一种基于小数据同步写的回写I/O调度器

小数据同步写普遍存在于的各种计算机环境中，并且可以由计算机系统的不同层次软件产生，从底层操作系统一直到上层应用软件都可以生成小数据同步写请求.然而，操作系统的文件系统是以块作为最小逻辑可寻址单位，小数据写将会导致严重的写放大问题，使得系统的I/O性能大幅降低.为了解决上述问题，我们提出了一种新的I/O调度器，并将它取名为“Hitchhike”.该调度器可以识别小数据写，并通过对其他数据块中的数据进行压缩，，将小数据嵌入到压缩出来的空间中，从而将小数据和该数据块一起写入到磁盘上，以异步回写的方式完成小数据的同步写，不仅有效缓解了磁盘的写放大问题，也大大提高了小数据同步写的效率.我们基于Linux中Deadline调度器实现了Hitchhike原型系统，并利用Filebench基准测试来测试调度器在吞吐量、I/O延迟等方面的性能.通过和传统I/O调度器的性能进行比较，我们可以发现Hitchhike调度器能够显著的提高小数据同步写性能高达48.6%.

Writeback I/O Scheduler Based on Small Synchronous Writes

Small and synchronous writes are pervasive in various environments and manifest in various levels of software stack, ranging from device drivers to application software. Given a block interface, small write can cause serious write amplifications, which can substantially degrade the overall I/O performance. To address this issues, we present a novel block I/O scheduler, named Hitchhike, in this paper. Hitchhike is able to identify small writes, and embed them into other data blocks through data compression. With Hitchhike, we can enable a writeback buffer for small synchronous writes, not only removing the write amplification, but also dramatically improving the performance of small synchronous writes. We implemented Hitchhike based on the Deadline I/O schedulers in Linux 2.6.32, and evaluated it by running Filebench benchmark. Our results show that compared to traditional approaches, Hitchhike can significantly improve the performance of synchronous small writes up to 48.6%.