Reiser4

Differences between revisions 1 and 2
Revision 1 as of 2006-06-13 09:58:16
Size: 2768
Editor: mail
Comment:
Revision 2 as of 2006-06-16 17:11:24
Size: 3361
Editor: 0x50c71cc7
Comment:
Deletions are marked like this. Additions are marked like this.
Line 41: Line 41:
 * SELinux + Reiser4 is a no-go due to their xattr implementation.
 * Reiser4 is not yet upstream and is likely not to be for a while, at the earliest Reiser4 will be in 2.6.19 depending on code review.
 * There are known, possibly unsolvable, deadlocks in the file-as-directory implementation.
Line 42: Line 46:

 * Due to the lack of upstream acceptance currently this comes down to whether or not Edgy will be able to ship a kernel with support and Reiser4 making it upstream in time to allow Ubuntu plenty of time to test and implement Reiser4 support. Therefore this is likely to be an Edgy+1 item.

Summary

Reiser4 is a fairly radical departure from Reiser3 and other filesystems. Benchmarks show some very average desktop performance in some use cases, but excellent performance in many areas. On-the-fly compression can increase bandwidth (and capacity). This filesystem should be available for Edgy, and its peculiar advantages could be exploited.

Rationale

Reiser4 shows good performance in a range of use cases. It should be a choice, especially for server installs. It is not supported in the mainstream kernel, apparently due to arguments about coding style.

The Reiser4 equivalent of user_xattr is a system where meta-data is stored as sub-files, so a file is actually a folder and a file. (A strange concept, but attractive on consideration, also approved technically by Mr. Torvalds.)

This could be a barrier, or an opprtunity to drench Ubuntu with rich meta-data and make desktop usability take off. Each file format has its own meta-data structures. This is annoying, and there are frequently severe limitations on meta-data storage and incompatibility between versions (e.g. id3 tags). Under the Reiser4 paradigm (and possible with use of user_xattr: is this possible?) the user can attach meta-data to any file, obviating the need for each file format to continue to redesign the wheel. See second use case.

Use cases

  • Penelope wants to try out Reiser4 on a new server build. She heard the performance was excellent. She installs Edgy because it is cutting edge, and expects Reiser4 to be there.
  • Horace has thousands of photos and mp3s. Using Edgy with Reiser4, the mp3 lyrics are stored as text files 'inside' the mp3 file/folders. His annotaions of pictures are stored in the same way. This replaces hidden .folders everywhere, and is much faster.

Scope

  • Reiser4 as option at install time

Design

  • Beagle would need rewriting to use Reiser4 meta-data scheme, instead of user_xattr. This could probably be abstracted fairly easily.
  • See how meta-data can be integrated more closely in ubuntu's operation, particularly in search. Replacing extensive .hidden folders with resier4 type structures is a much neater and more efficient solution.

Implementation

  • Include Reiser4 patches in the Ubuntu kernel build
  • Reiser4 tools are already in the repos.

Code

Data preservation and migration

Outstanding issues

  • SELinux + Reiser4 is a no-go due to their xattr implementation.
  • Reiser4 is not yet upstream and is likely not to be for a while, at the earliest Reiser4 will be in 2.6.19 depending on code review.
  • There are known, possibly unsolvable, deadlocks in the file-as-directory implementation.

BoF agenda and discussion

  • Due to the lack of upstream acceptance currently this comes down to whether or not Edgy will be able to ship a kernel with support and Reiser4 making it upstream in time to allow Ubuntu plenty of time to test and implement Reiser4 support. Therefore this is likely to be an Edgy+1 item.


CategorySpec

Reiser4 (last edited 2008-08-06 16:28:03 by localhost)