Add some form of Automatic Storage Tiering (AHSM)

Latest response

This can potentially bring enormous performance gains at relatively small cost and could give RHEL a unique selling point. What I mean by Automatic Storage Tiering is the movement of blocks between fast and slow storage, based on their usage (SSD <-> SCSI <-> SATA). There are allready some things floating around: There is flashcache from facebook, dm-cache, and this is also a nice initative: . I like the approach to make it a device-mapper module. This means it can work with LVM, raw disk, etc. another idea could be to bring it into the new gluster product, allthough I don't know wether that would fit.


This sounds like something I posted a few months ago.  I was speaking entirely about BTRFS taking on board this capability though.


Is it similar enough?


I would like it to be build into Device-Mapper, LVM. (block-level) Dedup should also belong there imho, just like snapshotting and (remote) mirroring (think dm-replicator). This kind of functionality does not belong in the filesystem, but below it. This way the customer can still choose the best FS for the job or even RAW for i.e. DB's.


I know a lot of people working on DM/LVM have @redhat in their email address, so RedHat is able to initiate these kind of functionalities.

I see your point.


BTRFS does extend down below the normal perception of filesystem, however, to effectively roll up some of the work that LVM performs at present.  It also provides a fair amount of stuff that LVM can't at the moment.  However - it's also not enterprise ready yet.


LVM won't be going away any time soon, but with all these functional elements being baked into enxt generation filesystems, the clock may have started ticking.  For sure, it'll be the best way to underlay my ext4 file systems etc. for now.