集群环境下一种基于交易模型的空闲资源分配方法

针对集群中地理位置分散的计算资源的使用不均衡现象，提出了一种基于交易模型的空闲资源分配方法。该算法解决了分布计算资源的自治性问题、异构支持问题、资源管理者域限制问题和动态控制问题。首先给出了交易模型的构架，其次给出了交易模型匹配算法，最后对该算法进行了系统的评价。该算法已经在一个大型网络作业管理系统中得到应用，取得良好的效果。

On Better Utilization of Idle Resources in Cluster Environment

Cluster environment often includes a number of workstations in addition to personal computers. Existing methods for utilization of idle resources are, in our opinion, not quite efficient. We propose a new method based on trade off model in order to achieve better utilization of such resources. Section 1 discusses this trade off model, which consists of provider, applicant and arbitrator. Essentially the provider provides a list of idle resources, the applicant applies for resources in the above mentioned list, and the arbitrator arbitrates when conflict arises. Finally, section 1 decomposes working process of this trade off model into five steps. Sections 2 and 3 describe how to implement our method based on trade off model. Essentially the model's applicant part uses SIMPLE DEAL algorithm developed by us to file application for resources; the model's arbitrator uses ARBITRATOR algorithm developed by us to determine the optimal idle resources and allocates them to the model's applicant part; once the applicant part accepts such allocation, it sends confirmation of such acceptance to model's provider; after successful confirmation, the model's applicant part utilizes the allocated idle resources to accomplish its tasks; meanwhile the main algorithm developed by us deletes the allocated resources in the list of idle resources provided by the model's provider. Fig. 3 shows that the resource sharing quality of our new method based on trade off model is quite near the ideal value of 1. Our new method has been tried on a certain small network job management system and can achieve an average of 70% utilization of idle resources.