Dimensioning of Multi-Granular Optical Networks

We present an ILP model for dimensioning optical networks that support wavelength and sub-wavelength switching. Results indicate significant reductions in cost and node-complexity with such multi-granular networks. Introduction Future networks will likely consist of some form of alloptical networking, implying data remains in optical form end-to-end. An apparent difficulty is that a single infrastructure must support a wide range of application requirements and their associated connection constraints. It is however well known that performance can become unacceptable when the switching granularity is insufficiently adapted to the traffic demands. An example is sending burst or packet-sized data on wavelength-granular switches. We therefore propose to offer switching at wavelength (circuit) and sub-wavelength (burst or even packet) level, a concept generally referred to as multi-granular switching. Previous work has demonstrated that multi-granular optical switching can improve bandwidth utilization. In this paper, we explore the possible cost and complexity savings when multiple switching technologies are incorporated, and compare them to more traditional, single-technology solutions. To this end, we propose a dimensioning algorithm to optimally install switching ports in the network. Results are presented to investigate total installation cost and node complexity of the network infrastructure. Multi-Granular OXC An essential device is the multi-granular optical crossconnect (MG-OXC). This device is in general composed of several different switching fabrics to support a wide range of switching speeds. In [1], the authors describe the design and architecture of an MG-OXC based on the combination of a slow and cheap MEMS-based switching block, and a fast but expensive SOA-based switching fabric. A number of key design choices are identified, and their effect on scalability and reconfigurability are studied. Of special importance for the study presented in this paper, is the distinction between a sequential and a parallel MG-OXC design. In brief, in the parallel approach, switching blocks are placed in parallel, which offers good performance with respect to scalability. The sequential approach, on the other hand, connects the output of slow switching fabric to the input of the fast switching block. In this way it becomes possible to reconfigure which wavelengths have access to the fast switching block, at an increase in number of slow switching ports. Furthermore, in [2] we have studied the performance behaviour of a single OXC through simulation, and showed that even a small number of fast and expensive wavelength ports are sufficient to provide significant performance improvements. In this paper, we present an ILP-based algorithm to dimension a network composed of MG-OXC, by incorporating the price ratio of fast over slow switching fabrics. Network Design Model Suppose the network is defined by the directed graph G(V,E), V the set of nodes, and E the set of directed links. Each wavelength has a fixed bandwidth B, identical for all links and wavelengths. Traffic between source s and destination d is represented by !sd, and C is the cost ratio of fast over slow switches. Finally, the possible paths between source s and destination d are determined in advance, and captured by the boolean parameters ! pl sd =1 iff link l is part of path p between source s and destination d, 0 otherwise. Furthermore, the following variables are introduced: • !p sd =1iff demand (s,d) uses path p, 0 otherwise • !sd =1iff demand (s,d) is switched fast, 0 if slow • integer variables x l and yl , representing the number of slow and fast wavelengths on link l We now state the problem as a non-linear model, in which the last constraint ensures that only single-path routing is used between source and destination. !(s,d), p : xp sd " #sd$p sd (1%& ) !l : xl = " pl sd x p sd