Date: Mon, 1 Nov 1999 21:20:21 -0500 (EST) From: Robert Watson <robert@cyrus.watson.org> To: Rodney <rr@xs4all.nl> Cc: freebsd-fs@FreeBSD.ORG Subject: Re: feature list journalled fs Message-ID: <Pine.BSF.3.96.991101211533.22399A-100000@fledge.watson.org> In-Reply-To: <19991031120514.A28103@xs4all.nl>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 31 Oct 1999, Rodney wrote: > here's my list of features I'd like to see in a > journalled fs. Have to admit this list is heavily > inspired ( ok , copied ) from the VxFS features, > apart from th buzz words, > some of them make sense, some of them don't > but it should give us some stuff to discus: > > 1) extent based allocation > coding this should be easy, it's just a address-lenght pair > identifying the starting block address and the length of the > extent. I've seen this coded up in qnxfs under linux. > I think the vsta filesystem does something similar. > 2) fast filesystem recovery , obviously > 3) acls would be nice , afs style ? > 4) online defrag and resizing (while user are online) > 5) online backup/snapshot > 6) vinum integration (vague) > 7) built features that make databases very happy > like msql/mysql/oracle. (vague) > > also b?trees for indexing sounds cool, thought the xfs > implementation seems quite heavy(they maintain 2 of them) > , ie over-kill ? > The way b+trees are use in the Be fs (bfs) might be more > appropriate. I guess I'd be interested in more seperation of the on-top semantics and filestore. I.e., some piece of code provides a transactional filestore--inodes, attributes, and blocks of data. On top of that, a semantics layer can build directories, acls, etc. This way the transactional implementation doesn't make a mess of the otherwise clean ufs-like behavior? I'm not sure how feasible that is, but it would be nice if possible. I'd also like to see things like ACLs implemented as attributes for storage purposes--while the VFS layer would expose vop_get/set_acl, et al, the top file system layer (not in traditional layering sense of the word) would convert these to internal vop_get/set_extattr calls. I'm really interested in an FS that provides transactional consistency over the inodes (or equiv) and attributes in an extensible way, allowing people to develop extensions (such as ACLs, MAC, etc) without having to understand the filestore in all its complexity. BTW, a useful thing to address would be consistency across layers in a stacked file system--something that I haven't really seen discussed anywhere... Robert N M Watson robert@fledge.watson.org http://www.watson.org/~robert/ PGP key fingerprint: AF B5 5F FF A6 4A 79 37 ED 5F 55 E9 58 04 6A B1 TIS Labs at Network Associates, Safeport Network Services To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-fs" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.3.96.991101211533.22399A-100000>