Research Of Energy-efficient Scheduling And Resource Management On Cloud Data Centers

With the development of Cloud Computing, the cloud data center is also becoming important.It is crucial to build the cloud data center that provides the high-quality, safe and reliable serviceto the dramatically increasing demands, so the cloud data center has been drawing more attentionfrom both industrial and academia researchers. Nowadays, the data center usually consists of t-housands of computing nodes (such as servers) and related equipments (such as network, coolingsystem). However, as the increasing of demands, the scale of data center is also growing. There-fore, cloud data center is strongly related to the people’s life, it is equivalent to an electric powerplant in our daily life.Although we have been benefiting a lot from the data centers, there are still many problemsas the increasing scale of the existing data centers. First, due to the variety of service requests, toimprove the performance of data center, the new systems (reconfigurable systems) is also appliedto data centers. However, the energy consumption in such system is extremely high. It requiresa higher cost to service providers, so it becomes an important concern. Second, high energy con-sumption is one of the most critical problems in data centers. This issue is not only making thedata center service provider to bear the huge cost (e.g., electricity cost, and other expenses), butalso raising the issue of high carbon emission rate that causes environmental pollution. To buildthe green data centers can greatly help to alleviate this problem, but green energy (such as solar,wind) is difficult to be predicted, it is unsustainable to ensure the supply of energy. In addition, thehigh energy consumption can also cause the reliability problem of the computing nodes. With therising of the temperature, the computing nodes in data centers are more vulnerable, so it is easy tocause the failure of computing nodes during the operation that resulting in data loss and leading tohuge economic consequences. In addition, Third, due to the enlarging of the data center scale, inorder to meet customers’ service needs, it is critical to design a resource management approach formaximizing service provider profit by appropriate placing requests.This thesis is motivated by these concerns. The following problems are addressed: energy-efficiency on multi-FPGA based systems on data center, green energy-aware efficient schedulingand reliability-aware scheduling on data center, efficient resource management on geo-distributed cloud data centers. The problems can be specifically as follows:(1) since the the reconfigurablesystem applications is introduced in the data center, this thesis also proposes a low energy con-sumption algorithm for the multi-FPGA based reconfigurable systems applied in data centers.(2)proposes a data center network-based energy-aware scheduling algorithm, this algorithm takes in-to account the energy loss that mainly draws from the data center network communication;(3)the reliability of computing nodes is associated with the temperature, proposes a reliability-awarescheduling algorithm. This algorithm not only can make advantage of both the data center tomaximize the use of solar energy, but also ensure the reliability of computing nodes;(4) for thegeo-distributed cloud data centers, this thesis presents an Ant Colony Optimization based resourcemanagement scheme to maximize the service provider profit by satisfying the customers demands;The main contributions of this thesis are as follows:This thesis studies the crucial problem of energy-efficient scheduling for reconfigurable sys-tems with multiple FPGAs. Several factors make the energy efficient scheduling particularlychallenging, including spatial allocation constraint, reconfiguration overhead, limited recon-figuration ports, and deadline satisfaction. These unique characteristics make energy effi-cient scheduling in multi-FPGA reconfigurable systems particularly challenging and none ofexisting solutions can be directly applied. This thesis takes on this challenge and proposesan energy-efficient scheduling algorithm called AEE based on ant colony optimization formulti-FPGA reconfigurable systems. A task placement scheme is devised which serves asthe heuristic function that derives the minimum global makespan, which is important to theant colony algorithm based proposed in the thesis. The scheme takes into account reconfig-uration overhead and places tasks for reducing the overall overhead. Then, based on AEE,an enhanced algorithm (eAEE) is devised to deal with the tasks with precedence and inter-dependencies. To evaluate the effectiveness of the two proposed algorithms, comprehensivetrace-driven simulations have been conducted and compared with other state-of-art algo-rithms. Experimental results demonstrate that AEE can successfully complete tasks withoutviolating deadline constraints and the energy dissipation is largely reduced. Also, eAEEconsumes energy less than an improved simulated annealing algorithm (iSA) with a largeproblem scale.It is highly challenging for datacenters to make use of green energy. First, the availability oftypical green energy is variable to dynamic changes of natural environments, e.g., weather.Second, although predictions can be made for the future availability of green energy, it isinevitable that such predictions have errors. Third, jobs are associated with strict deadlinesand it is required that jobs are completed before their deadlines. Fourth, since the reliabilityin a datacenter relies upon temperature, the awareness of temperature should be taken into account while maximizing the green energy. In this thesis, we consider online scheduling ofjobs whose arrivals to the datacenter system dynamically. In addition, we explicitly take thepower consumption of switches into account when scheduling jobs onto computing nodes.Two solar energy-aware algorithms called SEEDMin and SEEDMax, have been proposed.Then, we extend SEED to RSEED with the awareness of reliability. To evaluate the effec-tiveness of the proposed algorithms, comprehensive simulations have been conducted and theproposed algorithms are compared with other state-of-art algorithms. Experimental resultsdemonstrate that both SEEDMin and SEEDMax can significantly increase the utilization ofsolar energy without violating job deadlines and overall energy budget. The amount of solarenergy utilized by SEEDMin and SEEDMax is larger than that of two traditional schedulingalgorithms, Min-Min and Min-Max, respectively. Also, it can be seen that RSEED greatlyimproves the reliability by decreasing the temperature.It is crucial to design a request dispatching and resource allocation algorithm to maximize netprofit. The existing algorithms are either built upon energy-efficient schemes alone, or multi-type requests and customer satisfaction oblivious. They cannot be applied to multi-type re-quests and customer satisfaction-aware algorithm design with the objective of maximizingnet profit. This thesis proposes an Ant-colony optimization based algorithm for Maximiz-ing SP’s net Profit (AMP) on geographically distributed data centers with the considerationof customer satisfaction. First, using model of customer satisfaction, we formulate the u-tility (or net profit) maximization issue as an optimization problem under the constraints ofcustomer satisfaction and data centers; Second, we analyze the complexity of the optimalrequests dispatchment problem and rigidly prove that it is a NP-Complete problem. Third,to evaluate the proposed algorithm, we have conducted the comprehensive simulation andcompared with the other state-of-the-art algorithms. It has been shown that AMP maximizesSP net profit by dispatching service requests to the proper data centers and generating theappropriate amount of Virtual Machines (VMs) to meet customer satisfaction. Moreover, wealso demonstrate that the effectiveness of our approach when it accommodates the impactsof dynamically arrived heavy workload and various evaporation rate.