【作者】 赵军；

【导师】 王莹；

【作者基本信息】 西安电子科技大学 ， 信息安全， 2012， 硕士

【摘要】 当今社会，计算机系统被广泛应用在交通运输、医学、航海、航空等各个领域，人们对计算机可靠性提出了越来越高的要求。事实上，软硬件本身的特性决定了系统完全不发生失效是不可能的。考虑一个需要长时间运行的任务，在执行过程中，如果发生故障，那么不得不重新开始执行，这就造成了不必要的浪费。因此，能够在故障发生时容忍故障就显得尤为重要了。检查点技术就是这样一种有效的容错手段，被广泛用在计算机、数据库系统中，旨在提高系统可靠性。通过在任务运行过程中每隔一段时间设置检查点，避免失效发生时，大量的计算内容被丢失，改善系统的性能。针对1层恢复方案检查点设置开销较大的问题，Vaidya提出了所谓的2层恢复方案，旨在降低在任务运行过程中的检查点设置开销。在2层恢复方案中，存在设置开销不同的2种类型的检查点，即N-checkpoint和local checkpoint，分别被保存在远端存储器和本地磁盘中。设置一个local checkpoint的花销要低于设置N-checkpoint的开销。为了实现优化的性能，Vaidya通过数值搜索给出了指数失效分布下的检查点放置策略。本文提出一种新的2层检查点放置策略，确定整个系统运行过程中localcheckpoint和N-checkpoint放置。该放置策略不仅适用于故障分布服从指数分布的情形，也能适用于更复杂的分布类型，如weibull分布。结果表明，本文给出的策略能获得较好的性能。同时，本文分析了影响相邻N-checkpoint之间最优localcheckpoint数目的因素。结果表明2种类型检查点的设置开销之比和2种失效发生的概率比是影响其的因素。更多还原

【Abstract】 Today computer systems are widely used in transportation, medical, navigationand many other fields. Human beings’ demand on systems reliability is increasingsignificantly. In fact, software and hardware characteristics decide that it is impossiblesystems do not fails during the whole process. Considering a long-running task, if afailure happens during the running time, the task must be restarted. So fault tolerance isparticularly important. As an effective means for fault tolerance, checkpoint/restart iswidely used in computer and database systems to improve the system reliability. TheCheckpoint/Restart technology is used to avoid significant loss of computation when afailure happens to improve the system performance.Aiming at the high performance overhead problem in the one-level recoveryscheme, Vaidya proposed the so-called two-level checkpoint recovery schemes todecrease the overhead which is incured during the process of setting up checkpoints. Inthe two-level recovery scheme, there exist two types of checkpoints which areN-checkpoint and local checkpoint and saved in stable and local disk respectively. Theoverhead of setting a local checkpoint is lower than an N-checkpoint. To obtain theoptimal performance, Vaidya determine the checkpoint placement by numerical search.The paper presents a two-level checkpoint placement strategy which decides tohow to place local checkpoint and N-checkpoint in the whole process. The placementstrategy can not only apply to exponential failure distribution but also more complexfailure distribution type, such as Weibull. The result shows the scheme presented canperform good performance. Meanwhile, we also analyze the factors which affect theoptimal number of local checkpoints between N-checkpoints and the result shows thecheckpoint ratio and the probability is the influences factors.更多还原