Dariush Debian Diary

Motto: quoting xfs whitepaper from 1996 usenix conference: With today's 9 gigabyte disk drives it only takes 112 disk drives to surpass 1 terabyte of storage capacity (emphasis mine). http://oss.sgi.com/projects/xfs/papers/xfs_usenix/index.html

XFS: eXtended File System (SGI, Unix, Windows)

What are 'journaled filesystems': filesystems with journal.

Good:
1. Old and very well-tested: shipping since 1994
2. Very good performance on large IOS
3. Fully 64bit filesystem, has no problem with large files and discs, very well tested ( literaly decades ) with terabyte-range files and filesystems.
4. Security: damaged files gets zeroed...
5. Native quota
6. Native acl
7. Native EAs

Bad:
1. Relatively new to linux. But still - oldest journaling filesystem available on Linux
2. Slow on small files ( with 1-2k sized files even ext2 is way faster)
3. Convenience: damaged files gets zeroed... supposedly this is the issue with all journaling filesystems
4. Not very popular - issues with exotic software: quotes inside virtual servers, etc..
5. Not very simple codebase, stable and well-tested, but very big and relatively intrusive. (non-issue on 2.6, where lots of needed support was moved to lower layers, things like variable-size IO requests, delayed allocation )
6. Takes few % more of diskspace then others, it bit me when I tried to convert partition that was 99% full...
7. On related note - xfs is very slow when filesystem is 99.x% full,
9. No data=journal
10. You can't use it in raid1 both driectly on hardware and via md1

Features:
• Large file & filesystem support: 2^63-1 byte files supported.
• Extensive use of B+trees
• Scalable&multi-threaded - scales very well with added spindles and CPUs ( working instalations with up to 512cpus... ), mainly thanks to Allocation Groups.
• Variable block size: 512bytes to 64kbytes
• Realtime data subvolume
  • Data ( metadata+files data)
  • Log
  • Realtime
  Usefull utilities:
  • xfsdump,xfsrestore. (this is my personal top feature;)
  • xfs_growfs (fast and reliable. similiar utilities exist for ext2&stuff, but they seem kludgy, and are not supported by ext2/3 developers)
  • mkfs.xfs - very fast (this is surprisingly important when you get to terabyte range)

  History: 1993: Berkeley FFS - state of the art at the time. Irix uses EFS - extents-based FFS problems to solve Linux had the same problems to fix in ~1999:

  • Filesize limit(2 GB)
  • Filesystem limit(8GB)
  • Statically allocated metadata
  • Long recovery times
  • Slow operation on big directories
  • Lack of extended attributes
  • Problems with media streaming
  • General problems with IO speed, high-end hardware became so fast at IO that FFS couldn't keep up. Stuff like late allocation helped with that

  Filesystems generations:

  1. minix fs/ sys v
  2. ffs/ext2
  3. jfs/xfs/reiserfs
     • journaling(or equivalent)
     • flexible metadata structures
     • dynamic inode allocation
     • extents
  Things XFS brought to linux
  • XFS - the filesystem;)
  • DirectIO ( btw, besides normal uses for DirectIO, it can be used to bypass 16Terabytes filesize limit that exists on linux systems, with DirectIO it's possible to use 8388608TB files ;)

  How to install debian on XFS? http://people.debian.org/~blade/XFS-Install/ - woody cds...

  ---

  Generic: http://www.usenix.org/publications/library/proceedings/usenix2000/general/full_papers/seltzer/seltzer_html/index.html - Journaling vs Soft Updates http://www.mckusick.com/softdep/index.html
  [/Talk] permanent link