Hi Daniel, Thanks for your comments, much appreciated !
I'm with you, it's a volume manager problem rather than a filesystem problem. There is some nice things to do on the volume manager layer, like this heirarchical storage but also corrupted data autocorrection, and a modern IDA kernel level implementation (cleversafe like-but-better) and probably more. Dedup should also be a part of it (I'm not sure to see the full picture here). It would be so nice to see linux ahead in that area ! You really push my will to start coding again :) (fear of the amount of work is the main break I must admit) Regards, Michael B. On Thu, Mar 12, 2009 at 11:35 AM, Daniel Phillips <phill...@phunq.net>wrote: > On Thursday 12 March 2009, Michael Keulkeul wrote: > > Hi all, > > > > Would it be possible to have a dedicated device for fast syncronious > > transactions, and latter flush to slower storage with tux3, or is this > hard > > because of the design ? > > Sort-of copy the ZIL scheme of ZFS, and be able to recover all writes > after > > a failover or a power loss. > > > > Thanks in advance for your answers ! > > > > Michael B. > > Hi Michael, > > I expect Tux3 to be pretty fast for synchronous writes even without a > scheme like ZIL. But if it were to be done, I would think a generic > virtual block device that implements a heirarchical cache would make > more sense than doing something special in the filesystem. > > In general, I would like to view most of this "accelerate disk with > a small amount of expensive persistent storage" ideas as logical > volume manager problems. Unsolved ones, to be sure, but no actual > reason for that. > > As a volume manager solution, we could map part of the volume to fast > disk and most of the volume to normal, slow disk. Then we always > allocate new writes into the fast disk area, and migrate those blocks > off somewhere else later (yes, Tux3 will have block migration). A > good Master's thesis or two in there for someone. > > Regards, > > Daniel >
_______________________________________________ Tux3 mailing list Tux3@tux3.org http://mailman.tux3.org/cgi-bin/mailman/listinfo/tux3
