Date: Fri, 21 Jun 1996 15:36:05 -0400 From: Anne Brink <anneb@svl.tec.army.mil> To: Gary Kline <kline@tera.com> Cc: black@mr.net (Ben Black), questions@freebsd.org Subject: Re: LFS anyone? Message-ID: <199606211936.PAA11013@svl.tec.army.mil> In-Reply-To: Your message of "Fri, 21 Jun 1996 10:39:54 PDT." <199606211739.KAA15153@athena.tera.com>
next in thread | previous in thread | raw e-mail | index | archive | help
> According to Ben Black: > > Is anyone using LFS and if so how well does it work and how do I newfs a > > partition to support it? LFS isn't currently working on FreeBSD, since the vm/buffer cache merge. I looked into playing with the source code, but decided I haven't done enough kernel hacking yet to be able to handle doing a decent job. The FS code is NOT easy going. I read Dr. Margo Seltzer's papers on implementing LFS in BSD, and what she discussed as needed and decided it was out of my league at this point. It would be way cool to have, though, and should work much more cleanly in FreeBSD than in straight 4.4, if I understand her main points correctly. I'm told None of the funky 4.4 file systems work, since they need to be patched to handle it. Others more knowledgable than I may care to comment. As for LFS, You can newlfs a partition and mount it, but as soon as I would write to it or run the cleaner, I panicked my OS. Not what one generally wants out of a file system. /-: > I'll second this question and follow up by asking if > anyone can post some information about the FFS and the > Log Filesystem. ((Does the LFS exist in the BSD world?)) > > Seems like the LFS would make fsck's obsolete. Yes? No? Yes, LFS does some rollbacks after a crash, BUT, the file system is not checked. It's just assumed to be fine. Fsck actually does verify that you have not lost something in a crash. Since most systems won't fsck file systems that were brought down cleanly, you only see fsck when you either come up after a crash or are doing something silly playing with clri, I think it's a good thing (TM) to have around. (-; LFS I believe assumes that the FS is ok, and then runs the cleaner in the background after you get up. That's one of the things that Margo Seltzer had to deal with when porting LFS to BSD- BSD users demanded a bit more stability in the FS than Sprite was able to offer. IMO, LFS looks very, very good for a spool partition, where you can stuff things like syslogs and mail logs and things like that. If you have lots of random file accesses, especially over NFS which caches things in blocks, not whole files, you may lose a /lot/ of the LFS functionality, and possibly get worse performance than with FFS. The LFS model isn't consistent with the concept of cylinder groups, but uses the theory of locality which helps its efficiency. As for a regular user file system? Well, I'd want to run some benchmarks. (-; Someone asked for some references: Both Dr. Margo Seltzer at Harvard and Dr. John Ousterhout at UC Berkeley have some web pages and papers about LFS. There are some very interesting strong points and weaknesses to LFS. Ousterhout designed LFS to be the file system for Sprite. Seltzer has done serious work on porting LFS to 4.4BSD (some mods have to be done to get it to work in unix) and also done some benchmarking of LFS. Dr. Seltzer has written some interesting (and readable) papers about implementing LFS in UNIX, which can be gotten from her web page at Harvard. Likewise, Ousterhout's original LFS paper, which is at UC Berkeley, is a very good read. Check out: http://www.das.harvard.edu/users/faculty/Margo_Seltzer/usenix.195/ to get started, it points to Ousterhout's stuff as well. The two don't entirely agree on everything LFS, so there's some interesting and enlightening back and forth commentary. Her '95 Usenix paper is very readable. Here's a couple more fun papers: Seltzer, Bostic, McKusick & Staelin: An implementation of a Log File System for Unix. 1993 Winter Usenix proceedings. (on Seltzer's web page, also) J. Ousterhout & F. Douglis: beating the I/O Bottleneck: The case for Log- Structured File Systems (Operating Systems Review Jan 1989) Optimal Write batch Size in Log-Structured File Systems (Carson & Setia, Computing Systems, Spring 1994) (I have a couple other refs at home, but they're the basic Sprite papers of Ousterhout's, and should be on his web page, which is pointed to by Seltzer's.) > Clues? > > gary kline > -Anne Brink -- "Ivanova is always right. I will listen to Ivanova. I will not ignore Ivanova's recommendations. Ivanova is God. And if this ever happens again, Ivanova will personally rip your lungs out!" -- Ivanova (re: Babylon 5 mantra), "A Voice in the Wilderness I"
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199606211936.PAA11013>