements for the nodes at all levels. Moreover, on-line routing leads to non-optimal routes if compared with global routing performed off-line. Thus, it is necessary to find a reasonable combination of dynamic routing facilities and static routing.
5. Requirements for an enhanced signaling The development of the GMPLS requires a suitable extension of MPLS signaling and routing protocols in order to manage heterogeneous technology [11]. That means that the routing protocols, such as OSPF-TE, have to perform flooding of detailed and updated topology information and attributes for each link at different network layers and signaling protocols, as RSVP-TE, have to handle generalized label concept to support the establishment of LSPs at any hierarchical level [19] As a result, extended routing and signaling protocols have to cope with a huge and heterogeneous amount of information respect to a pure MPLS based network, leading to scalability issue. For instance, the overall number of links in an optical network can be several orders of magnitude bigger than in an MPLS network. To address such an issue the concept of link bundling has been introduced [12]. In fact, similar optical parallel links can be aggregated to form a bundle for the routing purpose. On the other hand the signaling of each individual component of the bundle requires a new protocol, introduced specifically for link management in the optical networks, named link management protocol (LMP) [12,20]. Specifically, LMP is responsible for i) establishing and maintaining control channel connectivity, ii) verifying the link physical connectivity, and iii) rapidly identifying link, fiber, and channels failure within the optical domain. However, the efficiency of the constraint based routing depends not only on the amount of disseminated information on network topology and resource availability, but also on the frequency of information updating. The more detailed and updated is the information collected in the link state
T13/6
data base, the better the routing decision is likely to be. A dynamic link state routing suffers this problem especially due to the fast changeability of the constraints to be considered. These issues could be addressed by inheriting mechanisms already used in the MPLS network, such as threshold methods that avoid excessive flooding or methods based on a timer setting an upper bound on flooding frequency [21]. However, in a GMPLS scenario, these mechanisms could be not sufficient to make a completely dynamic link state routing approach feasible. This is a further reason for using a hybrid routing approach, as mentioned in Section 2.2. In this way the dependence of route computation by the flooding information is relaxed.
6. Evolution of the optical networks towards TE solutions The evolution of optical transmission has led to the availability of WDM systems, which provide the means to transport hundreds of Gb/s onto the same optical fiber, for thousands kilometers without signal regeneration. This was made possible thanks to different technological enablers: i) the evolution of high speed transceivers, which are reached the capacity of 40 Gb/s on a single channel, ii) the evolution of the WDM component technology that make possible the multiplexing of many tens of optical channels onto the same fiber, and iii) the maturity of optical amplifiers’ technology that allowed to strongly reduce the cost of optical links. If optical transmission technology pushed massively to enhance the capacity of network links, the evolution of optical switching has led to the possibility to concretely realize optical network elements. In fact, optical add-drop multiplexers (OADMs), based on different types of technologies, provide the means for the realization of optical rings that could cover metropolitan areas of different geographical sizes. Furthermore, the availability of optical cross-connects (OXCs) allows the interconnections of different optical rings in order to realize a real optical layer in the transport network infrastructure. The continuous evolution of technology is allowing the final step: the possibility to realize real optical meshed networks, which represents the best options in order to build up whatever network topology, with all the advantages coming with the mesh architectures. Besides optical transmission and switching a third element is needed for future network infrastructure: the optical control plane, which will allow intelligence to be put onto the optical network layer. Put intelligence in optical network essentially means three things: i) the possibility of a flexible utilization of the optical bandwidth (i.e. the lightpaths), ii) a fast circuit provisioning, and iii) the possibility to manage failures in an effective way. Besides the IETF standardization body, the ITU has introduced the model of the automatic switched optical networks (ASON). In this model different control and management reference models are considered: i) the switched connection, ii) the permanent connection, and iii) the soft permanent connection. At the time being, the effort of research effort in both industry and academia is to provide concrete solution for the control plane architecture; its interoperation with the management plane; and in the way TE can be concretely applied. Relevant hot topics are, in fact, the way data flows are groomed into optical path (lightpaths) and routed, in the way routing and wavelength assignment problem is solved, in the multi-layer approaches to route data flows at the different hierarchical levels, and in the way data flows are protected and restored in case of failure of either network nodes or links.
7. Traffic engineering system for new generation optical networks So far most of the literature in optical networking has treated in depth the routing and wavelength assignment (RWA) problem in the optical layer of the network, independently of which layer is above that one. Furthermore, a lot of effort has been devoted on the need of wavelength conversion function in optical networks: whether this function is needed or not, what is the gain of having wavelength converters and what are the related costs, if there exist a reasonable trade off of having
T13/7
a limited number of converters and, in case, where those devices have to be located, and what technology is suitable for wavelength converters. Another relevant portion of the optical network literature is devoted to protection and restoration issues in optical networks. Even in those cases, most of the paper just considers the optical network layer. In the last few years, several papers deal with specific TE functions such as routing, wavelength assignment, and pre-emption algorithms [7,8,9,10] in an optical layer, possibly overlaid to an electrical layer. The “multi-layer” aspect is being considered in some way. In those papers the two sub-problems of i) design of logical topology of the optical network (i.e. the set of wavelength paths), and ii) the routing of the data flows at the IP/MPLS layer onto the logical topology, are solved in a separate way (e.g. in two different steps). Differently, a multi-layer approach would consist in simultaneously solving these two sub-problems. Our group is working on TE in multi-layer networks since few years ago. A TE strategy involving and combining specific TE functions in a multi-layer network has been reported in [12]. That paper also presented the main building blocks and the mode of operations, discussing the main characteristics of the system as a whole. Key building blocks of that solution were reported in [13,14,15]. In particular, the routing problem has been approached in a multi-layer fashion, in the aforementioned sense. The present paper reports for the first time, to the best of our knowledge, an integrated solution of TE in the technical details, the performance analysis of this TE system as a whole, aiming at both assessing its feasibility and evidencing its advantages with respect to a relevant example among traditional over-provisioning approaches. What are the main aspects the international research is being analyzing about TE? In the following some relevant topics are reviewed, with the attempt to identify the hot issues that still remain open. The main requirements for a TE system of a new generation network can be summarized as follows:
7.1.
a.
Optimize the use of network resources (e.g. the link bandwidth and the node throughput), by means of an “elastic” use of the bandwidth resource;
b.
Actualize the “bandwidth-on-demand” concept;
c.
Support different classes of service, including real time traffic (e.g. the CoS foreseen in the DiffServ scenario defined by IETF), and guarantee the required QoS.
Routing in a multi-layer fashion
One of the most significant features of the GMPLS paradigm is the possibility to control the entire set of network resources simu ltaneously, through one control plane. This means that the control plane knows all the information about the network nodes, both optical (e.g. OXCs) and electrical (e.g. IP/MPLS routers), and all the information about the status of occupancy of each link, either at the optical or the electrical layer. This practically signifies the possibility to make routing decision in a multi-layer fashion with the significant advantage of optimizing network resources in terms of link capacity, node throughput, number of data flows to be controlled and so forth. If this kind of statement is often cited in many occasions, it is not so well known how to do that in an effective and practical way. This is still a hot issue that many researchers is being considering. To address this point it is necessary to set the architectural problem in the right way and to approach the routing models and related algorithms by exploiting at the best the methods of modern operational research. For sure multilayer routing will remain a hot issue in the next years.
T13/8
7.2.
Off-line or on-line routing?
Another passionate debate relates to the approach of making routing and related pros and cons. There are basically two approaches. The off-line approach consents the possibility of exploiting the power of operational research algorithms that allow the routing solution to be found in a nearly optimal way. This method, combined with a multi-layer scheme could allow to optimize network resources for a given demand of traffic. For these reasons, off-line methods are particularly suitable for configuring an entire set of data paths, at different layers, in an optimal way, for a given network topology and a given traffic demand matrix expressed, for instance, in terms of set of data flows to be accommodated, with given bandwidth attribute, for a each pair of network nodes. The entity that could perform such an off-line path provisioning could be either a management system or similar network unit. Of course, this type of operation can be done each time the traffic demand changes significantly in a semi-permanent way. It is neither indicated to handle fast traffic variations, nor unpredictable traffic demand, nor even suitable to promptly accommodate, “on-demand”, new requests. Conversely, the on-line approach is suitable to promptly react to traffic changes, in a dynamic fashion. To properly works it needs routing protocols that flood throughout the network updated information about the status of the links and the nodes, in order to take appropriate routing decisions. Of course, the price to be paid in case on-line dynamic routing algorithms are used lays in a sub-optimal utilization of network resources with respect to off-line methods. Furthermore, if the routing time is comparable with the time related to the generation of new requests, instability could occur. It is clear that the choice of the right routing strategy depends on the network context it has to operate. Many times intermediate or hybrid approaches are mentioned in the literature; even though there are not many solutions in the literature very assessed. A hybrid solution in the framework of next generation Internet networks remains a challenging open issue to be addressed.
7.3.
Pre-emption and re-routing strategies
The fact that new generation infrastructures must be multi-service and multi-classes and thus must on one hand guarantee different levels of QoS and, on the other hand, make an effective use of network resources, lead to address another relevant subject: the possibility of pre-empt lower priority data flows at the advantage of higher priority ones. Moreover, if the goal of not loosing traffic and make an effective resource utilization is mandatory, it is necessary to foresee the possibility of re-routing traffic. Sometimes this latter topic is regarded as “bandwidth management” or, in a broader sense “traffic engineering”. The challenging issues in this case are: i) what are the criteria that lead to pre-empt data flows? ii) what is the price that can be paid in terms of data flows to be pre-empted and possibly lost? Preempted traffic have to be re-routed elsewhere? How traffic segregation has to be accomplished? How these issues can be applied in a multi-layer scenario? The answers to these questions represent several open issue that have to be addressed.
7.4.
Protection and restoration strategies
The considerations made so far regarding routing strategies and related algorithms can be essentially replicated in the considered topic. The multi-service mandate of next generation networks lead to the fact that the network has to offer different levels of protection and/or restoration against node or link failures. If on one hand the network will be required to provide the carrier class performance to mission critical services (let’s
T13/9
consider for instance the 50 ms time guaranteed by SDH system in case of failure), on the other hand there is no point to provide the same level of performance for low priority traffic (e.g. best effort). Even in this case network resources can be utilized to offer several levels of protection and restoration. All the questions of the previous sections remain valid. In particular, we could ask what traffic segregation? What levels of protections? How to distribute protection and restoration over the different network layers? Is there any escalation strategy from a layer to another in case of failure? Al those issues impact to the network architecture significantly. It is certain that there is a lot of research work to be done to provide convincing answers and practical solutions for real systems.
7.5.
What traffic engineering is really needed?
Last but not least it is necessary to find a reasonable and convincing answer to the question above. Besides the charm of the topics mentioned above that push many researchers to propose ideas and solutions, the most important aspect lays in finding strategies and related algorithms that can be concretely applied in real network to solve real problems of carriers and operators and that lead ultimately to save CAPEX and OPEX or even better to earn much more money! As a matter of fact, too often fascinating solutions are too complex to be applied and lead to small gains. Each time one find a TE solution should ask him/her self: why the solution I have found is much more convenient of an over-provisioning solution? How much I gain using it instead? How complex is the realization and the operation of such a solution? Even though these questions seem more uncomfortable and perhaps less challenging, from an academic point of view, with respect to the ones of the previous sections, they should be taken seriously into account when planning a research activity. A good answer to that question could have a dramatic impact in convincing operators to adopt systems employing traffic engineering.
8. References [1] D. Awduche et al. , “Requirements for Traffic Engineering Over MPLS”, IEEE Communications Magazine, IETF RFC 2702 , September 1999. [2] G. Ash, “Traffic Engineering & QoS methods for IP-, ATM-, & TDM-based multi-service network” Internet draft , October 2001 [3] D. Awduche et al., “Requirements for Traffic Engineering over MPLS”, IETF RFC 2702, September 1999 [4] E. Rosen, et al. “ Multiprotocol Label Switching Architecture”, IETF, RFC 3031, January 2001. [5] D. Awduche et al. “Overview and Principles of Internet Traffic Engineering”, IETF, RFC 3272, May 2002. [6] R. Coltun, "The OSPF Opaque LSA Option," RFC 2370, July 1998. [7] D. Katz, “Traffic Engineering Extensions to OSPF”, work in progress IETF draft [8] B. Davie et al. , “ MPLS Technology and Applications “, Academic Press, 2000. [9] D. Awduche et al, "RSVP-TE: Extensions to RSVP for LSP Tunnels", RFC 3209, December 2001. [10] E. Mannie et al. “Generalized Multi-Protocol Label Switching (GMPLS) Architecture” work in progress IETF draft . [11] A. Banerjee, et al., “Generalized Multiprotocol Label Switching: An Overview of Signaling Enhancements and Recovery Techniques”, IEEE Communications Magazine, July 2001, pp. 144151. [12] A. Banerjee, et al., “Generalized Multiprotocol Label Switching: An Overview of Routing and Management Enhancements”, IEEE Communications Magazine, January 2001, pp. 144-150.
T13/10
[13] K. Kompella et al. “LSP Hierarchy with Generalized MPLS TE”, work in progress IETF draft [14] OIF Architecture, OAM&P, PLL, & Signaling Working Groups: “ User Network Interface (UNI) 1.0 signaling specification; OIF200.125.3. [15] R. Dutta et al. “Traffic Grooming in WDM Networks: Past and Future”, IEEE Networks, November/December 2002, pp. 46-56. [16] J. Wang et al. “ Improved Approaches for Cost-Effective Traffic Grooming in WDM Ring Networks: ILP Formulations and Single-Hop and Multihop Connections”, IEEE, Journal of Lightwave Technology, vol. 19, n.11, 2001, pp. 1645-1653. [17] M. Settembre et al. “A Multi-Layer Solution for Paths Provisioning in New Generation Optical/MPLS Networks” accepted for publication on IEEE Journal of Lightwave Technology. [18] R. Sabella et al. “Strategy for Dynamic Multi-Layer Routing in New Generation Optical Networks based on the GMPLS Paradigm” submitted to IEEE, Journal of Lightwave Technology. [19] P. Ashwood et al. “ Generalized MPLS-Signaling functional description”, work in progress IETF draft, , October 2002. [20] J. P. Lang, “Link Management Protocol”, work in progress IETF draft, , May 2002 [21] A. Shaikh et al. ,“Evaluating the Overheads of Source-Directed Quality of Service routing”, International Conference on Network Protocols (ICNP, ’98), 1998.
T13/11
