The Design and Implementation of nilfs, a Log-Structured File System for Linux

Transcription

1 Linux nilfs Linux nilfs nilfs inode B-Tree inode nilfs The Design and Implementation of nilfs, a Log-Structured File System for Linux Yoshiji Amagai, Hisashi Hifumi, Ryusuke Konishi, Koji Sato, Seiji Kihara and Satoshi Moriai We developed a Log-Structured local file system nilfs for Linux operating system. The Log-Structured file system prevents data write operations from overwriting the disk blocks. So, it minimizes a damage of file data and file system consistency on hardware failure. The Log-Structured system is able to create a snapshot of the file system immediately and to check the file system speedy. The nilfs implements file and inode block management by B-Tree structure. The B-Tree structure enables the nilfs to create a huge file, to store a large number of files. This paper presents the design and implementation of the nilfs. 1. NTT OSS NTT Cyber Space Laboratories Open Source Software Computing Project inode inode Linux ext2 ext3 1) XFS 2) ReiserFS 3) JFS 4) ext3 1

2 2 Linux (Log-Structured File System, LFS) LFS Linux 2.6 nilfs (New Implementation of a Log-Structured File System) 2 3 LFS 4 5 LFS Linux ( 1 ) Linux (2) Linux (3) (4) (5) (6) ext3 (7) (8) (9) (10) blockgroup 0 blockgroup 1... blockgroup n group desc datablock bitmap 1 inode bitmap inode data block datablock placed fixed position (11) 3. LFS 3.1 UNIX 1 inode truncate truncate ext2/3 BSD FFS 5) 1

3 3 segment 0 segment 1... segment n header segment summary partial segment 0 new file blocks 2 partial segment 1 partial segment 2 new inode blocks LFS... check point 3.2 LFS LFS 2 LFS inode LFS inode inode inode (fsck) inode inode inode (Garbage Collection, GC) GC LFS GC LFS GC 4M 3.3 Sprite 6) 1990 Sprite-LFS 7) inode inode inode map inode map BSD OS Sprite-LFS BSD-LFS 8) BSD-LFS inode ifile inode inode GC BSD OS NetBSD 9) Linux kernel2.2 ext2 LinlogFS 10)

4 4 segment superblock management seg 0 seg 1 seg n... partial segment logical segment 3 nilfs 4. nilfs 1 1 nilfs inode 64 B-Tree 11)12) inode inode B-Tree nilfs GC sync pdflush GC 4.2 inode inode inode inode inode nilfs inode B-Tree B-Tree inode B-Tree B + -Tree 12) B*-Tree 12) B-Tree B + -Tree nilfs B + -Tree LFS B-Tree B-Tree B-Tree

5 5 inode B-Tree inode B-Tree inode B-Tree inode B- Tree inode B-Tree 1 B-Tree inode B-Tree inode inode B-Tree B-Tree B-Tree B-Tree inode B-Tree B-Tree nilfs ext segment summary file block data blocks file btree block 4 inode block check point inode btree block GC B-Tree inode inode B-Tree inode B-Tree inode 4.4 nilfs

6 6 nilfs inode B-Tree Linux2.6 (radix-tree) inode Linux2.6 block device inode 1 (bd inode) 5.2 nilfs Linux Linux LFS Linux B-Tree bd inode nilfs radix-tree radix-tree inode 5.3 sync pdflush (1) (2) inode B-Tree B-Tree radix-tree ( 3 ) inode radix-tree inode (4) B-Tree 5.4 (5) B-Tree inode inode B-Tree ( 6 ) B-Tree radix-tree (7) bio POSIX fsync fsync

7 7 5.4 dirty LFS B-Tree B-Tree LFS O SYNC nilfs 1 dirty B-Tree nilfs 1Mbyte 4Kbyte 265 O SYNC 4Kbyte write 1541 dirty O SYNC 5.5 GC nilfs GC B-Tree dirty dirty 5.6 nilfs LFS GC nilfs B-Tree inode inode B-Tree inode nilfs mount GC GC 1 B-Tree elephant 13) elephant landmark 5.7 rm LFS 6. iozone nilfs ext3 Pentium 4 3.0GHz 1G Linux kernel Ultra ATA 7,200rpm 8MByte 4KByte 32MByte 512MByte 1 write 2KByte 8KByte fsync

8 8 1 (MB/s) Table 1 continuous write EXT3 nilfs 2K 4K 8K 2K 4K 8K 32M M M (MB/s) Table 2 random write EXT3 nilfs 2K 4K 8K 2K 4K 8K 32M M M O SYNC (MB/s) Table 3 measurement with O SYNC option EXT3 nilfs 2K 4K 8K 2K 4K 8K 32M M M M M M MByte/sec ext3 ordered mode kernel2.6 anticipatory nilfs GC 1 nilfs ext3 2 O SYNC 3 nilfs ext3 nilfs ext3 GC 7. Linux nilfs B-Tree inode nilfs GPL i386 B-Tree 1) Stephen Tweedie: Journaling the Linux ext2fs Filesystem, LinuxExpo 98, ) Project XFS Linux, 3) ReiserFS, 4) JFS for Linux, 5) Marshall K. McKusick, William N. Joy, Samuel J. Leffler and Robert S. Fabry: A Fast File System for UNIX, Computer Systems, Vol.2, No.3, pp , ) J. K. Ousterhout, A. R. Cherenson, F. Douglis, M. N. Nelson and B. B. Welch: The Sprite Network Operating System, Computer Magazine of the Computer Group News of the IEEE Computer Group Society, ACM CR , Vol.21, No.2, ) Mendel Rosenblum and John K. Ousterhout: The Design and Implementation of a Log- Structured File System, ACM Transactions on Computer Systems, Vol.10, No.1, pp.26 52, ) Margo I. Seltzer, Keith Bostic, Marshall K. McKusick and Carl Staelin: An Implementation of a Log-Structured File System for UNIX, USENIX Winter, pp , ) The NetBSD Project, 10) Christian Czezatke and M. Anton Ertl: Lin- LogFS A Log-Structured Filesystem For Linux, Freenix Track of Usenix Annual Technical Conference, pp.77 88, ) R. Bayer and E. McCreight: Organization and Maintenance of Large Ordered Indexes, Acta Informatica Vol.1, Fasc. 3, pp , ) Douglas Comer: The Ubiquitous B-tree, ACM Computing Surveys Vol.11, No.2, pp , June ) D.J. Santry, M.J. Feeley, N.C. Hutchinson and A.C. Veitch: Elephant: The file system that never forgets Proceeding of IEEE Hot Topics in Operating Systems, pp.2 7, March 1999.