Time  Nick       Message
00:54 lhofhansl  Hi all... For kicks I instrumented the calls to db->loadBlock. It turns out that sqlite maxes out around 5.5k/s. (At that rate more emerge threads won't help).
00:55 lhofhansl  The frustum for a viewing_range of 1000 with fov 72 has about 130k blocks. Thus the best we can get is about 24s to load all the blocks. And with the right setting I get pretty close to that.
00:56 lhofhansl  I expect this to be worse with Postgres, as there is an actual rountrip to the server involved for each block.
00:57 lhofhansl  The only way to improve this is collect multiple blocks in one query ... SELECT * FROM blocks where pos IN (..., ..., ...). That'd be a major refactor, though.
00:58 lhofhansl  (note that most blocks are not sent to the client due to the optimize distance and server occlusion culling, but the DB, nonetheless has to read the blocks)
01:00 lhofhansl  And #13225 is ready for another look (last 5.7.0 blocker)
01:00 ShadowBot  https://github.com/minetest/minetest/issues/13225 -- Fix infinite viewing_range by lhofhansl
01:03 lhofhansl  With "And with the right setting I get pretty close to that." I mean within 2x, so less that 50s.
01:13 lhofhansl  Interestingly if I start from an empty map and let the scene (viewing_range = 1000, fov = 72) generate, I find 330k blocks in the database. Not what I had expected.
01:14 lhofhansl  Should have been around 2/3 * pi * ceil(1000/16)^1 * (1 - cos(72/2))
01:15 lhofhansl  ^3
10:46 Desour     !tell lhofhansl 5.5kB/s sounds *very* slow (i.e. slower than internet). or is that 5.5k calls (and hence 5.5k blocks) per second? and did you measure only the time it took inside db->loadBlock or did you also include the time outside of that (which would mean we could get more bandwidth by calling oftener)?
10:46 ShadowBot  Desour: An error has occurred and has been logged. Please contact this bot's administrator for more information.
10:46 Desour     uh oh
10:47 Desour     ~tell lhofhansl 5.5kB/s sounds *very* slow (i.e. slower than internet). or is that 5.5k calls (and hence 5.5k blocks) per second? and did you measure only the time it took inside db->loadBlock or did you also include the time outside of that (which would mean we could get more bandwidth by calling oftener)?
10:47 ShadowBot  Desour: An error has occurred and has been logged. Please contact this bot's administrator for more information.
11:08 pgimeno    yeah I interpreted it as being 5.5 Kblocks/s
11:10 pgimeno    still, I would test with PostgreSQL, SQLite is too crappy performance-wise
12:09 MTDiscord  <Andrey01> Rebased #13020 and #12828 now
12:09 ShadowBot  https://github.com/minetest/minetest/issues/13020 -- 3d line rendering. by Andrey2470T
12:09 ShadowBot  https://github.com/minetest/minetest/issues/12828 -- Add vector variation for ContentFeatures 'visual_scale' property. by Andrey2470T
13:11 MTDiscord  <x2048> Thank you!
13:44 MTDiscord  <MisterE> So, I remember that sfan said that desour's sound refactor was to be prioritized for 5.7, and desour rebased and seems to have fixed conflicts. Is it still on the table for 5.7 despite the feature freeze?
13:47 Desour     I'd say no. we're in feature freeze. and that PR is big. it seemed quite unrealistic for me in the first place to get it into 5.7 (at least back then when 5.7 was planned to be released in february or so)
14:08 rubenwardy no, refactors like that are not allowed to be merged during a feature freeze
14:10 rubenwardy Desour being a core dev definitely helps with the probability of that PR landing
15:51 lhofhansl  sfan5: I added two more questions to #13225, mind have a quick lock? Namely: (1) should we do occlusion culling at all? (2) if we do, gate it with the raytraced culling setting?
15:51 ShadowBot  https://github.com/minetest/minetest/issues/13225 -- Fix infinite viewing_range by lhofhansl
16:05 lhofhansl  !tell Desour That was 5.5k blocks/s
16:05 ShadowBot  lhofhansl: OK.
16:08 lhofhansl  Going to merge #13225 soon.
16:08 ShadowBot  https://github.com/minetest/minetest/issues/13225 -- Fix infinite viewing_range by lhofhansl
16:09 rubenwardy ShadowBot is nicer than MinetestBot
16:09 ShadowBot  rubenwardy: Error: not enough values to unpack (expected 2, got 1)
16:09 rubenwardy lol
16:09 rubenwardy ~tell Desour that was 5.5k blocks/s
16:09 ShadowBot  rubenwardy: OK.
16:09 rubenwardy huh
16:09 Desour     !tell lhofhansl test
16:09 ShadowBot  Desour: OK.
16:09 rubenwardy MinetestBot says "tell them yourself" or  "sure, whatever"
16:09 rubenwardy oh, it's not here
16:10 Desour     !tell _Desour test
16:10 ShadowBot  Desour: OK.
16:10 Desour     earlier shadowbot didn't want to do the tell
16:12 Desour     lhofhansl: how did you measure the time? did you measure the calls per second, or the time inside the function`
16:12 Desour     ?
16:14 lhofhansl  Desour: I added a scope profiler in EmergeThread::getBlockOrStartGen (on the loadBlock branch). Now I can how often it is called and how much time it takes on average (about 150 microsecs on my machine)
16:15 Desour     ah, ok
16:21 lhofhansl  Something we might want to do anyway... I'll file a PR :)
16:21 lhofhansl  Help with testing all the emerge settings.
16:37 lhofhansl  13225 is done.
17:14 lhofhansl  Desour: #13314
17:14 ShadowBot  https://github.com/minetest/minetest/issues/13314 -- Add some instrumentation by lhofhansl
18:40 Desour     merging #13313 in 15 min
18:40 ShadowBot  https://github.com/minetest/minetest/issues/13313 -- Fix rendertarget size being saved as screensize by Desour
18:57 Desour     merged.