Journaling file systems

1
Journaling file systems

• Alexander Goberman

2
In this presentation

• Traditional (static) file systems
• How they work
• What can go wrong?
• Possible solutions
• Journaling
• Overview of 4 main journaling FS
• Conclusions and recommendations

3
File system overview

• Important part of an operating system
• Data storage
• Data integrity
• Meta-data
• Performance

4
File systems vernacular

• Logical block - smallest unit of storage
• Logical volume - physical disk or subset of the
  physical disk space
• Block allocation
• Internal fragmentation
• External fragmentation

5
File systems vernacular (continued)

• Extent number of contiguous blocks
• Meta-data file systems internal data structure
• Inode - stores all of the information about a
  file except the data itself
• Directory special file

6
How file systems work

• Store data on hard drive
• Determine where file blocks should be stored
• Maintain locations
• Ext2fs map of inodes
• Change the contents of the file
• Edited data is stored in the same location
• New blocks are allocated if needed
• Meta-data is changed

7
When good file system goes bad

• Closer look at saving a modified file
• New blocks allocated
• Files inodes updated
• Data written into the blocks
• Assume power outage
• Midst of writing data into the blocks
• Midst of updating meta-data
• FSCK has to check the entire file system

8
Possible solutions

• Keep log of changes(NTFS, EXT3, ReiserFS, JFS)
• Log contains a list of all changes to either
  meta-data or data (or both) BEFORE they are
  actually done
• Pros Downtime due to FSCK reduced dramatically,
  corrupted files are more likely to be fixed
• Cons Same chances to loose the data

9
Possible solutions (continued)

• Keep a copy of a file(Elephant)
• Changes are done to a copy of a file
• Pros At least one consistent copy of a file
  exists at any given point of time, high
  reliability
• Cons Huge overhead, low performance
• Log based file systems
• Entire file system based on data from a log file
• Are not in use any more

10
Journaling

• Journaling is introduced to solve problems with
• File system availability
• File consistency
• File management
• Allows to do file modification as an atomic
  operation
• Changes either complete or no change is done
• All modern journaling systems use meta-data
  (data) logging

11
Journaling file systems

• Four major journaling file systems
• EXT3
• JFS
• ReiserFS
• XFS
• NTFS does not handle full-fledged journalingit
  logs alterations to files, but does not provide
  enough information to reverse them

12
EXT3

• An extension to EXT2
• Easy migration from Ext2 to Ext3, and back!
• Meta-data and data journaling
• The only FS that provides data journaling
• Has all of the characteristics of Ext2
• Is not the fastest or most efficient FS
• Data journaling large overhead

13
JFS (Journaled File System)

• Fully integrated journaling
• Extent-based addressing structures
• Dynamic inode allocation
• Two different directory organizations
• Small directories (up to 8 files or folders)
• Large directories (organized in B trees)
• Full 64-bit file system
• Min FS size 16 MB
• Max FS size 512 TB 4096 TB (depending on the
  block size)

14
ReiserFS

• If you need journaling, we are the most stable
  FS available Hans Reiser from NameSys
• Properties
• Meta-data journaling
• Uses a separate Btree for every directory
• Very efficient when managing small files
• Much more complicated indexing algorithms
• Does not use extents
• Allows use of extensions (in close future)

15
ReiserFS Pros and Cons

• Best support for the small files
• x8 x15 times faster then ext2 (files less then
  1K)
• Stores more data then Ext2 on the same disk space
• Small journaling overhead
• High potential of extensions
• Access Control Lists
• Arbitrary (user) meta-data

• Poor performance with large files
• Complicated indexing algorithms

16
XFS

• Initially created by the CGI for the IRIX OS
• Direct I/O
• Extent-based space management with either bulk
  preallocation or delayed allocation.
• Max. performance and min. fragmentation
• Journaled quota subsystem
• Extended attributes
• Access Control Lists
• Support for Hierarchical Storage Management
• Largest file system support (up to 18000 PB)
• Best speed and security performance

17
Summary and recommendations

• Journaling is a very useful feature!
• Which FS is better?
• Ext3 Meta-data Data Journaling, easy migration
  from ext2, reliable underlying FS
• JFS High integrity, fast, support for large FS
• ReiserFS Good performance, very high potential
  FS
• XFS Best mix of speed, security, and support for
  large file systems

18
Questions