2018-06-24 13:13:09
I have tried MrPlow and it seems to work. Only a message on the node appears and I don't know what does it mean. `INFO snowblossom.node.SnowUserService sendNewBlocks Error: io.grpc.StatusRuntimeException: CANCELLED: call already cancelled` Something I have to be worried? It seems every provided share by the miner is causing that message. Thanks
Ghadras
2018-06-24 13:13:44
not an issue, just clutters the logs
Rotonen
2018-06-24 13:31:32
`creating field: snow/mainnet/snowblossom.7 memfield=true, precache=0` what's the precache thingy, something to sorta 'grep the files' and push them into the fscache, but no hard guarantees they'll stay in there?
Rotonen
2018-06-24 14:24:26
hey guys, where does the node store the block files?
bottob88
2018-06-24 14:25:57
it has a db directory you point it at and that'll be either rocksdb or lobstack based on what you configured
Rotonen
2018-06-24 15:22:56
Check out mining tuning
Fireduck
2018-06-24 15:23:06
it is just a way to load some of the field into memory
Fireduck
2018-06-24 15:27:20
ah that thing, was wondering why that's an integer - forgot about that already
Rotonen
2018-06-24 15:27:37
the 'windows be gone' ward
Rotonen
2018-06-24 15:28:18
@Fireduck is there some inherent architectural choice which would prevent one from tweaking the thread count at run time? i'm thinking about looking at making the miner autotune
Rotonen
2018-06-24 15:29:40
nope, thread coutn at runtime is good
Fireduck
2018-06-24 15:30:44
so it's some run of the mill thread pool you've gone with and not manual juggle hell?
Rotonen
2018-06-24 15:32:03
it's so confusing trying to help people fiddle with the thread counts and/or to try to explain to them why something does or does not work right now for them, it'd be nice if that could just take care of itself
Rotonen
2018-06-24 15:59:48
There was a company that made a cake mix. Just add water, mix, bake.
Fireduck
2018-06-24 16:00:08
Customers didn't like it, they thought it wasn't real cooking.
Fireduck
2018-06-24 16:00:36
So they changed the recipe so that they had to add one egg as well.
Fireduck
2018-06-24 16:00:50
So water and egg, mix, bake.
Fireduck
2018-06-24 16:01:00
People liked that a lot more.
Fireduck
2018-06-24 16:01:32
But yeah, thread autotune would be good to do.
Fireduck
2018-06-24 16:01:49
But for now maybe makes people feel involved.
Fireduck
2018-06-24 16:05:10
the egg is the `-Xmx`
Rotonen
2018-06-24 16:06:01
i suppose it could just check the available heap space and 1) put the whole field in if it fits 2) start putting it in at one (1GB?) chunk at a time, until there's < 2GB heap space left
Rotonen
2018-06-24 16:06:35
Yeah, plenty of egg here already
Fireduck
2018-06-24 16:08:00
hmm, was it only a suggestion to break the snow fields into chunks? that'd play into that nicely - i remember talk of letting people use multiple drives as otherwise the requirements eventually get expensive
Rotonen
2018-06-24 16:08:21
and i suppose that'd also allow for distributed mining, but dunno if the share schemes for pools would overcomplicate
Rotonen
2018-06-24 16:08:42
but that's essentially the same bunch of trouble as balance management of electrical grids, so fair systems already exist
Rotonen
2018-06-24 16:10:27
Yeah, breaking up the fields is planned
Fireduck
2018-06-24 16:10:38
Probably 1gb chunks
Fireduck
2018-06-24 16:10:49
Ordered list of sources
Fireduck
2018-06-24 16:11:06
Use first source that has chunk
Fireduck
2018-06-24 16:11:48
i supposed 1GB is sane for various reasons
Rotonen
2018-06-24 16:14:53
I mention it in case you have thoughts on it
Fireduck
2018-06-24 16:15:38
Could make them arbitrary but it will be complicated for already confused users
Fireduck
2018-06-24 16:16:04
Yeah, just use dd with seek and skip to get the size you want
Fireduck
2018-06-24 16:16:18
I can see that going poorly
Fireduck
2018-06-24 16:19:06
oh man, that'd be the new `rm -rf /` or 'delete system32' - almost worth it for the marketing-by-meme ensuing
Rotonen
2018-06-24 16:20:27
'order your sources by speed' would be also smart then as the miner should offload from the slowest ones
Rotonen
2018-06-24 16:20:33
and yet again a simple enough egg
Rotonen
2018-06-24 16:25:40
Especially if they can manage a ram fs and put what they can there as first source
Fireduck
2018-06-24 16:27:07
mostly currently the field does not actually stay in the fs cache as it's a monolith, so just having it in 1G chunks would help people out a lot
Rotonen
2018-06-24 16:32:09
distributed mining looks too hard to be of consideration anytime soon, though:
https://www.usenix.org/legacy/event/fast10/tech/full_papers/eshel.pdf
Rotonen
2018-06-24 16:33:51
also i think i'm done with deeper systems hacks now, was still considering to make everything `POSIX_FADV_DONTNEED` unless whitelisted, but meh
Rotonen
2018-06-24 16:34:00
but for a rough overview:
https://github.com/Feh/nocache minimize caching effects
Rotonen
2018-06-24 19:02:49
https://github.com/snowblossomcoin/snowblossom/blob/753e0500088d17aa73d5e3b145ebd49b4f0278b3/example/scripts/sample0-build-from-source.sh#L48
Line 48 in </snowblossomcoin/snowblossom/commit/753e0500088d17aa73d5e3b145ebd49b4f0278b3|753e050>
systemctl link /var/snowblossom/source/example/systemd/snowblossom-node.service
GitHub
2018-06-24 21:20:20
What should I work on?
Fireduck
2018-06-24 21:29:58
hm
Clueless
2018-06-24 21:31:46
[snowblossomcoin/snowblossom] Issue closed by cluelessperson
GitHub
2018-06-24 21:31:57
[snowblossomcoin/snowblossom] Issue closed by cluelessperson
GitHub