A QoS-Oriented Resource Availability Evaluation Model In Computational Grids

Grid computing has emerged as the next-generation parallel and distributed computing methodology that aggregates dispersed heterogeneous and diverse global resources for solving various kinds of large-scale applications in science, engineering, and commerce. In such a decentralized, dynamic and autonomous environment, providing non-trivial QoS for end users is a major challenge. To address this problem, many mechanisms such as advanced reservation and QoS negotiation have already been presented. Howerver, none of past research efforts have been devoted to the following issues:(1) Due to software/hardware failure, security problem or usage policy defined by resource provider or VO (Virtual Organization), Grid resource may be unavailable during a period of time. Thus, what is the resource's next state according to its historical trace needs to be solved. (2) The knowledge of resource capability when it is serving for end users should be got before task scheduling. Hence, how to evaluate resource capability while considering its varying workload and heterogeneous metrics should be resorted.With the aim of providing acceptable QoS in Computational Grids under Service Oriented Architecture (SOA), we present an availability evaluation model to provide effective and accurate resource information for QoS-oriented task scheduling. Listed below are our original contributions.1. Resource availability degree evaluation:towards availability-dgree enhanced job execution service deploymentFirstly, resource availability degree is defined and quantified from application perspective. We analyze that noise exists in availability degree historical trace got from currently measurement methods. Then, a Model for Resource Availability Prediction Based on Grey Model (MRAPGM) is presented. In MRAPGM, noise is filtered out from resource availability degree historical data employing wavelet analysis, and then Grey model is utilized to predict resource availability degree during a future period. We conduct extensive experiments to determine the wavelet and related parameters in MRAPGM. On the basis of prediction of resource availability degree and replication technique, a definition called Availability-degree Enhanced Job Execution Service (AEJES) and its constructing method is proposed. A Availability-degree Enhanced Model for Job Execution (AEMJE) is derived from AEJES. The model gets a balance between resource provider's profits and user's QoS guarantee, which not only avoids system performance degradation due to inappropriate replication number, but also satisfies user's QoS requirement.2. Queue-theory-based dynamic resource capability evaluationA Resource Serving Model based on Queuing Theory (RSMQT) and performance metrics for characterizing resource or resource set capability are presented. Orienting QoS requirements, a Resource Performance Evaluation Mechanism based on Queuing Theory (RPEMQT) is derived from RSMQT. In RPEMQT, resource performance metrics are monitored and traced, then resource or resource set dynamic capability is evaluated from user's and provider's perspective repectively. Theoretical analyses and simulation results show that Queueing Theory based resource serving model is suitable to describe the working model of grid resources, and utilizing the proposed resource evaluation mechanism in resource selection can provides enhanced satisfaction for end users.3. Resource filtering policy based on resource capability evaluationBased on AEJES capability metrics derived from dynamic resource evaluation, an adaptive and dynamic AEJES organization topology is proposed using PSO-based clustering algorithm. The AEJES services with similar or same QoS are gathered into one Logical Service Cluster (LSC) to decrease time complexity of task scheduling. Then, a service filtering algorithm is proposed in light of service clustering. We validate impact of similaritis between services in LSC on efficiency of service selection, and determine some parameters such as optimal LSC number, clusering period through extensive simulations. The experiment results also show that the time complexity for prior well-established task algorithm reduces significantly while using our LSC.4. Resource co-allocation and task scheduling based on resource availability evaluationOn AEJES candidate set obtainied from service filtering, we describe different players on behalf of their profits as a Game in AEMJE, and introduce a co-allocation scheme. Furthermore, a QoS-guaranted workflow scheduling algorithm is proposed based on the new scheme. The experiment results show that the task scheduling algorithm provide better QoS level while employing resource availability evaluation.We construct a model for evaluating and quantifying resource availability degree and its dynamic capability, which provides efficient and accurate resource availability information for QoS-constraint task scheduling. Our theoretically sound and practically feasible model provides more satisfied QoS for commercial uses than previous works, which promotes the newly emerged grid applications to business mode.