11:27 <beneroth> abu[m], reducing fragmentation by having specific database files for specific entities (instead of having all entities with similar blocksize in same single database file) should (theoretically) improve probability of hitting filesystem cache, agreed?

11:28 <beneroth> or is this academic, either go full benchmark and optimize for real or just keep one file per blocksize for simplicity (and storage space optimization) ?

11:29 <abu[m]> Only if those near entities are likely to be accessed at the same time

11:30 <abu[m]> More important is not to spread single objects across distant blocks

11:30 <abu[m]> they are always accessed at the same time

11:30 <beneroth> yeah agreed

11:30 <abu[m]> And to have tree nodes close to each other

11:31 <abu[m]> Best one tree per file

11:32 <abu[m]> So as far as I measured, the index trees are the critical part, not the objects

11:32 <beneroth> T, I see. Though I have way to many trees for that.. probably only worthwhile to give the most used trees their complete own file

11:32 <abu[m]> That's why 'create' imports one tree after the other

11:32 <beneroth> ok, yes makes sense

11:33 <beneroth> I do a big dump import, with a single commit at the end.. this might still spread the tree all over the place, right?

11:33 <beneroth> yeah I guess it does

11:33 <abu[m]> yes

11:33 <beneroth> I see. thanks

11:34 <abu[m]> Instead of a single commit at the end, one every 10000 objects or so seems best

11:34 <abu[m]> (at (0 . 1000) (commit))

11:34 <beneroth> T, though no influence on external symbol / tree fragmentation I guess

11:34 <abu[m]> right

11:34 <beneroth> allocated per-need / first-ask-first-served

11:35 <abu[m]> If the whole DB fits into RAM, a single commit may be fine though

11:35 <beneroth> so 'create' would be best re-allocation for lowest fragmentation, or a file per index

11:35 <abu[m]> Yes

11:35 <beneroth> thank you :)

11:39 <beneroth> afk

11:39 <abu[m]> cu ☺/

14:12 <beneroth> re

14:13 <beneroth> abu[m], I've got 205 indexes in a database here. So not sure if giving each its own file is really good idea (possible though) :D

14:15 <abu[m]> Is it really so critical?

15:51 <beneroth> abu[m], certainly not :D

15:52 <beneroth> I know the handful important index, will give them their own files, and the rest will be assigned according to size :)

15:52 <abu[m]> Should be just perfect 😉

15:52 <abu[m]> Yes, that's best

15:53 <beneroth> 😉