仿真网格中预防死锁的资源分配方法及其仿真分析  

Method and Simulation of Deadlock Free Resource Co-allocation in Simulation-grid

作　　者：张传富[1] 查亚兵[1] 刘云生[1] 张童[1] 黄柯棣[1] 

机构地区：[1]国防科学技术大学机电工程学院自动化所,湖南长沙410073

出　　处：《系统仿真学报》2006年第9期2588-2592,共5页Journal of System Simulation

基　　金：国防预研基金(51404010403KG0155)

摘　　要：仿真网格中资源协同分配是资源组织和调度的一个重要组成部分,如何避免仿真应用之间的死锁是资源协同分配过程中需要解决的重要问题。通过对仿真网格中死锁原因的分析,对死锁的特点进行了描述。根据解决死锁问题的不同策略,提出了仿真网格中基于原子事务预防死锁的资源协同分配方法,并根据应用的不同资源分配方案给出了具体算法。通过理论分析并结合仿真试验研究了该算法在资源协同分配过程中的性能,试验证明使用该方法不仅能够避免应用资源分配过程中的死锁,与其它方法相比,还能获得好的资源分配性能。Simulation applications designed to execute on simulation grid frequently requires the simultaneous co-allocation of multiple resources in order to meet performance requirement, In particular, the problem of co-allocating simulation grid resources that span multiple administrative domains was complicated by the possibility of deadlock. Motivated by this, the reason for application deadlock was analyzed and the deadlock in simulation grid was characterized. Further, a new method for fast co-allocation of simulation grid resources was developed based on different strategies dealing with deadlocks, and some detailed algorithms were proposed according to different co-allocation schemes. The proposed method based on Atomic Transaction is free from deadlock. Experimental results demonstrate that the proposed method yields a significant performance improvement over the other existing deadlock prevention method.

关 键 词：仿真网格 原子事务 资源协同分配 死锁 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]