Active versus passive OPS architectures for metro rings: a network dimensioning point of view

Optical Packet Switching is a promising technology for metro environments. We discuss two ring architectures (with/without active components allowing for spatial reuse) and compare them in terms of resources required for a given traffic demand. † This work has been supported by the European Commission through the IST-project DAVID (IST-1999-11387), and by the Flemish Govern-ment through the IWT GBOU-project “Optical Networking and Node Architectures”. C. Develder is supported as a Research Assistant of the Fund for Scientific Research – Flanders (F.W.O.–Vl.), Belgium. Introduction Next generation metro area networks (MANs) should provide high bandwidth in a flexible manner: they should efficiently exploit available resources, support multiple traffic types and offer rapid provisioning. Optical Packet Switching (OPS), with its packet-level granularity and hence efficient and flexible bandwidth sharing, fulfils these requirements very well [1]. In the European DAVID project [2], multiple MAN architectures are compared. Here, we outline the DAVID metro ring architecture and discuss two different MAN optical packet add/drop multiplexer (OPADM) designs: a Passive one, and an Active one. This paper focuses on the impact of these design choices on the resources needed to build a MAN network interconnecting a given set of nodes, with a given traffic demand from one node to another. MAN ring architectures In the DAVID concept, sketched in Fig. 1, the MAN comprises slotted WDM rings collecting traffic from several optical packet add/drop multiplexers (OPADMs). Rings are interconnected by a buffer-less Hub, which also provides access to a backbone (WAN). The rings constitute a shared medium, requiring a medium access control (MAC) protocol [3] to arbitrate access to the slotted channels. One wavelength, λc, is a dedicated control channel. Fig. 1: Network architecture and Passive OPADM DAVID proposes two OPADM architectures. The first one limits the use of advanced optical technologies, choosing commercial and mature ones instead [4]: it uses couplers and off-line filters to minimize physical cascadeability issues. The structure of this Passive OPADM is depicted in Fig. 1. The wavelength spectrum is separated for upstream (transmitters, Tx) and downstream (receivers, Rx): the Hub will perform conversion from Tx to Rx spectrum. The second, Active OPADM proposal of Fig. 2 — considered as longer term approach— allows an incoming packet to be erased from the ring, and to replace it with a new one. Because of this erasing capability, there is no need for spectral separation of Rx and Tx signals. This also allows for spatial reuse: whenever the path from source to destination does not cover the whole ring, the same wavelength can be re-used, as for A-C and D-E in Fig. 2. To limit the tuneability range of the Rx/Tx elements, a waveband concept is introduced: a Rx/Tx board provides access to a set of only B wavelengths (with one Rx/Tx per band). Fig. 2: Network architecture and Active OPADM A network dimensioning point of view The objective of this paper is to compare the architectures in terms of the amount of resources (which will to a great extent dominate the CAPEX) required to set-up a given demand between a given set of MAN nodes. Therefore, we developed a network planning algorithm starting from an ILPformulation of the planning problem. Yet, the many degrees of freedom hamper the finding of ILP solutions within reasonable time. Hence, we provided heuristic solutions using a tabu-search approach to find the minimal number of resources needed to fulfil a given traffic demand. The cost indicators used are the following: (i) Rx/Tx capacity: the total number of Rx/Tx elements used, summed over all OPADMs, (ii) link capacity: the number of wavelengths effectively used per link, summed over all physical links, (iii) nr. of lambdas: the number of wavelengths used per ring, summed over all rings. The first criterion is an indicator of the OPADM costs, while the last will impact the Hub dimension and thus its cost. Note that this dimensioning study is only a single (but quite important) facet of an in-depth assessment of the pros and cons of Active and Passive architectures. This paper therefore is to be complemented with e.g. studies on the architectures’ capabilities to deal with dynamic traffic in a network with given amount of resources, as eg. in [3]. Set-up of the case study To assess the resource requirements of the OPADM architectures, we covered a wide range of demand patterns. The demand patterns are the following (where D[i,j] denotes the bandwidth required between OPADMs i and j): (i) Uni: a uniform demand pattern, where between each two OPADMs a bandwidth d needs to be set-up (D[i,j]=d); (ii) Serv: there is one server node s, which dominates the demand matrix (D[i,s]=D[s,i]=2d, other D[i,j]=d); (iii) Neigh3: each node only communicates to 3 other nodes (D[i,i+1]= D[i,i+2]=D[i,i+3]=d, rest is zero); (v) David: a demand matrix based on real-life traffic, provided by the operators participating in DAVID. The impact of space reuse The main difference between the active and passive architectures from a conceptual point of view is the space reuse capability of the Active structure. Fig. 3 presents dimensioning results of the dimensioning for Passive and Active with wavebands of a single wavelength. (Note that B=1 amounts to having no waveband concept; B>1 is discussed in the next section.) 0% 100% 200% 300% 400%