Vasiliki L. Kakali

A Novel Adaptive Framework for Wireless Push Systems Based on Distributed Learning Automata

A novel adaptive scheme for wireless push systems is presented in this paper. In this wireless environment two entities play the most important role: the server side and the client side that is connected to the system. The server side is responsible to broadcast an item per transmission in order to satisfy the clients’ requests. The performance of the server side depends on item selections. Hence, the server broadcasts an item and the clients are satisfied if the transmitted item was the desired one. In this work, a set of learning automata try to estimate the client demands in a distributed manner. More specifically, an autonomous learning automaton is utilized on each client group, since the clients are gathered into groups based on their location. The output of each automaton is combined in order to produce a well-performed transmission schedule. Concurrently, a round robin phase is adopted, giving the opportunity to the non-popular items to be transmitted. In this manner, the various client demands are treated fairly. The introduced technique is compared with a centralized adaptive scheme and the results indicate that the proposed scheduling framework outperforms the centralized one, in terms of response time and fairness.

On Performance Improvement of Wireless Push Systems via Smart Antennas

In environments with locality of client demands, the use of multiple directional antennas at the Broadcast Server has been shown to increase performance. In many cases however, such broadcasting systems fail to exploit the full potential of the multiple antennas as they do not take into account the geographical distribution of clients within the coverage area of the system. This letter proposes an adaptive smart antenna-based wireless push system where the beamwidth of each smart antenna is altered based on the current placement of clients within the system area. Coupled with a modification of the broadcast schedule, the proposed approach significantly increases the performance observed by the system clients.

A New Class of Wireless Push Systems

Data broadcasting is an efficient way of delivering information over asymmetric wireless environments, and push systems can provide high scalability and client hardware simplicity. Many environments, however, are characterized by a priori unknown client demands and groups of clients that are located at the same region and have similar demands. The main disadvantage of the approaches in such cases is the lack of fairness, because groups with few members have much lower performance than groups with many clients, because the performance per group (directly) depends on the group size. In this paper, we propose a fair push system where the performance of each client is independent of the total number of clients that are located in the same region. This condition is achieved without the overall performance of the system being significantly affected. Moreover, the proposed fair system is extended to a priority-based system that multiplies the performance of a specific group, depending exclusively on its priority level in relevance to the rest of the groups.

A novel HMM-based learning framework for improving dynamic wireless push system performance

Abstract A new machine learning framework is introduced in this paper, based on the hidden Markov model (HMM), designed to provide scheduling in dynamic wireless push systems. In realistic wireless systems, the clients’ intentions change dynamically; hence a cognitive scheduling scheme is needed to estimate the desirability of the connected clients. The proposed scheduling scheme is enhanced with self-organized HMMs, supporting the network with an estimated expectation of the clients’ intentions, since the system’s environment characteristics alter dynamically and the base station (server side) has no a priori knowledge of such changes. Compared to the original pure scheme, the proposed machine learning framework succeeds in predicting the clients’ information desires and overcomes the limitation of the original static scheme, in terms of mean delay and system efficiency.

An adaptive energy-efficient framework for time-constrained optical backbone networks

Summary Power management has emerged as a challenge of paramount importance having strong social and financial impact in the community. The rapid growth of information and communication technologies made backbone networks a serious energy consumer. Concurrently, backbone networking is deemed as one of the most promising areas to apply energy efficient frameworks. One of the most popular energy efficient techniques, in the context of backbone networks, is to intentionally switch off nodes and links that are monitored underutilized. Having in mind that optical technology has thoroughly dominated modern backbone networks, the function of switching off techniques entails fast operation and rigorous decision-making because of the tremendous speed of the underlying optical media. This paper addresses this challenge by introducing a novel, adaptive, and efficient power management scheme for large-scale backbone networks. The proposed framework exploits traffic patterns and dynamics in order to effectively switch off the set of network entities in a periodic fashion. An adaptive decision-making algorithm is presented to maximize the network energy gains with respect to time constraints as well as QoS guarantees. The conducted simulation results reveal considerable improvements when applying the proposed framework compared with other inflexible energy efficient schemes. Copyright

IFAISTOS: A fair and flexible resource allocation policy for next-generation passive optical networks

In modern, competitive, and dynamic access networks the underlying bandwidth distribution mechanism has to be capable of understanding user requirements, meeting stringent quality of service (QoS) demands, and satisfying a broad spectrum of user traffic dynamics. Undoubtedly, optical fiber is the dominant transmission medium enabling practical and cost-effective optical infrastructures in the last mile. Passive optical networks (PONs) represent one of the most promising player towards the fiber to the home (FTTH) vision allowing users to experience high quality, demanding multimedia services and applications. The 10-gigabit-capable passive optical network (XG-PON), one of the latest PON standard, incorporates a set of profound conditions a contemporary PON should ensure. Fairness provisioning constitutes one of the most critical features a PON should provide. However, ensuring fairness in an access network with numerous different users, requesting multiple traffic flows in any time, is not a straightforward task. In this work, we focus on the fairness issue by devising an adaptive, efficient, and fair dynamic bandwidth allocation (DBA) scheme called Insistent FAIr STrategy prOcesS (IFAISTOS). IFAISTOS investigates and maintains user traffic profiles. Overloaded users are carefully treated by gaining greater granting windows than other users; however bandwidth monopolization is prevented. Fairness is ensured for all users in terms of traffic load and average delay. A steering, adaptive mechanism records user traffic profiles by changing and defining bandwidth weights proportional to individual traffic needs. Extensive simulation results reveal the efficacy of the proposed DBA in terms of fairness and average packet delay.

On forecasting the ONU sleep period in XG-PON systems using exponential smoothing techniques

Power management has been advanced on a crucial factor in the design of modern access networks. Furthermore, the proliferation of optical networking in the last mile led major Telecom unions, such as the International Telecommunication Union (ITU), to emerge energy consumption as a critical objective of the next generation passive optical networks (NG-PONs). In particular, the standardization of the 10-gigabit-capable PON (XG-PON) entails well-defined specifications towards power management and energy reduction, especially regarding the power control of optical terminal devices such as the optical network units (ONUs). In this way, the optical line terminator (OLT) along with ONUs are able to cooperate with each other in order to succeed energy reduction, by applying doze or cyclic sleep periods to idle ONUs. However, the sleep period determination remains a quite challenging research area. In this study, we endeavor to provide XG-PON networks with an effective forecasting mechanism that is capable of estimating the time duration of the forthcoming sleep session. To this end, we apply the exponential smoothing technique to best estimate the sleep duration based on the monitoring time series observations. The obtained evaluation results sound quite promising, since the proposed model accomplishes to advance the trade-off between the energy reduction and network efficiency.

Simulation techniques for evaluating energy-efficient heuristics for backbone optical networks

Energy efficiency in backbone optical networks can reduce the carbon footprint while preserving performance levels. When embedded in computational logic, core functional traits like lightpath routing,will be performed aiming at better resource utilization from an energy perspective. Simulating the procedure of energy-aware lightpath routing and establishment by means of software tools can lead to decisions about the characteristics a heuristic method should include to fulfil its purpose, i.e., reduce energy consumption. Also, elaborate comparison and evaluation between heuristic methods can be achieved through simulation. The main internal parts of software simulator design and implementation will be presented in this chapter, along with network environment parameters that affect their performance. Finally, an energy-efficient heuristic method is designed, implemented and simulated with the purpose of reducing energy consumption by adhering to all presented simulation principles.

An adaptive power management scheme for Ethernet Passive Optical Networks

Undoubtedly, energy consumption in communication networks poses a significant threat to the environmental stability. Access networks contribute to this consumption by being composed of numerous energy inefficient devices and network equipment. Passive Optical Networks (PONs), one of the most promising candidates in the field of access networking, should avoid this bottleneck in the backhaul power consumption by lowering the energy use of the optical devices. In this paper, we move towards that direction by introducing an energy efficient power management scheme that encompasses two major goals: a) to reduce the energy consumption by allowing the optical devices to enter the sleep mode longer, and b) to concurrently maintain the network performance. To this end, we focus on the energy consumed by the optical network units (ONUs). The intelligence of the ONUs is stimulated by enhancing the decision making in determining the duration of the sleep period with learning from experience mechanism. Learning automata (LAs) are charged to address this challenge. The evaluation of the proposed enhanced power management scheme reveals considerable improvements in terms of energy savings, while at the same time the network performance remains in high levels.

ARES: An Adaptive, Resilient, Estimation Scheme for Enforcing Bandwidth Allocation in XG-PON Systems

Passive Optical Networks (PONs) constitute the dominant architecture in the last mile that effectively realize the Fiber To The Home/Building/Curve (FTTH/B/C) paradigm. It combines a cost-effective infrastructure with an effective data delivering, where multiple users are able to use high-quality services. The latest new generation PON (NG-PON) standard, known as 10-gigabit-capable passive optical network (XG-PON), stands a very promising framework that incorporates 10 Gbps nominal speed in the downstream direction. In the opposite, all users have to share the upstream channel, where multiple upstream traffic flows are delivered to the Central Office (CO), using a channel of 2.5 Gbps rate. Having in mind that in dense, urban areas the number of users is quite large, an efficient Dynamic Bandwidth Allocation (DBA) scheme is mandatory to guarantee unhindered high-quality service delivery. In this work, a resilient coordination scheme is presented that intends to ensure high-efficient traffic delivery under pressing traffic conditions. In order to achieve that, a sophisticated machine learning model is proposed that coordinates the Optical Networks Units (ONUs) based on their traffic profile. The proposed, Adaptive Resilient Estimation Scheme (ARES), contributes in a twofold way. First, it succeeds to provide balanced resource allocation, under heavy traffic circumstances, by isolating idle ONUs. Second, it manages to effectively adjust the amount of fixed bandwidth allocated to Alloc-IDs based on their traffic behavior. Simulation results demonstrate that ARES offers considerable improvements in terms of average upstream packet delay and traffic received, while the estimation accuracy attains at high levels.