Applying Quality of Service Prediction in WDM Optical Networks Panagiotis G. Sarigiannidis Department of Engineering Informatics and Telecommunications University of Western Macedonia Kozani, Greece sarpan@uowm.gr Abstract—A dynamic prediction scheme is presented in this paper, named PROPHET. The purpose of the proposed technique is to predict, given two different classes of quality of service (QoS), the total amount of the demanded transmission requests per QoS class. PROPHET is constructed based on Hidden Markov Chains (HMC), modeled by an ergodic framework. The prediction objective is to reduce the amount of time spent in computing the transmission schedule by predicting traffic requests. The evaluation of the predictor is realized in a Wavelength Division Multiplexing (WDM) single-hop network with star topology. Furthermore, PROPHET is compared to a previous prediction-based scheme, called POSA. Simulation results indicate that the novel technique supports efficient predictable QoS, since it operates more accurately than POSA. Keywords-prediction; WDM networks; optical networks; scheduling. I. INTRODUCTION Network traffic prediction is an important and valuable issue in network management and configuration. Accurate predictions may lead to various benefits such as time benefit regarding transmission scheduling, utilization benefits in a medium exploitation manner, throughtput improvements in a network design etc [1]-[7]. Usually, most of the traffic prediction models do not take into account QoS matters, such as the differentiation of the packets’ priority. In this work, a novel prediction technique is presented, which supports QoS. The introduced PRiOritized Prediction Hidden-chained Ergodic Technique (PROPHET) constitutes a novel way of introducing estimation procedure by applying a distributed Hidden Marcov Chains (HMCs) [8]-[9], in order to predict the upcoming nodes’ requests per service class. The suggested scheme is evaluated into a single-hop WDM optical network with star topology. A reservation-based Multiple Access Control (MAC) protocol is applied in order to address the channel allocation process [10]-[11], by creating a transmission program, known as transmission schedule. PROPHET tries to predict the traffic requests that each node sends to all other in order to coordinate the transmissions without any collisions. The main purpose of the PROPHET framework is to provide the (predicted) data requests in advance, enabling the overlapping between the schedule computation phase and the data transmission phase. In this way, the schedule computed during the current data phase (let it be frame f) is used for the next data phase (frame f+1). The result of this overlapping is the reduction of the schedule computation time, earning less spent control time, solving a crucial problem for optical access networks. The advantage of PROPHET scheme is that supports QoS, by predicting separately the real- and non-real traffic requests. In this context the performance of the prediction modules is counted in terms of prediction accuracy. In other words, the prediction accuracy emerges from the portion of the number of successful predictions divided by the total predictions. Simulation results indicate that the novel QoS- based prediction framework supports more accurate estimations than the previous POSA scheme [12]. The rest of this paper is organized as follows. Section II describes the network background, Section III introduces the PROPHET framework, Section IV shows and makes remarks about the simulation results, and finally Section V concludes the letter. II. NETWORK MODEL In the adopted WDM star network, a set of N nodes are interconnecting to a passive star coupler via two-way optical fibers [13]-[15]. Each node is equipped with a tunable transmitter and a fixed receiver in order to send data packets on any available channel and to accept data on a specific channel, known as home channel. The network is supported by W data channels of the same bandwidth, while there is no separate control channel. For a more realistic scenario it is considered that the number of nodes is larger than the number of channels (N > W), hence N/W nodes share the same home channel. The system is synchronous and so the network operates in a slotted mode with a timeslot equal to the transmission time of one fixed-length packet. Each node maintains W queues, one queue per channel, with upper limit equal to K (fixed-length packets) in order to simulate a deterministic prediction model. In this way, a request, consisting of a variable-length packet, is generated at a node and it is sorted in the corresponding queue according to the destination’s home channel. A scheduling algorithm functions in a distributed manner, so each node broadcasts control packets to all channels in a TDM-based fashion. This phase is called reservation phase. Next, the scheduling algorithm calculates the transmission schedule and the data phase is followed. During the data phase nodes apply the common schedule and transmit their packets on the appropriate channel. At the end of the data phase, the current