Light-trails are a promising candidate for grooming of traffic in the optical layer. We investigated the performance of light-trails via theory, simulation and experiment focusing on its efficiency. Introduction The growth of IP communication induces network dynamism that requires the optical layer to provide a high degree of re-configurability. Reconfigurable networks can be further classified as: being able to juxtapose a virtual topology on a given physical topology and to provide dynamic bandwidth allocation to nodes on demand. Metro ring networks represent a strong application case for reconfigurable networking. Rings are popular due to their resiliency and management ease. With the advent of new IP-based services, peer-to-peer traffic has substantially grown and is expected to grow in the future. Such peer-topeer traffic when provisioned through lightpath based optical networks does not fully utilize the available bandwidth. Peer-to-peer traffic is particularly characterized by dynamic, sub-lambda granularity and service provisioning based on unicast and multicast requirements. The recently proposed concept of light-trails [1-3] provides a solution that enables sub-lambda communication at the optical layer while allowing for dynamic bandwidth allocation and facilitating unicast and optical multicast traffic. Light-trails allow wavelength bandwidth to be timeshared amongst multiple nodes in an efficient way. A network based on light-trails is defined as Shared Wavelength Optical Network (SWON) and was proposed in . Such a network is a good candidate for emerging Ethernet aware optical transport. In this paper, we study performance of light-trails and quantify relations between the bandwidth and efficiency. We then report results from our light-trail test-bed. Light-trail based SWON Design Considerations A light-trail is a generalization of a lightpath such that multiple nodes can communicate with one-another along the path. Due to the combination of a unique nodal architecture and an out-of-band (OOB) control protocol, a light-trail allows communication between nodes by time-sharing of a wavelength without optical switching. The node architecture shown in Fig. 1 has the capability to drop and continue as well as passively add a wavelength channel. These two features enable a light-trail to be analogous to an optical bus. In a light-trail, the out-of-band control channel enhances the known properties of an optical bus. Communication between nodes within a light-trail is done through the establishment of connections (using the OOB channel), which do not involve any optical switch configuration and hence can be dynamically setup and torn down. Creation, deletion and extension/contraction of a light-trail however does involve optical switching of the wavelength. A lighttrail based network exemplifies an embedded reconfigurable virtual topology  that provides spatial wavelength reuse of non-overlapping light-trails. Light-trails can be setup using the GMPLS signaling scheme . Connection set up and tear down is done using OOB signaling. Due to the drop and continue aspect of node architecture, a light-trail readily provides for optical multicasting. Dynamic provisioning of connections over a light-trail, sans switching is a key precursor to peer-to-peer traffic. In order to avoid collision at any given time only one node actually transmits data in a light-trail (though there can be several destinations at the same time). The light-trail solution provides an efficient method for optical traffic grooming as shown in . To facilitate such dynamic setting up and tearing down of connections over a light-trail we make use of burstmode optics. A light-trail has two constraints for functioning. Firstly, the total input bandwidth averaged over time and summed over all the nodes must be less than the bandwidth provided by the wavelength on which the light-trail resides. This can be understood from a multipoint flow model, where the total flows into a hose have to be less than the output rate or would result in congestion. Secondly, to physically function there must be a guard band between two successive connections over a light-trail. Naturally, it is desired that the guard band is small to increase light-trail efficiency. Coupler Lightpath: new wavelength for each connection Single – and Multi-casting using light-trails: creating sub-lambda communication over single wavelength Convener Node End Node Optical combiner or splitter Node Architecture Optical ON/OFF Switch Rline Rin Rin Rin Rin Fig. 1 Light-trail based SWON architecture The contribution of this article is to deduce light-trail efficiency and derive the relation between the constraints for light-trail communication and verify these through simulation and experiments. For this purpose, we define the following parameters assuming round-robin type scheduling of connections within a light-trail. Tg: the guard time between two successive transmissions. Tslot(i): the connection duration made available to the i node; Rin(i): the input rate in bits/sec at the i th node, which is bandwidth flow into a node; Rline: the line-rate or light-trail bandwidth in bits/sec, and k: number of nodes in a light-trail Since nodes have to wait to provision their connection as other nodes seek the light-trail bandwidth, the criteria for finite queueing in a light-trail is given by (1).