Date: Wed, 15 Jul 1998 22:44:29 +0100 From: "Alton, Matthew" <Matthew.Alton@anheuser-busch.com> To: "'Terry Lambert'" <tlambert@primenet.com> Cc: FreeBSD-fs@FreeBSD.ORG Subject: RE: LFS & soft updates Message-ID: <31B3F0BF1C40D11192A700805FD48BF901776611@STLABCEXG011>
next in thread | raw e-mail | index | archive | help
> -----Original Message----- > From: Terry Lambert [SMTP:tlambert@primenet.com] > Sent: Wednesday, July 15, 1998 4:10 PM > To: Alton, Matthew > Cc: FreeBSD-fs@FreeBSD.ORG > Subject: Re: LFS & soft updates > > > Just for fun - are lfs and Kirk McKusick's Ganger-Patt soft update > > scheme necessarily mutually exclusive? I don't know anything > > about soft updates yet. Just wondered if I should look at the > > GP papers before designing LFS v.2. > > Note: Doing this work would be a good Master's Thesis project; > solving the general problem such that you could stack FS layers > arbitrarily using the same underlying graph code would probably be > worth a PhD Dissertation. > [snip] > If you want soft updates in the LFS code, which might make *some* > sense, in terms of reducing the work for the cleaner, and adding > implicit write gathering, and similar benefits to LFS, then you > should do this: > [snip] To a first approximation, I'm parsing: "Well, it would be a real fine piece of academic calisthenics but I don't think you'll get much mileage out of it in the real world." ;-) Jeez, I do rem- ember "studying" this stuff one weekend and getting good and swamped in graph theory and thinking that I had come up with a better model but then realizing that it was only better for the special class of objects that I was obsessed with at the time. As a practical expedient I'm going to see if my "generalized seg- ment bubble sort" fever dream is valid and useful for the special case of append-only operations. It takes advantage of a small predictive capability to switch the positions of consecutive seg- ments should this be deemed a win by the algorithm. If the thing requires a prohibitively far-sighted prediction or if it winds up simply restating everything I've learned so far on the subject in the general case, well, then... back to the drawing board. > 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?31B3F0BF1C40D11192A700805FD48BF901776611>