【作者】 王晓翠；

【导师】 方娟；

【作者基本信息】 北京工业大学 ， 计算机科学与技术， 2010， 硕士

【摘要】 随着网络技术的飞速发展及云计算概念的提出,用户对存储系统的性能及安全性等方面都提出了更高的要求,云存储也成为目前各存储厂商进军的领域之一。为了提高存储系统的性能,存储系统提供商分别在存储客户端和存储服务器端增加了大容量的内存(main-memory buffers)空间作为缓存(caching)使用。然而,随着硬件的升级和缓存空间的不断增加,系统表现出来的性能却未随之提高,有时甚至出现下降的情况。研究表明,有两个原因限制了这种层次缓存有效性的发挥:第一,随着缓存能力的增加,存储客户端的命中率不断提高,致使到达存储服务器端请求的时间和空间局部性较差,从而导致基于局部性的算法如LRU等对缓存的利用率降低。第二,大量数据被存储客户端和服务器端冗余存储,使整个系统表现出来的性能远远没有达到系统中层次缓存的总和应该达到的性能。因此,如何才能更有效的利用缓存空间成为目前最为广泛研究的问题。本文针对当前存储系统的研究现况分析发现,目前存储系统多是直接应用单级存储系统中的预取算法,这些算法直接应用于多级存储系统将带来很多局限,如太过保守的预取算法不能有效的利用底层的缓存空间,而太过积极的预取算法又有可能穿越多层,产生大量的冗余数据,从而浪费了存储空间等。预取协调算法在不改变原有预取算法的基础上,只通过分析当前数据流的请求方式和Cache的状态来调整预取算法的积极性以提高Cache的利用率。本文通过对当前的预取协调方法进行详细分析,发现其存在两个问题:第一,对于随机请求的影响较小,不能明显提高服务器端的利用率。第二,只能用在同构的系统结构中,即在客户和服务器端使用相同的预取算法。针对预取协调算法存在的两个缺陷,本文提出了相应的改进方案,不仅提高了其在随机请求下的效率,还可将其应用在异构的存储系统架构当中。为了对改进后的预取协调算法进行验证,本文设计并实现了一个两级存储系统实验平台,分别使用不同的预取算法对改进的预取协调算法进行验证。实验结果表明,改进后的算法无论是对顺序访问模式还是随机访问模式,都能够有效的提高存储系统的性能。该算法还可以应用在异构的系统环境之中,这种异构的系统性能几乎完全取决于服务器端使用的预取算法的性能。预取协调算法为提高存储系统的性能提出了一种新的解决方案,不需要任何的额外花费,具有一定的理论价值和应用价值。更多还原

【Abstract】 With the development of network technology and introduction of concept of cloud computing, users have put forward higher requirements in performance and security of storage system. Recently, cloud storage has become a new area that the storage vendors joined. In order to improve the performance of storage system, storage providers are using a large buffer as caching in both client and server. However, as hardware upgrades, the capacity of cache is increasing, the performance of storage system is not increasing, or decreasing in sometimes. Researches show that there are two reasons limit the effectiveness of the multi-level cache architecture. First, with the increase of cache capacity, the hit rate in storage client is increasing, then, the requests sent to server has poor temporal locality. This result in the algorithms based-locality such as LRU has a lower utilization. Second, the large amount of data was stored client-side and server-side. These redundant data made the whole system performance lower than the total amount of caches in the hierarchy should be demonstrated. So, how can improve the utilization of cache space become the most extensively studied problems.The paper makes a research on the storage system. The results show that the prefetching algorithm applied to multi-level storage system is the algorithm that used in single system. These algorithms directly applied to multi-level storage system also brings a lot of limitations, such as overly conservative prefetching algorithm will not be able to effectively use the lower-level cache space, while overly aggressive prefetching algorithm will be compounded across levels and generate large amounts of wasted prefetch. Thus wasting a lot of storage space. Prefetching coordination algorithm can improve the utilization of L2 Cache by analyzing the mode of the requests and forecasting the state of L2 Cache. And it does not change the original prefetching algorithm. Based on a detailed analysis to the current prefetching coordination algorithm, the thesis found two problems: first, it has little effect for the random request, and cannot improve the utilization of L2 Cache significantly. Second, it can only be worked with homogeneous combinations of prefetching algorithms at multiple levels.For the two defects of perfetching coordination algorithm, the thesis put forward a new improvement algorithm. It can not only improve its efficiency in the random request, can also be used in heterogeneous storage systems architecture. In order to validate the improved coordination algorithm, this thesis designs and implements a two-level storage system test bed, and uses different prefetching algorithm in it. The results show that the new algorithm can improve the performance of storage system in both random and sequential requests. It can be used in heterogeneous system environment, and the performance of this kind of heterogeneous system depends almost entirely on the prefetching algorithm that used in server side. Prefetching coordination algorithm provides a new method in improve the performance of storage system, and does not require any additional cost. So, it has very important theoretical and practical value.更多还原