Talk:Olpcfs: Difference between revisions

From OLPC
Jump to navigation Jump to search
No edit summary
(Fix attributions)
 
(6 intermediate revisions by 4 users not shown)
Line 1: Line 1:
If you haven't already, consider taking a look at the Linux version of Venti as well, it is available in the plan 9 from user space suite: http://swtch.com/plan9
If you haven't already, consider taking a look at the Linux version of Venti as well, it is available in the plan 9 from user space suite: http://swtch.com/plan9
Also consider using 9P instead of FUSE as it can give you implicit distributed file system capabilities -[[User:Ericvh|Ericvh]] 15:34, 21 March 2008 (EDT)
Also consider using 9P instead of FUSE as it can give you implicit distributed file system capabilities -[[User:Ericvh|Ericvh]] 15:34, 21 March 2008 (EDT)
: I will probably consider 9P for a 2nd implementation, mostly due to API issues. I'm not interested in building a distributed file system, however. [[User:CScott|CScott]] 13:15, 9 May 2008 (EDT)


In order to guard against database corruption, I suggest that any important file attributed be stored as xattrs on the file (in addition to within the index). Whether the file is a landmark version and maybe MIME type is important. Preview is not important. Tags might be important. [[User:Jpritikin|Jpritikin]] 10:12, 29 April 2008 (EDT)
In order to guard against database corruption, I suggest that any important file attributes be stored as xattrs on the file (in addition to within the index). Whether the file is a landmark version and maybe MIME type is important. Preview is not important. Tags might be important. The most trustworthy database is the kind you can 'rm -rf' and regenerate. [[User:Jpritikin|Jpritikin]] 10:12, 29 April 2008 (EDT)
: Too far down this path and we'll end up wasting a lot of space on redundant content. There is a degree to which you are completely correct, however. [[User:CScott|CScott]] 13:15, 9 May 2008 (EDT)

==Reiser4 comparison==
Has there been a response to this comment from http://lists.laptop.org/pipermail/devel/2008-April/013260.html? Or is this a sufficient [http://www.linux.com/articles/56093 review]? --[[User:FGrose|FGrose]] 21:22, 9 May 2008 (EDT)

:"I looked at the OLPCFS page and it became very clear that OLPCFS is virtually identical to what Reiser4 was initially designed to be. Almost all of the listed OLPCFS features were present in the first Reiser4 designs and almost all were repeatedly vetoed by the Linux VFS guys (Al Viro and Christoph Hellwig) for design reasons, not code reasons. So if you want OLPCFS, try early Reiser4 snapshots (maybe for Linux 2.6.4 or so) and add one or two features on top of it.

:Sorry. I wish the idea would fly, but you may be in for some very tough criticism if you intend to have anything with a concept like OLPCFS merged in the Linux kernel."
::Regards, Carl-Daniel Hailfinger
This doesn't seem to be a technical argument at all. The linked reviews of Reiser4 all refer to personality, code style, maturity, issues. Besides, olpcfs can live quite well as a FUSE or 9P filesystem, so kernel mergeability is moot. [[User:CScott|CScott]] 15:28, 12 May 2008 (EDT)

Latest revision as of 19:28, 12 May 2008

If you haven't already, consider taking a look at the Linux version of Venti as well, it is available in the plan 9 from user space suite: http://swtch.com/plan9 Also consider using 9P instead of FUSE as it can give you implicit distributed file system capabilities -Ericvh 15:34, 21 March 2008 (EDT)

I will probably consider 9P for a 2nd implementation, mostly due to API issues. I'm not interested in building a distributed file system, however. CScott 13:15, 9 May 2008 (EDT)

In order to guard against database corruption, I suggest that any important file attributes be stored as xattrs on the file (in addition to within the index). Whether the file is a landmark version and maybe MIME type is important. Preview is not important. Tags might be important. The most trustworthy database is the kind you can 'rm -rf' and regenerate. Jpritikin 10:12, 29 April 2008 (EDT)

Too far down this path and we'll end up wasting a lot of space on redundant content. There is a degree to which you are completely correct, however. CScott 13:15, 9 May 2008 (EDT)

Reiser4 comparison

Has there been a response to this comment from http://lists.laptop.org/pipermail/devel/2008-April/013260.html? Or is this a sufficient review? --FGrose 21:22, 9 May 2008 (EDT)

"I looked at the OLPCFS page and it became very clear that OLPCFS is virtually identical to what Reiser4 was initially designed to be. Almost all of the listed OLPCFS features were present in the first Reiser4 designs and almost all were repeatedly vetoed by the Linux VFS guys (Al Viro and Christoph Hellwig) for design reasons, not code reasons. So if you want OLPCFS, try early Reiser4 snapshots (maybe for Linux 2.6.4 or so) and add one or two features on top of it.
Sorry. I wish the idea would fly, but you may be in for some very tough criticism if you intend to have anything with a concept like OLPCFS merged in the Linux kernel."
Regards, Carl-Daniel Hailfinger

This doesn't seem to be a technical argument at all. The linked reviews of Reiser4 all refer to personality, code style, maturity, issues. Besides, olpcfs can live quite well as a FUSE or 9P filesystem, so kernel mergeability is moot. CScott 15:28, 12 May 2008 (EDT)