How's WAL utilize disk when it shares the same device with DB? Say device size 50G, 100G, 200G, they are no difference to DB because DB will take 30G anyways. Does it make any difference to WAL? Thanks! Tony > -----Original Message----- > From: Zhenshi Zhou <deaderzzs@xxxxxxxxx> > Sent: Wednesday, August 26, 2020 11:16 PM > To: Tony Liu <tonyliu0592@xxxxxxxxxxx> > Cc: Anthony D'Atri <anthony.datri@xxxxxxxxx>; ceph-users@xxxxxxx > Subject: Re: Re: Add OSD with primary on HDD, WAL and DB on > SSD > > Official document says that you should allocate 4% of the slow device > space for block.db. > > But the main problem is that Bluestore uses RocksDB and RocksDB puts a > file on the fast device only if it thinks that the whole layer will fit > there. > > As for RocksDB, L1 is about 300M, L2 is about 3G, L3 is near 30G, and L4 > is about 300G. > For instance, RocksDB puts L2 files to block.db only if it’s at least 3G > there. > As a result, 30G is a acceptable value. > > Tony Liu <tonyliu0592@xxxxxxxxxxx <mailto:tonyliu0592@xxxxxxxxxxx> > > 于2020年8月25日周二 上午10:49写道： > > > > -----Original Message----- > > From: Anthony D'Atri <anthony.datri@xxxxxxxxx > <mailto:anthony.datri@xxxxxxxxx> > > > Sent: Monday, August 24, 2020 7:30 PM > > To: Tony Liu <tonyliu0592@xxxxxxxxxxx > <mailto:tonyliu0592@xxxxxxxxxxx> > > > Subject: Re: Re: Add OSD with primary on HDD, WAL > and DB on > > SSD > > > > Why such small HDDs? Kinda not worth the drive bays and power, > instead > > of the complexity of putting WAL+DB on a shared SSD, might you > have been > > able to just buy SSDs and not split? ymmv. > > 2TB is for testing, it will bump up to 10TB for production. > > > The limit is a function of the way the DB levels work, it’s not > > intentional. > > > > WAL by default takes a fixed size, like 512 MB or something. > > > > 64 GB is a reasonable size, it accomodates the WAL and allows > space for > > DB compaction without overflowing. > > For each 10TB HDD, what's the recommended DB device size for both > DB and WAL? The doc recommends 1% - 4%, meaning 100GB - 400GB for > each 10TB HDD. But given the WAL data size and DB data size, I am > not sure if that 100GB - 400GB will be used efficiently. > > > With this commit the situation should be improved, though you > don’t > > mention what release you’re running > > > > https://github.com/ceph/ceph/pull/29687 > > I am using ceph version 15.2.4 octopus (stable). > > Thanks! > Tony > > > >>> I don't need to create > > >>> WAL device, just primary on HDD and DB on SSD, and WAL will > be using > > >>> DB device cause it's faster. Is that correct? > > >> > > >> Yes. > > >> > > >> > > >> But be aware that the DB sizes are limited to 3GB, 30GB and > 300GB. > > >> Anything less than those sizes will have a lot of untilised > space, > > >> e.g a 20GB device will only utilise 3GB. > > > > > > I have 1 480GB SSD and 7 2TB HDDs. 7 LVs are created on SSD, > each is > > > about 64GB, for 7 OSDs. > > > > > > Since it's shared by DB and WAL, DB will take 30GB and WAL will > take > > > the rest 34GB. Is that correct? > > > > > > Is that size of DB and WAL good for 2TB HDD (block store and > object > > > store cases)? > > > > > > Could you share a bit more about the intention of such limit? > > > > > > > > > Thanks! > > > Tony > > > _______________________________________________ > > > ceph-users mailing list -- ceph-users@xxxxxxx <mailto:ceph- > users@xxxxxxx> To unsubscribe send an > > > email to ceph-users-leave@xxxxxxx <mailto:ceph-users- > leave@xxxxxxx> > > _______________________________________________ > ceph-users mailing list -- ceph-users@xxxxxxx <mailto:ceph- > users@xxxxxxx> > To unsubscribe send an email to ceph-users-leave@xxxxxxx > <mailto:ceph-users-leave@xxxxxxx> > _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |