摘要
后拷贝迁移算法需要在地址空间不完整的情况下恢复进程运行,导致进程由于缺页错误过多而频频中断,严重影响了迁移的整体效率.针对这一问题,提出了Pre-record算法.该算法延长进程在源节点上的执行时间并对该过程中访问的地址空间页面加以记录,迁移时源节点优先迁移被记录的页面,然后继续推送剩余地址空间页面,保证在缺页错误发生频率最高的进程恢复运行初期能够获取所需内存页面.分析及实验结果表明,该算法能够有效降低迁移过程中缺页错误的发生率,进而提高了进程迁移的整体效率,并具有冻结时间短、剩余依赖度低等优点.
Post-copy algorithm needs process to resume running while address space is incomplete. Due to the frequently occurring page faults, it would cause process executing intermittently which may significantly reduce migration efficiency. So, a new migration algorithm named as Pre-record was proposed to solve this problem. Based on the proposed algorithm, process execution on host node would be prolonged and the visited memory pages would be recorded. Host node will migrate these recorded pages preferentially, and then continue to push the rest of the address space. This can guarantee that the needed memory pages couId have been acquired at the beginning phase of process resume running in which phase page faults have the highest frequency. Analysis and simulation experiments show that Pre-record algorithm can reduce the occurrence rate of page faults effectively and improve the efficiency of process migration, and it also has the advantages of short freeze time and less residual dependence.
出处
《东北大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2016年第1期29-33,共5页
Journal of Northeastern University(Natural Science)
基金
国家自然科学基金资助项目(61272177)
中科院网络化控制系统重点实验室开放课题(WLHKZ2014007)
关键词
地址空间
分布式系统
进程迁移
后拷贝算法
缺页错误
address space
distributed system
process migration
post-copy algorithm
pagefault