شبکه سیاست های مدیریت منابع برای بار متعادل و صرفه جویی در انرژی با استفاده از نظریه صف تعطیلات

The resource management is the central component of grid system. The analysis of the workload log file of LCG including the job arrival and the resource utilization daily cycle shows that the idle sites in the Grid are the source of load imbalance and energy waste. Here we focus on these two issues: balancing the workload by transferring jobs to idle sites at prime time to minimize the response time and maximize the resource utilization; power management by switch the idle sites to sleeping mode at non-prime time to minimize the energy consume. We form the M/G/1 queue model with server vacations, startup and closedown to analysis the performance metrics to instruct the design of load-balancing and energy-saving policies. We provide our Adaptive Receiver Initiated (ARI) load-balancing strategy and power-management policy for energy-saving. The simulation experiments prove the accuracy of our analysis and the comparisons results indicate our policies are largely suitable for large-scale heterogeneous grid environment.

A Grid [1] is a very large scale, generalized distributed network computing system that can scale to Internet-size environments with machines distributed across multiple organizations and administrative domains. The emergence of a variety of new applications demands that Grids support efficient data and resource management mechanisms. Resource Management [2] is central component of a grid system. Its basic responsibility is to accept requests from users, match user requests to available resources for which the user has access and schedule the matched resources. Applications may request resources from the Grid. Such resource requests are considered as jobs by the Grid. Depending on the application, the job may specify quality of service (QoS) requirements. The resource management is required to perform resource management decisions while maximizing the QoS metrics delivered to the clients [3]. With the Grid becoming a viable high-performance alternative to the traditional supercomputing environment, various aspects of effective Grid resource utilization are gaining significance. With its multitude of heterogeneous resources, a proper scheduling and efficient load-balancing across the Grid is required for improving the performance of the system. Due to uneven job arrival patterns and unequal processing capacities and capabilities, the processors in one grid site may be overloaded while others in a different grid site may be under-utilized. It is therefore desirable to dispatch jobs to idle or lightly loaded site in the grid environment to achieve better resource utilization and reduce the average job response time. This is a natural extension of the existing work on load-balancing in a traditional distributed system. Grid computing tends to push high performance. Unfortunately, the “last drop” of performance tends to be the most expensive; that is, the last 10% increase in performance requires a disproportionally large amount of resources. The Earth Simulator, one of the world’s fastest supercomputers with 640 computing sites, consumes 7 MW of power [4]. In particular, energy consumption – and the resultant heat dissipation – is becoming an important limiting factor; reducing energy saves money and increases reliability, among other things. In this paper, we focus on the two aspects of resource management of grid environment: load-balancing and energy-saving. By analyzing the Large Hadron Collider Computing Grid (LCG) log file [5], we learn the statistical properties of the job arrival daily cycle which shows that there is an activity peak around midday. Even at that period, there are more than 10% sites remain idle which is source of load imbalancing. Meanwhile, few jobs arrive at non-prime time and at least 40% sites are idle which causes the energy consumption becoming critical concerns. Queuing systems in which the server works on primary and secondary customers arise naturally. As far as the primary customers are concerned, the server working on the secondary customers is equivalent to the server taking a vacation [6]. The vacation can be explained as idle sites receiving load for load-balancing or switching the idle sites to sleeping mode for energy-saving. We form the M/G/1 queue model with server vacations, startup and closedown to analysis the performance metric. We present our policies based on the analysis. Adaptive Receiver Initiated (ARI) load-balancing strategy for grid environment considers the job migration cost, resource heterogeneity and network dynamics when load-balancing is considered. Power management is energy-saving policy which explore the tradeoff between the site energy consume and QoS request satisfy. The simulation experiment proves the performance of our policies. The rest of the paper is organized as follows: In Section 2, we will present the related work on load-balancing and energy-saving in recent year. In Section 3, we analyze the LCG log file. In Section 4, we develop a novel queuing analytical model. In Section 5, we introduce our two polices and give the experiment results in Section 6. Finally, Section 7 concludes the paper.

In this paper, we first analyze of the LCG log file. The job arrival daily cycle and the fraction of resource state introduce the problems of load-balancing and energy-saving. Queuing system with vacation model can be used to describe the utilization of idle fraction of a site for different purpose. A tradeoff is induced between the site idle fraction utilization and the job QoS requirements. Here, we develop a novel queuing analytical model with server vacations, startup and closedown to deduce the performance metrics which enable us to explore the tradeoff. ARI is a modified version of RI in which we consider the job migration cost, resource heterogeneity and network dynamics when load-balancing is considered. Power-management policy will switch the idle sites to sleeping mode at non-prime time for energy-saving considering the QoS request of new arrival jobs. The simulation results show that ARI surpasses other exist load-balancing strategies especially the case for large-scale heterogeneous grid environment. The power management policy will switch most idle sites to sleep at non-prime time with acceptable response time. In the future, we will extend this work by providing fault tolerance into the resource management system as fault tolerance is a very important characteristic for any distributed systems.