Hi Octave, According to http://www.linuxworld.com.au/index.php/id;1928390333;fp;4;fpid;1 unionfs is 'supported' on Solaris. Does that help you?
Octave Orgeron wrote On 09/12/05 10:03,: > Hi Dan, > > It was not clear to me either after reading more about the unionfs as > to if it would solve the "real-world" case that I brought up. > Block-level union is definitely required to solve this case. > Admittedly, this real world case hits pretty close to home for me. > Because of a lack of such a feature in Solaris 10, there is a good > chance I may have to support Linux in the near future. Apparently, they > have this feature in the LVM "copy-on-write" volumes. Even though it's > definitely not considered "production ready", it's one of those "we'll > save money" features that can be hard to over-look. Especially in our > case since we go through TB's of storage pretty quickly. > > But now that I know ZFS will have "space-efficient" r/w clones, I may > be able to defend the castle as it were;) I know this must be a > question you get a lot.. but here it is.. Is there a way to test out a > pre-release version of ZFS? I realize we'll have to sign NDA's and what > not, but it's definitely something I'd be interested in. > > Thanks again Dan! > > Octave > > --- Dan Price <[EMAIL PROTECTED]> wrote: > > >>On Sun 11 Sep 2005 at 01:32PM, Octave Orgeron wrote: >> >>>I do believe that lofs does solve a lot of issues around wasted >>>diskspace where zones are concerned. However, I do see a use for a >>>union style FS. While a lofs mount of a database out to seperate >>zones >> >>>is usefull for reads, it does not solve the issue of writes. Going >>>back to my example of 50 developers needing their own copy of a >>500GB >> >>>database, it does not make sense to copy the 500GB database 50 >>times. >> >>>It does make sense, however, to only store the changes they want to >>>make to that database. To extend this to zones, each developer >>could >> >>>have their very own zone and a virtual copy of the database to work >>>with. This would be extremely useful for shops that do a lot of >>data >> >>>mining and research, like financial companies:) >> >>Octave, >> >>I think you make a good point about a significant real world use case. >>However, it wasn't clear to me from (admittedly a skimmed reading) the >>unionfs documentation that it actually implements block level union; I >>was under the impression that when a file diverged, unionfs would "copy >>up" the file. The unionfs paper I found (Wright, et. al) seems a bit >>vague on this point [*]. >> >>So in this case, existing union filesystems may not do what you'd like; >>you'll get 50 copies of your giant database. But don't despair--(well, >>maybe despair a little)-- when ZFS eventually arrives, it will nicely >>solve this problem, providing space-efficient r/w "clones." >> >>I suppose someone on the alias with a linux box can do the experiment >>and let us know. >> >> -dp >> >>[*]To be clear, I've skimmed the unionfs paper referenced and the >>docs >>I could find. I'm not an authority. Please correct me if I'm >>mistaken >>here. >> >>-- >>Daniel Price - Solaris Kernel Engineering - [EMAIL PROTECTED] - >>blogs.sun.com/dp >> -------------------------------------------------------------------------- Jeff VICTOR Sun Microsystems Technical Specialist OS Ambassador -------------------------------------------------------------------------- _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org