A Novel Reordering Write Buffer to Improve Write Performance of Log-Structured File Systems
IEEE Transactions on Computers
WOLF--A Novel Reordering Write Buffer to Boost the Performance of Log-Structured File System
FAST '02 Proceedings of the 1st USENIX Conference on File and Storage Technologies
Proceedings of the twentieth ACM symposium on Operating systems principles
HyLog: A High Performance Approach to Managing Disk Layout
FAST '04 Proceedings of the 3rd USENIX Conference on File and Storage Technologies
Online reorganization of databases
ACM Computing Surveys (CSUR)
WOLF: a novel reordering write buffer to boost the performance of log-structured file systems
FAST'02 Proceedings of the 1st USENIX conference on File and storage technologies
HyLog: a high performance approach to managing disk layout
FAST'04 Proceedings of the 3rd USENIX conference on File and storage technologies
Efficiently identifying working sets in block I/O streams
Proceedings of the 4th Annual International Conference on Systems and Storage
Data migration in RAID based on stripe unit heat
ICMLC'05 Proceedings of the 4th international conference on Advances in Machine Learning and Cybernetics
Hi-index | 0.00 |
Abstract: I/O is a major performance bottleneck in modern computer systems. Modern disks use the Zone-Bit-Recording (ZBR) technology to increase the capacity. A direct consequence of ZBR is that outer tracks have higher data transfer rates. Because LFS uses large disk transfers, access times are mainly determined by the data transfer rate. This paper presents a novel performance-oriented data reorganizing scheme, called PROFS, which boosts the I/O performance of LFS (Log-structured File System). Our scheme reorganizes data on the disk during LFS garbage collection and system idle periods. By putting active data in the faster zones and inactive data in the slower zones, we can significantly improve the read and write performance. The reorganization overhead is minimal. We studied three different algorithms that are read-optimized, write-optimized and balanced. Simulation results based on five real-world and two synthetic traces showed that such algorithms can dramatically reduce the I/O response time for both reads and writes compared with the most recent LFS systems with adaptive methods. Our algorithms improve the LFS system's write latency by up to 24.5% and the read latency by up to 21.3%.They also improve the file system throughput by up to 32%.