2018-12-23 00:47:47
yeah, that should be 90k or so

Fireduck
2018-12-23 00:47:50
running enough threads?

Fireduck
2018-12-23 00:47:58
I'd expect to need 200 or 300

Fireduck
2018-12-23 01:27:26
I was getting that on my 3xx motherboard

Fable
2018-12-23 01:27:37
but not on the z270

Fable
2018-12-23 01:28:14
270 is less than 300

Fireduck
2018-12-23 01:28:32
Wait, what?

Fable
2018-12-23 01:28:44
I am talking about motherboard series, not threads

Fable
2018-12-23 01:29:49
where do you put nolisten=true for arktika to prevent port binding

Fable
2018-12-23 01:30:52
It doesn't do anything when you put it in the miner.conf

Fable
2018-12-23 01:31:17
A dumb joke

Fireduck
2018-12-23 01:31:32
One sec, it might be called something else

Fireduck
2018-12-23 01:32:52
nolisten=true in the arktika config file

Fireduck
2018-12-23 01:33:28
not preventing port binding

Fable
2018-12-23 01:33:33
when I do that

Fable
2018-12-23 01:34:13
or at least the second instance wont run

Fable
2018-12-23 01:34:16
sorry

Fable
2018-12-23 01:34:48
what error do you see?

Fireduck
2018-12-23 01:34:53
hold on

Fable
2018-12-23 01:35:11
and what version?

Fireduck
2018-12-23 01:38:51
1.3.3

Fable
2018-12-23 01:39:35
Exception in thread "main" java.io.IOException: Failed to bind
at io.grpc.netty.NettyServer.start(NettyServer.java:252)
at io.grpc.internal.ServerImpl.start(ServerImpl.java:162)
at io.grpc.internal.ServerImpl.start(ServerImpl.java:77)
at snowblossom.miner.Arktika.<init>(Arktika.java:147)
at snowblossom.miner.Arktika.main(Arktika.java:65)
Caused by: java.net.BindException: Address already in use: bind
at sun.nio.ch.Net.bind0(Native Method)
at sun.nio.ch.Net.bind(Unknown Source)
at sun.nio.ch.Net.bind(Unknown Source)
at sun.nio.ch.ServerSocketChannelImpl.bind(Unknown Source)
at io.netty.channel.socket.nio.NioServerSocketChannel.doBind(NioServerSocketChannel.java:128)
at io.netty.channel.AbstractChannel$AbstractUnsafe.bind(AbstractChannel.java:558)
at io.netty.channel.DefaultChannelPipeline$HeadContext.bind(DefaultChannelPipeline.java:1358)
at io.netty.channel.AbstractChannelHandlerContext.invokeBind(AbstractChannelHandlerContext.java:501)
at io.netty.channel.AbstractChannelHandlerContext.bind(AbstractChannelHandlerContext.java:486)
at io.netty.channel.DefaultChannelPipeline.bind(DefaultChannelPipeline.java:1019)
at io.netty.channel.AbstractChannel.bind(AbstractChannel.java:254)
at io.netty.bootstrap.AbstractBootstrap$2.run(AbstractBootstrap.java:366)
at io.netty.util.concurrent.AbstractEventExecutor.safeExecute(AbstractEventExecutor.java:163)
at io.netty.util.concurrent.SingleThreadEventExecutor.runAllTasks(SingleThreadEventExecutor.java:404)
at io.netty.channel.nio.NioEventLoop.run(NioEventLoop.java:465)
at io.netty.util.concurrent.SingleThreadEventExecutor$5.run(SingleThreadEventExecutor.java:884)
at io.netty.util.concurrent.FastThreadLocalRunnable.run(FastThreadLocalRunnable.java:30)
at java.lang.Thread.run(Unknown Source)

Fable
2018-12-23 01:41:40
1.3.3? That was ages ago

Fireduck
2018-12-23 01:41:50
Yes

Fable
2018-12-23 01:41:53
no listen is only in newer ones

Fireduck
2018-12-23 01:42:34
Do I have to bazel build it or is there a link

Fable
2018-12-23 01:44:44
updating that now

Fireduck
2018-12-23 01:44:53
if it is older than 1.4.2, wait a minute and download again

Fireduck
2018-12-23 01:55:32
The new version prints nice IO use messages

Fireduck
2018-12-23 01:56:21
still only get 18k on my z270

Fable
2018-12-23 01:56:31
cpu benches 300k

Fable
2018-12-23 03:20:04
Okay, this is going to be too much for me

Fable
2018-12-23 03:20:20
I now realize that you need a server setup to do this

Fable
2018-12-23 03:20:58
Can't work without xeons

Fable
2018-12-23 03:21:33
The cpu load is too high for a simple rig setup

Fable
2018-12-23 03:21:34
I've had to move m.2 drives around

Fireduck
2018-12-23 03:21:48
mostly I've have good luck with modern boards with m.2 on board over pcie adapters

Fireduck
2018-12-23 03:21:52
well when you run multiple instances of the miner you cant hit the single instance benchmark speed

Fable
2018-12-23 03:22:06
why would you run mujltiple instances?

Fireduck
2018-12-23 03:22:11
and the other instances start to slow

Fable
2018-12-23 03:22:17
well you wouldnt if you didnt have to

Fable
2018-12-23 03:22:40
So how would you run multiple nvmes on one arktika instance?

Fable
2018-12-23 03:22:48
what I mean is that with arktika, you should be able to configure a single one to use whatever resources you have

Fireduck
2018-12-23 03:22:57
yeah, split up the chunk files and put some on each nvme

Fireduck
2018-12-23 03:23:02
one layer for each nvme

Fireduck
2018-12-23 03:23:25
that should work well, but I admit I haven't tried that personally

Fireduck
2018-12-23 03:23:42
I didnt think that would work I thought that was for when the field started to get too large

Fable
2018-12-23 03:23:52
but now, yes, thanks

Fable
2018-12-23 03:24:07
it is also for distributing the load among block devices

Fireduck
2018-12-23 03:24:43
block devices?

Fable
2018-12-23 03:24:54
in this case the nvmes

Fireduck
2018-12-23 03:25:01
but could be ssd

Fireduck
2018-12-23 03:25:04
snowplows?

Fable
2018-12-23 03:25:05
or remote ram servers

Fireduck
2018-12-23 03:25:14
dumb joke

Fable
2018-12-23 03:25:19
ha

Fireduck
2018-12-23 03:26:58
I also thought you said that you couldn't use range parameter with layertype=file

Fable
2018-12-23 03:27:12
so that led to my assumption

Fable
2018-12-23 03:30:02
yeah, no range

Fireduck
2018-12-23 03:30:06
it uses whatever files are there

Fireduck
2018-12-23 03:30:10
so you have to move things around

Fireduck
2018-12-23 03:30:18
how would splitting up the chunks give you the full speed for each device running one miner isntance?

Fable
2018-12-23 03:30:20
I plan on adding the range in though, it will make it easier to tune

Fireduck
2018-12-23 03:30:29
yeah okay

Fable
2018-12-23 03:31:21
I guess, so you are renting resources to do this then?

Fable
2018-12-23 03:31:28
and just testing small samples

Fable
2018-12-23 03:36:26
I have a small memory mining fleet

Fireduck
2018-12-23 03:36:57
It is a single Dell r900 with 256gb ram and network workers

Fireduck
2018-12-23 03:37:06
Total about 1mh

Fireduck
2018-12-23 03:37:34
how would you configure that

Fable
2018-12-23 03:37:38
with the snow files

Fable
2018-12-23 03:37:51
just two snowfields?

Fable
2018-12-23 03:40:58
A single field 7

Fireduck
2018-12-23 03:41:23
More network limited than memory bus

Fireduck
2018-12-23 03:41:45
So it was registered ram

Fable
2018-12-23 03:43:55
splitting the chunks up still only gives you the speed of one 512 nvme?

Fable
2018-12-23 03:45:40
so yeah

Fable
2018-12-23 03:46:05
Should get combined speed

Fireduck
2018-12-23 03:46:06
not really viable unless the miner can cut cpu usage for same thread count

Fable
2018-12-23 03:46:16
60k

Fable
2018-12-23 03:46:27
What is the CPU?

Fireduck
2018-12-23 03:46:55
i9 7940x

Fable
2018-12-23 03:47:02
benches 860k

Fable
2018-12-23 03:47:12
Ok, cool

Fireduck
2018-12-23 03:47:51
just leave the whole snow field?

Fable
2018-12-23 03:47:57
How many nvme do you have?

Fireduck
2018-12-23 03:48:02
4

Fable
2018-12-23 03:48:21
Put 32 chuncks on each

Fireduck
2018-12-23 03:49:00
im running field 8

Fable
2018-12-23 03:49:34
Have you seen the benchmark_layer option?

Fireduck
2018-12-23 03:49:39
fake

Fable
2018-12-23 03:49:40
yes

Fable
2018-12-23 03:50:01
Fake layer is a different thing

Fireduck
2018-12-23 03:50:09
Fake layer tests cpu

Fireduck
2018-12-23 03:50:13
okay

Fable
2018-12-23 03:50:16
then no

Fable
2018-12-23 03:51:40
benchmark_layer=0

Fireduck
2018-12-23 03:51:45
benchmark_threads=512

Fireduck
2018-12-23 03:51:57
that will just do as much IO on that layer as possible using that many threads

Fireduck
2018-12-23 03:52:04
that is a good way to see how much IO you can do on that layer

Fireduck
2018-12-23 03:52:24
so that would be good to see if your NVME can read as fast as spec says regardless of any CPU issues

Fireduck
2018-12-23 03:52:29
so delete other layers temporarily to test one

Fable
2018-12-23 03:52:33
in the config

Fable
2018-12-23 03:52:52
you can leave them in

Fireduck
2018-12-23 03:53:02
set benchmark_layer to the one you want to test

Fireduck
2018-12-23 03:55:27
read ops/s and read_bw

Fable
2018-12-23 03:55:36
582310 , 2275

Fable
2018-12-23 03:55:48
but says no hashing

Fable
2018-12-23 03:56:12
yeah, it is doing ont hashing

Fireduck
2018-12-23 03:56:18
no hashing

Fireduck
2018-12-23 03:56:22
just testing bandwidth

Fireduck
2018-12-23 03:56:34
I only did 1 benchmark layer though

Fable
2018-12-23 03:56:35
should I do 4

Fable
2018-12-23 04:01:09
yeah, it only supports one at time

Fireduck
2018-12-23 04:01:21
which is enough to let you know if the layer works as fast as you think it should

Fireduck
2018-12-23 04:01:51
if you can split the chunks up on a rig whats the point of 512gb drives

Fable
2018-12-23 04:02:16
I should be able to put the whole field on each drive and get the full hashrate with one miner instance, no?

Fable
2018-12-23 04:02:25
well, that will continue to work up to a much larger field size if split 4 ways

Fireduck
2018-12-23 04:02:41
still only 60k though

Fable
2018-12-23 04:02:59
if arktika supports that sort of load balancing, yes. You should absolutely be able to get nearly 400k with the rig you have

Fireduck
2018-12-23 04:03:17
assuming those nvmes are attached such that they can all be read at full rate at once

Fireduck
2018-12-23 04:03:22
so put the whole field on each drive?

Fable
2018-12-23 04:03:36
no, put 1/4th on each drive

Fireduck
2018-12-23 04:04:02
in some situations, you do need the data spread across the drive to get full rate, but I don't think we are there

Fireduck
2018-12-23 04:04:30
I think 64gb on each drive will be enough to spread it across whatever banks or chips or whatever there are

Fireduck
2018-12-23 04:05:03
So if you run four arktika instances you don't get good rates? (one per drive)

Fireduck
2018-12-23 04:05:18
can you show me one of the configs you were using for that?

Fireduck
2018-12-23 04:05:26
That should work as well

Fireduck
2018-12-23 04:08:30
the cpu load was too high

Fable
2018-12-23 04:08:45
my system has 128gb ram, I was running the bat with:

Fable
2018-12-23 04:08:59
@echo off
title miner
start /B /LOW java -jar -Xmx100g Arktika.jar configs/miner.conf
pause

Fable
2018-12-23 04:09:21
and then 512 threads on each instance

Fable
2018-12-23 04:09:32
but the cpu load is way too high with that many threads

Fable
2018-12-23 04:09:39
but it takes that many threads to get 60k

Fable
2018-12-23 04:09:47
interesting

Fireduck
2018-12-23 04:09:47
otherwise I only get 20~k

Fable
2018-12-23 04:09:55
I haven't done much arktika testing on windows

Fireduck
2018-12-23 04:10:02
I don't know why the cpu load is so high though

Fable
2018-12-23 04:10:12
unless you are mining xmr

Fable
2018-12-23 04:10:13
but that -Xmx100g is way too high if you aren't trying to fit it in ram

Fireduck
2018-12-23 04:10:21
probably remove that entirely

Fireduck
2018-12-23 04:10:24
okay

Fable
2018-12-23 04:10:38
can you show me the miner.conf?

Fireduck
2018-12-23 04:11:01
mine_to_address=e82d6vzegevqdey6y05625hfsm87jgtrl5k85j42
network=snowblossom
node_host=2338
pool_port=23380
memfield=false
min_depth_to_disk=6
layer_count=1
layer_0_type=file
layer_0_path=f:\\snow
layer_0_threads=64
snow_path=f:\\Snow
pool_host=http://findsnow.org
selected_field=8
nolisten=true

Fable
2018-12-23 04:11:41
that was from before, except I changed the threads to lower cpu usage

Fable
2018-12-23 04:12:46
ok. Windows has a thing where it likes each thread to have its own file descriptor open, I don't think I did that in arktika like I did in poolminer

Fireduck
2018-12-23 04:12:59
might be related, but hard to say

Fireduck
2018-12-23 04:13:36
I should try and balance the info queues with threads per layer right

Fable
2018-12-23 04:13:50
so if they are out of balance i need to change my threads on a layer?

Fable
2018-12-23 04:14:12
Im still only getting 70K~

Fable
2018-12-23 04:14:18
64chunks on each

Fable
2018-12-23 04:14:50
yeah, in your case the queues should be about even

Fireduck
2018-12-23 04:14:59
but they will vary a good bit in normal operation

Fireduck
2018-12-23 04:15:24
but still only 70k, so what is the solution you think?

Fable
2018-12-23 04:15:43
are all the CPU cores maxed out at 70k?

Fireduck
2018-12-23 04:15:49
what linux distro should I use on the rig to run multiple instances

Fable
2018-12-23 04:16:25
and yes

Fable
2018-12-23 04:16:26
I like debian, but the only things that actually matter are the kernel and the java version

Fireduck
2018-12-23 04:16:37
all cpu cores maxed with 512 threads and 4 drives

Fable
2018-12-23 04:16:39
so any distro

Fireduck
2018-12-23 04:16:53
so painful to use on rigs though

Fable
2018-12-23 04:17:27
if you don't mind, keep it windows and I'd like to try some things with you over the next few days

Fireduck
2018-12-23 04:17:28
unless you have everything automated and can control all rigs at once

Fable
2018-12-23 04:18:55
hopefully it can run all the drives with the full field on one instance to save ram usage

Fable
2018-12-23 04:19:08
but that would interfere with how it already is setup I think so

Fable
2018-12-23 04:26:52
That would be a pretty intense change to arktika and I think the main problem is about the file descriptors

Fireduck
2018-12-23 04:27:05
So might try something else first

Fireduck