Research On Distributed Tive Access Control For Data-Intensive Networks

The data-intensive network not only is the important application in the next genera-tion network, but also is the key technology of the next generation network, the existing prototypes of the next generation network have developed their data-intensive network models. It is very attractive to current network services, especially the multimedia appli-cations, that the data-intensive network is linearly scalable over the growth of network data sizes. However, data access is unpredictable in this environment for the reasons of large-scale data storage and loosely coupled nature of such networks. Especially, the data transitions in data-intensive networks spend much cost, most of the application-s should be allocated to the servers which has their required data objects. Today, an increasing number of network applications require not only considerations of computa-tion capacity of servers but also accessibility for adequate job allocations. An effective and adaptive mechanism of access control in this environment can significantly improve the service performance and resource utilization of the data-intensive networks. This article studies the access control of data-intensive networks from five aspects which are resource balance, accessibility estimation, adaptive server deployment, adaptive data deployment and wireless access, respectively. The detailed work is as follows.1. The paper first considers the impact of resource balance on access con-trol. Considering ordinary centralized access control model, we denote the state of a server by the its resource load level vector, thus, the state of the network is a resource load level matrix. Consequently, the state space of the network has small sizes and the computation is reduced for the reason that the number of network states is irrelevant with the number of users. In this case, we model the access procedure in the network as a Markov model, to get the optimal access control policy which is based on the resource load level, we optimize the access control policies with the method of gradient iteration.2. Considering the popular model of data-intensive networks where the work nodes is separated with the data nodes, we use the client cluster based on IP to describe the users’ behaviors, and allocate the adaptive work node cluster to server the client cluster whose members have similar behaviors, thus, the front end of the network is divided into many relevant small-scale access networks. Then, we use the accessibility estimation and stochastic control to implement the access control mechanism for these small-scale networks.3. Since the data transitions in data-intensive networks spend much cost, we consider the adaptive data deployment between the work nodes and data n-odes to improve the data-intensive networks. We propose an asynchronous distributed algorithm with shared memories which performs the non-critical operations asynchronously, and performs the critical operations simultane-ously. Compared to general approximation optimal data deployment, this work increases the communication complexity, but reduces the time com-plexity by the times of the number of the data servers.4. In the wireless environment, the accessibility of the user is determined by the accessibility of the access network, not by the access latency of the data deployment after the user’s requirement arrives at the access network. Con-sequently, the spectrum sensing and prediction are important techniques in this environment where the spectrum is directly related to the access band-width. Since the sensing information vector is usually sparse in common cases, we introduce the compressive sensing algorithm to reduce the com-munication bottleneck of the collaborative sensing.5. The spectrum prediction based on the sensing information of a small amount of channels can obtain the better channel selection policy than the decision only on the sensing information when the sensing budget is inade-quate. If the channels evolve according to the independent and statistically identical two-state discrete time Markov chains, we can naturally model the prediction procedure as the partly observed Markov decision progress for the reason that the ordinary Markov model has the huge dimension and the sensing budget is inadequate for observing all the channels. The simple and robust one-step optimal policy which focuses on maximizing the immediate reward is optimal in the case of practical interest. We prove the optimality of one-step optimal policy under infinite horizon discounted and average criterions and the assumptions of multi-channel sensing and sensing error.The article propose a number of simulations for the above work. The simulation results show us that the proposed schemes have good access control effect on the data- intensive networks, consequently, the efficiencies of the proposed schemes are validat-ed.