It's not you, it's me. My adventure with Vast.AI

This is kind of some rambling observations and a plea for help, since right now I'm feeling pretty dumb. I know I'm missing something quite obvious and have no idea what. Spent about 3 hours off and on trying to get things going (between shots of rum at the end).
So to start off I am by no means a Linux expert, so this is most likely user error.

I saw a machine for rent at less than 10 cents an hour with 2 1060s, so figured I'd at least give it the benefit of the doubt. If anything I could learn a thing or two. For the record I'm talking about this.
Apparently they use Docker images, so I figured I'd throw this NVIDIA-savvy one on there.

Right off this was new to me. Apparently the home directory is simply the contents of the boinc folder. There was no remote_hosts.cfg, which already had me wondering what was going on, so I created one. Edited the gui_rpc_auth file, while I was at it, then restarted boinc with the only command I know.
/etc/init.d/boinc-client restart which seemed to work.
Something to note here is that I did not need sudo for any of the commands - I had to install Nano and curl in order to grab the public IP address and actually edit anything.
What a disappointment when I tried to connect remotely and failed. I figured they had a firewall, and since I was so lost anyway I decided to just throw a Ubuntu docker image on it instead. The boinc docker was far too confusing simply because I didn't know where to input the commands starting with docker run -d \. They allow you to use ssh or pass the arguments directly onto docker, and I figured SSH was what I wanted. Maybe not?

So - get Ubuntu docker image (I think 1804) up and running, Install Boinc. This looks much more familiar. Edit /etc/boinc-client/gui_rpc_auth.cfg and /etc/default/boinc-client so Boinc would start allowing remote access and I had the right password. Still no go. I figured fine, let's try, just for giggles, port 80. I was of course restarting using the command I gave above with each change.
So, by the end of it I had this in the Boinc-client:
BOINC_OPTS="--allow_remote_gui_rpc --gui_rpc_port 80"
Still nothing.
Even more odd is when I'm actually in the /etc/boinc-client directory, typing Boinccmd --get_state gave me no output. I tried typing Boinc on its own and it seemed to start the actual manager (I thought I had restarted it earlier). But I was getting all of the output about ram, don't do work while gpu is suspended, etc. Still no go on connecting with ports 80, 8080, 43 or the port that I was given to SSH in with, not that I thought that would do much good. I didn't actually catch if it even picked up on the gpu since there was so much info flying by, but I digress.

Am I having a blond moment here or is this really going to be that impossible to connect to? Am I even editing the right files? I'm able to connect to it through ssh and it's able to make http connections to get its own public IP, so it has to be me. I just have no clue where I'm screwing up. Unless it really is Blocking all Boinc connections and is a lost cause.

Any help appreciated here.
I'm going to press the return key to send this, then have another drink. :)

Did you add this to your cc_config.xml?

<allow_remote_gui_rpc>1</allow_remote_gui_rpc>


If 1, allow GUI RPCs from any remote host (see Controlling BOINC remotely).

I did not. I will give that a shot and see what happens.
Tried that and still getting the same issues.
I wonder if port forwarding via ssh would do anything?
Still so totally new to this. Maybe I got in over my head.

I don't run Linux from the command line because that is above my pay grade.

From the command line to start/stop BOINC use:

sudo service boinc-client stop
sudo service boinc-client start

The gui command that Z mentioned for your cc_config requires BOINC to be restarted for it to become active.

Quote:

---

Originally Posted by Bryan

I don't run Linux from the command line because that is above my pay grade.

From the command line to start/stop BOINC use:

sudo service boinc-client stop
sudo service boinc-client start

The gui command that Z mentioned for your cc_config requires BOINC to be restarted for it to become active.

---

Okay, I must have been restarting the wrong part of Boinc. Upon further inspection Boinc isn't able to access the GPUs inside the Ubuntu docker image, despite vast claiming gpu drivers are installed. Can't remember the specific error message and apparently the log didn't save.
So maybe I need to try the Boinc docker image again regardless.
Not all that experienced in the terminal myself.
Further inquiries tell me that they are behind a firewall, but I should be able to use local port forwarding to theoretically bounce 31416 back to me and make it available using PuTTY? Have never done that before.

Sometimes I'm too stubborn for my own good...

Finally figured it out last night!
Using image: nvidia/opencl:devel-ubuntu18.04
Installed the latest boinc (7.16.1). 1070ti is happily crunching Collatz at 7.3 cents an hour, at least until I get outbid. I was not able to figure out remote.
I have the following in /root/onstart.sh, since when the instance comes back up after a higher bidder stops/destroys theirs, I can start crunching again.
/etc/init.d/boinc-client start
Does that look right?
We'll see how this actually holds up, since I'm only 12 hours in. You can also get on demand (so no bidding involved) instances.
1070ti seems to be holding right around 410-420 seconds per WU with optimizations, about 35-40 slower than my 1080.

How big of a difference do the RTX cards like 2070 and 2080 actually make over the 1070, 1070ti, 1080, and 1080ti? 1080ti seems to still be a very sexy card all things considered...I guess it makes a difference on what project you're using and what parts of the gpu are being stressed the most. Not always about clock speed.

I think the 2080 makes around 15M/day on Collatz and my 1080Tis make around 11M so there is a pretty big difference. Obviously the 2080Ti would do even better.

Quote:

---

Originally Posted by wolfman1360

Finally figured it out last night!
Using image: nvidia/opencl:devel-ubuntu18.04
Installed the latest boinc (7.16.1). 1070ti is happily crunching Collatz at 7.3 cents an hour, at least until I get outbid. I was not able to figure out remote.
I have the following in /root/onstart.sh, since when the instance comes back up after a higher bidder stops/destroys theirs, I can start crunching again.
/etc/init.d/boinc-client start
Does that look right?
We'll see how this actually holds up, since I'm only 12 hours in. You can also get on demand (so no bidding involved) instances.
1070ti seems to be holding right around 410-420 seconds per WU with optimizations, about 35-40 slower than my 1080.

How big of a difference do the RTX cards like 2070 and 2080 actually make over the 1070, 1070ti, 1080, and 1080ti? 1080ti seems to still be a very sexy card all things considered...I guess it makes a difference on what project you're using and what parts of the gpu are being stressed the most. Not always about clock speed.

---

So, $1.75 a day for a VM running a 1070ti? I think that's cheaper than my power bill without having had to even buy any hardware. :)

Where are you renting the GPU Wolfman?

Quote:

---

Originally Posted by Bryan

Where are you renting the GPU Wolfman?

---

I'm renting it from here.
Regarding the power bill...definitely feeling you. A few years ago I was living in a place where the utilities were included. Now that I'm paying mine I'm noticing a difference with everything running 24/7.

Maybe I should put all of this into my bedroom and sleep with the window open in these cold Canadian winters...nothing like -40 F to help keep things cool.

For $1.20/hour you can rent 8X 2080Ti. That would put up some numbers on Collatz =))