Resource management for virtualized networks

Network Virtualization has emerged as a promising approach that can be employed to efficiently enhance the resource management technologies. In this work, the goal is to study how to automate the bandwidth resource management, while deploying a virtual partitioning scheme for the network bandwidth resources. Works that addressed the resource management in Virtual Networks are many, however, each has some limitations. Resource overwhelming, poor bandwidth utilization, low profits, exaggeration, and collusion are types of such sort of limitations. Indeed, the lack of adequate bandwidth allocation schemes encourages resource overwhelming, where one customer may overwhelm the resources that supposed to serve others. Static resource partitioning can resist overwhelming but at the same time it may result in poor bandwidth utilization, which means less profit rates for the Internet Service Providers (ISPs). However, deploying the technology of autonomic management can enhance the resource utilization, and maximize the customers’ satisfaction rates. It also provides the customers with a kind of privilege that should be somehow controlled as customers, always eager to maximize their payoffs, can use such a privilege to cheat. Hence, cheating actions like exaggeration and collusion can be expected. Solving the aforementioned limitations is addressed in this work. \n \nIn the first part, the work deals with overcoming the problems of low profits, poor utilization, and high blocking ratios of the traditional First Ask First Allocate (FAFA) algorithm. The proposed solution is based on an Autonomic Resource Management Mechanism (ARMM). This solution deploys a smarter allocation algorithm based on the auction mechanism. At this level, to reduce the tendency of exaggeration, the Vickrey-Clarke-Groves (VCG) is proposed to provide a threat model that penalizes the exaggerating customers, based on the inconvenience they cause to others in the system. To resist the collusion, the state-dependent shadow price is calculated, based on the Markov decision theory, to represent a selling price threshold for the bandwidth units at a given state. \n \nPart two of the work solves an expanded version of the bandwidth allocation problem, but through a different methodology. In this part, the bandwidth allocation problem is expanded to a bandwidth partitioning problem. Such expansion allows dividing the link’s bandwidth resources based on the provided Quality of Service (QoS) classes, which provides better bandwidth utilization. In order to find the optimal management metrics, the problem is solved through Linear Programming (LP). A dynamic bandwidth partitioning scheme is also proposed to overcome the problems related to the static partitioning schemes, such as the poor bandwidth utilization, which can result in having under-utilized partitions. This dynamic partitioning model is deployed in a periodic manner. Periodic partitioning provides a new way to reduce the reasoning of exaggeration, when compared to the threat model, and eliminates the need of the further computational overhead. \n \nThe third part of this work proposes a decentralized management scheme to solve aforementioned problems in the context of networks that are managed by Virtual Network Operators (VNOs). Such decentralization allows deploying a higher level of autonomic management, through which, the management responsibilities are distributed over the network nodes, each responsible for managing its outgoing links. Compared to the centralized schemes, such distribution provides higher reliability and easier bandwidth dimensioning. Moreover, it creates a form of two-sided competition framework that allows a double-auction environment among the network players, both customers and node controllers. Such competing environment provides a new way to reduce the exaggeration beside the periodic and threat models mentioned before. More important, it can deliver better utilization rates, lower blocking, and consequently higher profits. \n \nFinally, numerical experiments and empirical results are presented to support the proposed solutions, and to provide a comparison with other works from the literature.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.