초록

용어보기논문에서 용어와 풀이말을 자동 추출한 결과로, 시범 서비스 중입니다. AI-Helper

Abstract ▼ AI-Helper

Recently, SSDs (Solid State Drives) are being widely used instead of hard disks as storage media in various computing environments such as PC and data center. Since the SSDs have the completely different characteristics from hard disks, the existing operating systems optimized for hard disks should ...

Recently, SSDs (Solid State Drives) are being widely used instead of hard disks as storage media in various computing environments such as PC and data center. Since the SSDs have the completely different characteristics from hard disks, the existing operating systems optimized for hard disks should be redesigned for SSDs. In this paper, we propose a new I/O scheduler that fully exploit the following characteristics of SSDs. First, since the SSD consists of many NAND flash packages and each package is able to perform its own I/O operation, almost SSDs provide some parallel I/O operations to improve the I/O performance. In order to exploit the parallelism of SSDs, we first analyze the size of parallel I/O in SSDs and make use of it in determining the maximum request size in I/O scheduler. Second, unlike hard disks, SSDs do not have data access overhead such as seek time and rotation delay. It means that the data transfer time is dominant in SSD's I/O time and consequently it will be proportional to I/O request size. In order to minimize the average response time of I/O requests, we schedule the I/O requests in size order and service from the I/O request with the shortest size. Finally, the performances of read and write are asymmetric in SSDs, and it should be considered in designing the I/O scheduler. To this end, we manage read requests and write requests separately, and then service read requests prior to write requests because read requests should be serviced synchronously. We evaluate our scheduler by implementing the prototype on a Linux 2.6.30 kernel. Through the extensive experiments, we show that it significantly improves the performance in terms of average response time for common workloads.

Keyword