> -----Original Message----- > From: linux-btrfs-owner@xxxxxxxxxxxxxxx <linux-btrfs- > owner@xxxxxxxxxxxxxxx> On Behalf Of Goffredo Baroncelli > Sent: Monday, 6 April 2020 5:04 AM > To: Achim Gratz <Stromeko@xxxxxxxx>; linux-btrfs@xxxxxxxxxxxxxxx > Subject: Re: [RFC][PATCH v2] btrfs: ssd_metadata: storing metadata on SSD > > On 4/5/20 10:22 AM, Achim Gratz wrote: > > Goffredo Baroncelli writes: > >> This is an RFC; I wrote this patch because I find the idea > >> interesting even though it adds more complication to the chunk allocator. > >> > >> The core idea is to store the metadata on the ssd and to leave the > >> data on the rotational disks. BTRFS looks at the rotational flags to > >> understand the kind of disks. > > > > My comment really is only about his aspect of your proposal: I would > > consider a more general way of introducing a tiering of disks so that > > one can discern between slower and faster SSD as well. > > This is a further step. I didn't mind to a tiering of fast (NVM ?) and slow SSD. > For that there are some unused fields in the superblock which can be used. > > However now my first concern is if this is > a) really useful > b) I introduced some functional regression > > Regarding a), there is an increment of performance; however stacking btrfs > over bcache leads to an even bigger gain; of course stacking btrfs over > bcache complicates the configuration for the raid setup > > Regarding b) the only regression that I found is that the logic behind the > allocation of disks in RAID5/RAID6 became more complex. But I am not sure if > this can be called regression. This is definitely useful for me. I've been meaning to implement it for ages, but was a bit afraid to try as kernel development is not my area. I've also been thinking about trying bcache, but I remember hearing about people having corruption issues when used with btrfs. That was a while ago, so I presume they are fixed now. Crazy idea - is there a way btrfs could hook into bcache or dm-cache directly? I know that's a layering violation, but btrfs doesn't use the normal layering paradigm anyway. Paul.
