应用控制的Web服务器磁盘缓冲方法

针对大数据负载时磁盘I／O阻塞造成的Web服务器性能下降的问题，提出了应用程序控制缓冲（ACC）方法．其核心是，缓冲跟踪模块根据应用程序的文件访问过程来跟踪内核中的文件缓冲状态，缓冲控制模块进行缓冲替换和预取，保持文件缓冲有足够的空闲空间．这样，服务器可在用户空间控制文件缓冲，从而准确判断文件是否在缓冲之中，并依此来调度请求，以提高处理器和磁盘的I／O并行度．同时，服务器可采用适应自身特点的缓冲和预读策略，以提高缓冲的命中率．作为示例，将ACC在Flash服务器中实现，实现中选用了“金字塔选择”缓冲算法．实验表明，在大数据负载下使用ACC的Flash服务器性能有很大的提高，即便在数据负载稍大于物理内存空间的情况下，服务器的吞吐率仍可提高约24．4％，而当数据负载超出物理内存2～3倍时，吞吐率可提高3～4倍。

Application-Controlled Caching for Web Server

Focusing on the problem that under workloads of large dataset,the performance of a Web server is dropped due to the disk I/O blockings,an application-controlled caching(ACC) approach is presented.In ACC,the cache tracking module traces the status of file caches in kernel according to the files access process of applications,and cache replacement and prefetch are performed in cache control module to keep enough free space for file caches.With ACC,servers can control the file caches in user space,and determine whether a file is in caching,then schedule requests based on this to improve the parallelism of the processor and disk I/O.Meanwhile,servers can choose a suitable policy of caching and prefetching to increase cache hit ratio.ACC is implemented in Flash server as a demonstration,in which pyramidal selection scheme is adopted.Experiments show that the performance of Flash with ACC can be largely enhanced under workloads of a large dataset.Even when the dataset load is slightly larger than the physical memory space,its throughput is improved by about 24.4%;when the dataset load is 2-3 times larger,the throughput improved by 3-4 times.