Research On Resource Scheduling Mechanism In Grid Market

With the continuous developments of Grid technology, numbers of business Grid projects pop up dramaticaly. As a typical provider with free computing resources, Grid gradually turns into a commercial Internet resources sharing middleware, and provides resources and services as public goods. It means that Grid has already reached a new plane of development in Grid market. In Grid market, it makes full use of Economics to allocate and manage resources, which called the non-cooperative Grid. Comparing with the traditional Grid, Grid market has its own driven characteristics which encourage a number of resource providers and resource consumers to join in. It not only brings a wealth of resources and a large number of users, improves the efficiency of Grid, reduces the total cost of the community, but also provides fresh impetus for the further development of Grid. Both at home and abroad, Grid market, a typical open network representative, has become one of the most active research areas in distributed system.The Grid resource exchanging environment is strongly similar to real economy environment. Grid resource allocation and management is similar to the real social and economic activities too. Therefore, current market mechanism and management methods can effectively solve the problems of job scheduling and resource allocation in Grid market. Economics largely promotes development of the Grid market. However, the Grid market resources are becoming more heterogeneous, dynamic and autonomous, and the optimized goal changes into customer-centric to challenge the problems of job scheduling in Grid market.The Grid market has large scale, strong heterogeneous, dynamic, autonomous resources and complex QoS requirements of users. This dissertation focuses on job scheduling and resource allocation in Grid and does the following researches.In this dissertation, the concept and history of Grid are introduced firstly. Then the concept and characteristics of Grid market are discussed and the status of the Grid market research is given. Main problems in job scheduling and resources allocation are analyzed. Due to the new characteristics of Grid market, such as more heterogeneous, dynamic, autonomous and customer-centric objective, we have deeply investigated the following problems which are also discussed later:low resource discovering efficacy; low resource availability; unsatisfied QoS requirements of users.Based on the formal description of a typical scheduling, we discuss the new characteristics of scheduling problems in Grid market and analyse the effects to job scheduling and resource allocation. Chapter2 introduces the formal description and notions of scheduling, and then describes the framework of Grid scheduling. At last, the new characteristics of job scheduling and resource allocation in Grid market are discussed, which are the bases of the following researches.Aiming at the resource instinctive characteristic of decentralized, dynamic and heterogeneous, a type-aware multi-attribute resource discovery mechanism based on super nodes is proposed in Chapter 3. Setting aside the traditional nodes partition, the mechanism treats resources as independent entities and partitioned them into several resource groups according to their types and attributes similarity. This partition approach improves the resources'purity'of a group leading to accurately searching. Type is part of the identities of resource group managers, and this makes it possible to quickly locate certain type of resource group, and leads to higher performance of resource discovery. Simulation shows that the type-aware multi-attribute resource discovery mechanism can increase the accuracy of searching to a certain extent. It can effectively decrease average hop counts, response time and messages brought by resource discovering. At the same time, the performance can be preserved when the system scales up.An accurate evaluation method of resource availability risk is proposed in Chapter 4. It can efficiently evaluate the availability of Grid resources and provide a new point of view for job scheduling in Grid market. The concept of Grid resource availability risk is put forward first and an evaluation approach of availability risk is also proposed based on the historical traces. Based on probability theory and statistic method, we predict and evaluate the availability risk of resource, and propose an accurate evaluation method of resource availability for single resource and resources set. Besides, a Grid job scheduling algorithm based on availability risk evaluation is also proposed. Simulations show the evaluation approach can accurately select available resources set and reduce job failure rate efficiently.By introducing trust mechanism into Grid market in Chapter 5, a trust-filtered approach for resource selection is proposed. This approach first computes the trust value of Grid market resource based on the trust model, and determines the available resource set based on users'trust requirement. Then we compute the cost and risk cost of available resources to get each opportunity cost. At last, a minimal opportunity cost algorithm is proposed. Simulations show this approach is an efficient incentive for both provider and consumer. It effectively guarantees profit of providers with more reliable resources to encourage more providers to join in Grid market. At the same time, it can efficiently reduce job failure rate and the cost of users.A fuzzy decision based multi-QoS batch scheduling algorithm is proposed in Chapter 6. The method resolves the problem that the users could not supply accurate QoS requirements information in Grid market and meets users'QoS requirements.A formalization of multi-QoS based batch scheduling is presented first, and it maps fuzzy users'QoS requirements to resources based on fuzzy decision theory to determine the weights of multi-QoS based on AHP algorithm. A fuzzy decision based multi-QoS batch scheduling algorithm is proposed at last. Simulations show the approach can efficiently meet users'QoS requirements without accurate QoS parameters. Comparing with traditional QoS based batch scheduling algorithms; this approach has a better accomplishment ratio of one-time job and lower fluctuation.By analyzing the characteristics of Grid market such as large-scale, heterogeneity, unreliability and customer-centric, this dissertation gives deeply analysis of the following problems, including low resource discovering efficacy; low resource availability; and unsatisfied users'QoS requirements. To solve these problems, a type-aware multi-attribute resource discovery mechanism, an accurate evaluation of resource availability risk method, a trust-filtered approach for resource selection, a fuzzy decision based multi-QoS batch scheduling algorithm are proposed in this dissertation. They give a new point of view and fine prospect for solving the problems of Grid market scheduling.