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:
- Old and very well-tested: shipping since 1994
- Very good performance on large IOS
- Fully 64bit filesystem, has no problem with large files and
discs, very well tested ( literaly decades ) with terabyte-range files
and filesystems.
- Security: damaged files gets zeroed...
- Native quota
- Native acl
- Native EAs
Bad:
- Relatively new to linux. But still - oldest journaling filesystem
available on Linux
- Slow on small files ( with 1-2k sized files even ext2 is way
faster)
- Convenience: damaged files gets zeroed... supposedly this is the
issue with all journaling filesystems
- Not very popular - issues with exotic software: quotes inside
virtual servers, etc..
- 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 )
- Takes few % more of diskspace then others, it bit me when I tried
to convert partition that was 99% full...
- On related note - xfs is very slow when filesystem is 99.x% full,
- No
data="">
- 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:
- minix fs/ sys v
- ffs/ext2
- 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
