A Novel Signaling Approach to Encompass Physical Impairments in GMPLS Networks Filippo Cugini (1), Nicola Andriolli (2), Luca Valcarenghi (2), Piero Castoldi (2) 1 : CNIT National Lab. of Photonic Networks, Via Cisanello 145, 56124 Pisa, Italy, e-mail:
[email protected] 2 : Scuola Superiore Sant’Anna, Via Cisanello 145, 56124 Pisa, Italy, e-mail:
[email protected],
[email protected],
[email protected] Generalized Multi-Protocol Label Switching (GMPLS, [2]) has been introduced to dynamically set up all-optical end-toend connections, i.e. lightpaths, within domains of transparency. GMPLS is a protocol suite able to manage the common control plane between edge IP routers and intermediate alloptical NEs. However end-to-end signal quality has to be preserved and only limited size domains of transparency can be set up. Currently, GMPLS does not take into account the evaluation of the signal quality. So far, to practically achieve adequate signal quality for every possible route, the domain of transparency has been statically designed during the planning process considering every worst case situation, e.g. every lightpath between the farthest nodes which traverses the most impaired links. This has determined extremely small domains of transparency and the presence of numerous expensive electronic transponders. In [1] two possible solutions to dynamically estimate the signal quality are evaluated. The first solution is based on an extended version of current GMPLS distributed control plane. This solution introduces additional physical information into the routing protocol, e.g. Open Shortest Path First with Traffic Engineering extensions (OSPF-TE). However OSPF-TE already suffers from stability problems and the introduction of further extensions appears to be practically unfeasible. The second approach evaluated in [1] proposes a centralized solution: the routing algorithm runs on a centralized network server which contains the physical network parameters. This solution however presents several drawbacks: scalability and central database failures are the most significant, but even fast restoration becomes difficult adopting such a centralized routing system. Because of the aforementioned drawbacks even this approach is practically unfeasible. In this paper we introduce a novel approach based on an extended version of the distributed GMPLS control plane to encompass linear physical impairments in GMPLS networks. Extensions required to handle physical impairments during the lightpath set up are introduced into signaling (i.e., Resource ReSerVation Protocol (RSVP-TE) and management (i.e., Link Management Protocol (LMP) [3]) protocols.
Abstract – Next generation networks are expected to be characterized by domains of transparency in which a GMPLS common control plane is adopted to manage both the IP layer and the optical transport layer. Within domains of transparency the end-to-end lightpath signal quality has to be guaranteed. However, currently, GMPLS does not take into account the evaluation of physical impairments. Thus just limited size domains of transparency are practically achievable. In this study a novel approach to dynamically estimate the end-to-end signal quality is proposed. The method is based on the introduction, into current GMPLS signaling and management protocols, of extensions which encompass the physical parameters that characterize the optical transport layer. The proposed approach allows to detect whether lightpaths cannot be set up because of accumulated linear physical impairments during their signaling phase. Numerical results show that the lightpath blocking probability due to physical impairments decreases by implementing suitable routing policies, such as multiple attempts. Thus few very impaired links can be introduced into large domains of transparency without affecting the overall network performance. An experimental implementation based on PXC and IP routers is also presented to validate the feasibility of the proposed approach.
I. INTRODUCTION Currently deployed optical networks are still based on optical connections terminated at each network node by optoelectronic transponders. Because of their high cost, fixed bit rate, and fixed protocol format, optoelectronic transponders limit the network evolution. The introduction of transparent Network Elements (NE), e.g. Photonic Cross-Connects (PXC), has boosted the network evolution from the technological viewpoint by allowing the introduction of domains of transparency [1], i.e. all-optical subnetworks at whose edges optical signals undergo optoelectronic conversion. This has allowed to overcome the limits and the high costs due to optoelectronic transponders. From the protocol viewpoint a paradigm called This work was supported in part by the European Union under IST e-Photon/One Network of Excellence and by the Italian Ministry of Education and University (MIUR) under FIRB project “Enabling platforms for high-performance computational grids oriented to scalable virtual organizations (GRID.IT)” IEEE Communications Society Globecom 2004 Workshops
369
0-7803-8798-8/04/$20.00 ©2004 IEEE
the data link, to correlate the link property information (e.g. interface switching type) and locate link failures. The GMPLS protocol suite implements a distributed control plane which permits the introduction, mainly in the network core, of different types of all-optical Network Elements (NE), e.g. transparent PXC. This determines a reduction of the amount of expensive optoelectronic transponders and the introduction of all-optical subnetworks, namely domains of transparency. The end-to-end lightpath provisioning over domains of transparency based on GMPLS protocols assumes that every possible route eligible by the routing protocol is characterized by the adequate signal quality. Indeed no physical information are included in current version of the GMPLS protocol suite.
Routes from source to destination are dynamically computed without taking into account physical impairments. Only upon lightpath establishment through the reservation protocol the amount of impairments are dynamically computed. At the destination node the lightpath setup request can either be accepted or blocked based on the amount of accumulated impairments. This solution, besides being distributed, does not require modifications to the routing protocol. Moreover the proposed solution, by introducing the dynamic evaluation of the signal quality, potentially allows to both set up large domains of transparency and reduce the number of optoelectronic regenerators. Indeed simulation results show that even though the domain of transparency dimensions would not allow some lightpaths to be set up because of accumulated physical impairments (causing a non-zero lightpath blocking probability), the proposed approach allows to decrease the lightpath blocking probability experienced in successive lightpath set up attempts. The solution is practically feasible and an experimental demonstration based on IP routers with Gigabit Ethernet interfaces and Photonic Cross-Connects (PXC) is shown.
III. IMPACT OF PHYSICAL IMPAIRMENTS ON DOMAINS OF TRANSPARENCY
Without loss of generality this study considers a GMPLS network consisting of two types of network elements: IP routers with WDM interfaces (PSC+LSC interfaces) and transparent PXC with LSC interfaces. At the network edge routers elaborate data and control packets at the electrical level. The core of the network consists of a domain of transparency made of PXCs. Edge routers run GMPLS protocols over an out-of-band control plane. OSPF-TE with standard GMPLS extensions is utilized for distributed route calculation without considering physical impairments. RSVP-TE with standard GMPLS extensions is used for lightpath provisioning. Link Management Protocol (LMP) runs between adjacent NEs to maintain channel connectivity and link property correlation. Because physical impairments are not considered by the routing algorithm, the adequate signal quality between every source and destination pair has to be a priori guaranteed considering the lightpath traversing the most impaired links. Indeed the domain of transparency has to be dimensioned during the planning process in order to set to zero the lightpath set up blocking probability induced by physical impairments. Fig.1 shows a simple case example in which a single linear physical impairment is considered. The domain of transparency is a 6x6 Manhattan street topology with transparent PXC in each network node. Nodes are connected through bidirectional WDM links. All the links but three (i.e., the deprecated links depicted with saw-toothed lines) introduce the same amount of impairment (e.g., attenuation). So far, to properly dimension the domain of transparency worst case routes between the farthest nodes, e.g. from node 1 to node 36, have to be statically evaluated during the planning process. For example in the network depicted in Fig.1, 252 different shortest paths can be utilized to connect node 1 and 36. Among the 252 paths, the paths passing through all the three deprecated links have to be avoided by the routing algorithm. However if the routing algorithm does not take into account physical impairments, the domain of transparency, because of few potential routes that determine
II. GMPLS-BASED OPTICAL NETWORKS GMPLS (Generalized Multi-Protocol Label Switching) is an extension of the MPLS control plane which provides a common control plane between the IP layer and the optical transport layer. GMPLS extends the MPLS support of Packet Switching Capable (PSC) interfaces to other types of interfaces, i.e. Time-Division Multiplex Capable (TDM), Lambda Switch Capable (LSC), and Fiber-Switch Capable (FSC) interfaces. TDM interfaces forward data based on the time slot. An example of such an interface is an interface on a SONET/SDH Cross-Connect. LSC interfaces treat data based on the wavelength on which the data is received, e.g. an interface on a Photonic Cross-Connect (PXC) that can operate at the level of an individual wavelength. FSC interfaces forward data based on the fiber on which the data is received. An example of such an interface is an interface on a PXC that can operate at the level of a single fiber. In order to properly manage all types of interfaces, GMPLS specifies several extensions to the routing protocol (i.e., OSPF-TE) and signaling protocol (i.e., RSVP-TE). Moreover the GMPLS protocol suite introduces a novel network protocol called Link Management Protocol (LMP). OSPF-TE is extended mainly to include into the Traffic Engineering extensions the management of the link protection type, e.g. shared or dedicated, the shared risk link group to describe the relationship between links with a shared vulnerability, and the interface switching capability descriptor (e.g., PSC, LSC). RSVP-TE is extended mainly to encompass the different types of label which refers to the different types of interfaces. LMP is introduced to maintain the out-of-band control channel connectivity, to verify the physical connectivity of
IEEE Communications Society Globecom 2004 Workshops
370
0-7803-8798-8/04/$20.00 ©2004 IEEE
1
presents significant disadvantages when compared to a distributed model. Rapid restoration in case of (multiple) failures is not achievable on a centralized routing system and even scalability and database failure are fundamental issues. Moreover this approach is not consistent with the distributed routing philosophy that has determined the success of the Internet. Because of these drawbacks even this approach is practically unfeasible. V. PROPOSED GMPLS SIGNALING APPROACH
36
This study proposes an approach to dynamically estimate the signal quality during the set up process and, as a consequence, to reduce the amount of electronic regenerators and realize large domains of transparency. The proposed approach introduces local databases in each Network Element (NE) to store the physical parameters that characterize the NE itself and the links connected to its interfaces. Network Elements automatically maintain local database synchronization and dynamically evaluate the lightpath signal quality by extending the GMPLS distributed control plane. Local database synchronization is guaranteed between adjacent NEs by exchanging LMP packets over the out of band control plane. Proper extension is introduced into the LMP link property correlation procedure. LMP Link Summary message is extended to contain the physical parameters specified in the local database. No modification are introduced in the OSPF-TE routing protocol which elaborates routes ignoring physical impairments. The RSVP-TE signaling protocol is extended to dynamically estimate the lightpath signal quality during the set up process. The lightpath set up process collects the physical parameters that characterize every traversed NE and link from the source node to the destination. The source node indeed generates an extended version of the RSVP PATH message containing the physical information of the transmitting interface and of the first attached link. Every traversed network element, before propagating the message, updates these parameters by adding its own local values (only linear effects are considered). Admission control at the destination node compares the overall accumulated parameters with the local parameters that characterize its receiver interface. If the accumulated parameters are within an acceptable range, the lightpath set up request is accepted and a RSVP RESV message is sent back to the source node. Otherwise the lightpath request is rejected and a proper RSVP ERROR is sent to the source node. In case the request is rejected, further set up attempts following different routes are realized in order to avoid the block induced by physical impairments. The proposed signaling method allows, in domains of transparency statically characterized by lightpath blocking probability greater than zero, to minimize the blocking probability of subsequent set up attempts. Indeed, by
Fig. 1: 6x6 Manhattan Street network with three deprecated links
excessive signal degradation, has to be reduced, even if all other paths guarantee adequate signal quality. As a consequence, even considering just one single physical impairment, several optoelectronic transponders have to be introduced thus increasing the overall network cost. In order to overcome these issues, the common control plane between the IP layer and optical layer can be properly extended. IV. PREVIOUS WORK In [1] an overview of the issues for a common control plane between the IP layer and the optical transport layer is presented. Two different approaches are evaluated to include the management of physical parameters into the common control plane. The first approach refers to an extended version of current GMPLS distributed control plane. The solution is based on the introduction of physical parameters into the routing protocol at the IP layer, e.g. OSPF-TE. Additional state information have to be introduced in order to allow the routing algorithm to evaluate the impairments that characterize every selected route. Many different linear and non-linear impairments can be considered. Linear impairments, such as Amplified Spontaneous Emission (ASE) and Polarization Mode Dispersion (PMD, [4]), are independent of signal power and they affect wavelengths individually. On the contrary nonlinearities (e.g. Four Wave Mixing, FWM [5]) are power dependent and they induce crosstalk between different wavelengths. This OSPFbased solution, even excluding nonlinear effects, requires to handle a significant amount of parameters. Moreover current OSPF-TE routing protocol is already affected from stability and scalability problems, especially in case of failures. As a consequence the OSPF-based solution proposed in [1] appears to be practically unfeasible. The second approach evaluated in [1] refers to a centralized model derived from the centralized solutions adopted in current optical transport networks. Routing is done centrally using a centralized database containing the physical information and the complete network view. This solution allows to obtain the best performance in terms of management of physical parameters and resource allocation, however it IEEE Communications Society Globecom 2004 Workshops
371
0-7803-8798-8/04/$20.00 ©2004 IEEE
0,08
excluding rejected routes, the blocking probability of subsequent attempts decreases to zero if at least one path is practicable, and the lightpath set up is guaranteed. Further set up attempts however increase the amount of required signaling. For this reason an accurate evaluation of the domain of transparency is still necessary during the planning process.
Blocking probability
1 link
VI. VALIDATION OF THE PROPOSED APPROACH In this section the proposed approach is evaluated through simulations considering the 6x6 Manhattan street topology with PXC in each network node (Fig. 1). Nodes are connected through bidirectional WDM links. To evaluate just the effect of physical impairments, links have infinite wavelengths. Every link is characterized by a critical linear impairment: to take into account this effect a suitable parameter αi is associated to each link i. The value of this parameter on network links is randomly distributed, with average value α and 25% excursion range. Since this impairment is additive, it determines a total effect on a lightpath equal to the sum of the parameters αi of the traversed links. We assume that a maximum value αMAX = 11·α
0,06
0,04 After two attemp ts After one attemp t
0,02
0 α
1.5α 2α 2.5α Deprecated Link value α'
3α
Fig. 2: Average blocking probability due to physical impairments after the first and second set up attempt
(1)
can be accepted to successfully establish a lightpath. On the considered 6x6 topology all the shortest paths between the farthest nodes (e.g. from node 1 to 36) traverse 10 hops: using these values every possible shortest path between any source-destination pair can be set up. However, if few deprecated links, characterized by a parameter α’ higher than α, are present in the network, some shortest paths cannot be available due to the physical impairments (the number of impaired links cannot be too high, otherwise the supposed connectivity among all nodes in the transparency domain is no more guaranteed). In this case, for each source-destination pair, only a fraction of the possible paths (e.g. between node 1 and 36 there are 252 different shortest paths) can be selected. Lightpaths whose routes accumulate physical impairments that do not exceed the maximum value αMAX are accepted, otherwise they cannot be set up because of excessive impairment accumulation. In the following we have studied the effect on network blocking performance of some deprecated links, exhaustively considering all their possible positions: in particular the presence of 1, 2 and 3 deprecated links (with a parameter α’ ∈ [α, 3·α]) has been evaluated, placing them in all the locations of the 6x6 topology which guarantee the connectivity among any couple of nodes. Fig. 2 shows the average blocking probability as a function of α’. Uniform lightpath requests are assumed between each source destination pair. Lightpath routes are computed utilizing Dijkstra’s routing algorithm. Dotted lines represent the average blocking probability due to physical impairments that statically characterize the domain of transparency. Adopting the proposed method, the dotted lines represent also the average IEEE Communications Society Globecom 2004 Workshops
2 link 3 link
blocking probability at the first dynamic set up attempt. Continuous lines show the average blocking probability obtained performing the second set up attempt upon the complete exclusion of the first rejected route. Results show that the proposed method, exploiting successive set up attempts, allows a significant blocking probability reduction even when three deprecated links, introducing heavy physical impairments, are present in the network. Moreover since the second attempt is often successful, the amount of required signaling and the setup time are limited. This drives the consideration that few very impaired links can be introduced into domains of transparency without affecting the overall network performance VII. IMPLEMENTATION FOR GIGABIT ETHERNET CONNECTIONS
We applied the proposed approach in a simple lightpath set up process between a source Router A and a destination Router C. Assume that there is a domain of transparency between the two routers and multiple routes are available from A to C. One of them, which traverses the PXC B, is shown in Fig. 3. Routers are equipped with Gigabit Ethernet (GbE) 1000BaseLX interfaces which operate at 1310nm. For these interfaces the physical impairments determine a maximum operative distance of 10 km. Local database is created and maintained in every node by the network manager. Local database on Routers describes the physical parameters that characterize the GbE interface, i.e. maximum permitted distance, operating wavelength, minimum launch power, receiver saturation and sensitivity. Router A and C
372
0-7803-8798-8/04/$20.00 ©2004 IEEE
PXC B Att. @ 1550nm [dB] Router A Gigabit Ethernet 1000BaseLX Max length on SMF [m]
10000
Wavelength [nm]
1270
Max Wavelength [nm]
1610
Receiver saturation [dBm] Receiver sensitivity [dBm]
-9 -3 -20
Router C Gigabit Ethernet 1000BaseLX Link 2
1310
Min launch power [dBm]
1
Min Wavelength [nm]
Link 1
Type
G.655 2500
Type
G.655
Length [m]
Length [m]
5000
Att. @ 1310nm [dB]
3
Att. @ 1310nm [dB]
5
Att. @ 1550nm [dB]
2
Att. @ 1550nm [dB]
3
LMP
Router A
Max length on SMF [m]
10000
Wavelength [nm]
1310
Min launch power [dBm]
-3
Receiver sensitivity [dBm]
-20
LMP
PXC B
RSVP PATH
Router C
RSVP PATH
RSVP PATH - PHY Descr. Obj
RSVP PATH - PHY Descr. Obj
Length [m]
5000
Length [m]
7500
Wavelength [nm]
1310
Wavelength [nm]
1310
Power [dBm]
-9
Receiver saturation [dBm]
-14
Power [dBm]
-18
Fig. 3: Experimental implementation of the proposed signaling approach
uted GMPLS control plane. Extensions required to handle physical impairments are introduced into signaling (i.e., RSVP-TE) and management protocols (i.e., LMP) and no modifications to the routing protocol are applied. Numerical results showed that the proposed approach, based on the dynamic evaluation of the signal quality, significantly decreases the blocking probability due to linear physical impairments. This allows to build large domains of transparency and to limit the number of required optoelectronic transponders. Moreover an experimental implementation of the proposed solution based on PXC and routers equipped with Gigabit Ethernet interfaces has been presented for the validation of the proposed approach.
maintain also a local database containing the physical information (type, length, attenuation, etc.) describing the attached links, 1 and 2 respectively. PXC B maintains a local database containing its own parameters (attenuation, wavelength range, etc) and links 1 and 2 descriptions. LMP packets are exchanged over an out of band control plane to maintain synchronization between adjacent NE databases thus preventing incongruence between neighbors. Link 1 parameters are exchanged between router A and PXC B and link 2 parameters between PXC B and router C. Assume that a connection request from source node A to destination node C arrives. The routing algorithm has not been modified. It is not aware of physical impairments and it selects the route passing through PXC B. To set up the lightpath a RSVP PATH message starts from node A. It contains the proposed extension which carries the significant physical information collected from the GbE transmit interface and the attached link 1. PXC B updates the parameters with its own parameters (attenuation) and with information describing link 2 (length and attenuation). The message is then propagated to destination router C. Admission control module running on router C evaluates the received information and it decides if the connection can or can not be established. The permitted distance is below the maximum value and the receiver sensitivity admits the collected power value. In this case the lightpath request can be established and a RSVP RESV message will be sent back to the source node.
REFERENCES [1] J. Strand, A. L. Chiu, and R. Tkach, “Issues for routing in the optical layer”, Communications Magazine, Feb 2001. [2] L. Berger, “Generalized Multi-Protocol Label Switching (GMPLS) Signaling Functional Description”, RFC 3471, Jan 2003. [3] J. Lang, “Link Management Protocol (LMP)”, internet draft, draft-ietf-ccamp-lmp-10.txt, Oct 2003, work in progress. [4] M. Ali, D. Elie-Dit-Cosaque, and L. Tancevski, “Network optimization with transmission impairments-based routing”, ECOC 2001. [5] I. E. Fonseca, R. Almeida, M. Ribeiro, and H. Waldman, “Algorithms for FWM-aware routing and wavelength assignment”, IMOC 2003. [6] J. Strand and A. Chiu, “Impairments and other constraints on optical layer routing”, internet draft, draft-ietf-ipoimpairments-05.txt, May 2003, work in progress.
VIII. CONCLUSIONS In this study a novel approach to allow the implementation of large domains of transparency has been introduced. The solution is based on an extended version of the distrib-
IEEE Communications Society Globecom 2004 Workshops
373
0-7803-8798-8/04/$20.00 ©2004 IEEE
