集群环境下一种基于交易模型的空闲资源分配方法

On Better Utilization of Idle Resources in Cluster Environment

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 。

Cluster environment often includes a number of workstations in addition to personal computers. The methods for utilizing idle resources are not quite efficient. We propose a method based on trade-off model to achieve better utilization of such resources. This trade-off model 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. The working process of this trade-off model is decomposed into five steps. Essentially the model's arbitrator uses ARBITRATOR algorithm to determine the optimal idle resources and allocates them to the model's applicant part. Once the applicant part accepts such allocation, it sends acceptance confirmation to model's provider; after successful confirmation, the model's applicant part utilizes the allocated idle resources to accomplish its tasks. Another main algorithm deletes the allocated resources in the list of idle resources provided by the model's provider. The resource-sharing quality of the trade-off model is quite near the ideal value of 1. It has been tried on a certain small network job management system and can achieve an average of 70% utilization of idle resources.