Optical Switching and Networking 7 (2010) 12–27 Contents lists available at ScienceDirect Optical Switching and Networking journal homepage: www.elsevier.com/locate/osn Cost-effective heuristics for planning GMPLS-based transport networks Nabil Naas, H.T. Mouftah * School of Information Technology and Engineering, University of Ottawa, 800 King Edward Ave, Ottawa, Ontario, Canada, K1N 6N5 article info Article history: Received 28 January 2008 Received in revised form 28 August 2009 Accepted 29 August 2009 Available online 10 September 2009 Keywords: MILP formulation GMPLS transport network planning Multi-granular optical network RMGPA problem Heuristic optimization abstract With the ever-increasing traffic in WDM transport networks, the development of GMPLS (or multi-granular) transport networks becomes essential to avoid the cost explosion of OXCs. Much work has been devoted to the development of Multi-Granular Optical Cross- connect (MG-OXC) architectures and network design and planning methods. Extending these efforts here, we address a new problem of planning the GMPLS-based transport net- work by (1) considering the whole traffic hierarchy defined in GMPLS; and (2) allowing the bifurcation of multi-granularity traffic demands among different physical routes. We call such a problem the Routing and Multi-Granular Paths Assignment (RMGPA). The objective of the problem is to minimize the total weighted port count in the transport network. Due to the computational complexity of the problem, only very small-sized planning problems can be solved exactly through Mixed Integer Linear Programming (MILP) optimization. In this paper, we propose efficient heuristics that are capable of solving large-sized planning problems in a reasonable amount of time. © 2009 Elsevier B.V. All rights reserved. 1. Introduction WDM networks using wavelength routing by Optical Crossconnets (OXCs) have been widely recognized as the most feasible architectural solution for transport net- works [1,2]. However, increasing the number of wave- lengths (λs) per fiber to accommodate the exponentially growing Internet traffic along with the dramatic increase in both broadband and wireless applications will result in (1) an increased number of OXC ports, (2) increased OXC footprint, (3) increased OXC power consumption, (4) in- creased difficulty in managing wavelengths, and (5) more severe cross-talk requirements for WDM demultiplexers in OXCs [1,3]. In order to overcome these drawbacks, the number of ports in the switching fabric can be reduced by routing a group of consecutive wavelengths, called a waveband, together through a single port. Further port re- duction is possible if consecutive wavebands are grouped * Corresponding author. Tel.: +1 613 562 5800x2173; fax: +1 613 562 5664. E-mail addresses: nnaas@site.uottawa.ca (N. Naas), mouftah@site.uottawa.ca (H.T. Mouftah). again into one fiber and routed through a single port. Re- ducing the port count is also one of the main factors con- tributing to the cost reduction of OXC [4]. An OXC that uses this multi-granular switching concept is referred to as a Multi-Granular OXC (MG-OXC) [4]. With the introduction of Generalized Multi-Protocol Label Switching (GMPLS) [5], the multi-granular switch- ing concept becomes essential to maintain the cost and complexity of the OXCs at a reasonable level. GMPLS is equipped with the ability to provision multi-granularity flows, as depicted in Fig. 1, by labeling them as follows [6]. First, a sub-wavelength-switched path is labeled with its frame/cell/packet header or the serial number of the TDM time slot. Second, a wavelength-switched path is labeled with its wavelength number and is switched according to its wavelength number. Next, a waveband-switched path is labeled and switched as a whole according to its wave- band number. Moreover, a fiber-switched path is labeled and switched as an entity based on its fiber code (or ID) on the unidirectional physical link. However, once the multi-granular switching paradigm is taken into consideration, most of the existing Routing and Wavelength Assignment (RWA) design and planning methods become economically infeasible [3]. Therefore, 1573-4277/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.osn.2009.08.009