BOINC Pentathlon 2023

[nadoteg]

Just gotta say, as someone new to this whole thing, I gotta give mad props to @SkillzTA for coming over here and being so sportsmanlike. Granted, you all are miles ahead of us, but I think part of what makes this so fun is seeing this sort of love for the overall community. 

[SkillzTA]

7 minutes ago, leadeater said:

Sure, happy to be the first idiot to run Einstein on CPU during the Pentathlon. Plus I'm a little curious to see what I have is equivalent to GPU wise.

 

That's the kind of attitude LTT_T needs! Full steam ahead! 

[SkillzTA]

3 minutes ago, nadoteg said:

Just gotta say, as someone new to this whole thing, I gotta give mad props to @SkillzTA for coming over here and being so sportsmanlike. Granted, you all are miles ahead of us, but I think part of what makes this so fun is seeing this sort of love for the overall community. 

 

While the competition is fun. The science that is getting done is above all the most important thing. These events bring awareness to those who don't know what BOINC is and bring new crunchers to otherwise forgotten BOINC projects. I love them.

 

With that being said, one of the GPUUG members has posted some helpful information for you guys here too; but I'll let him reveal himself if he wishes.

[Captainmarino]

What do we think about bunkering YoYo in case tomorrow is bonus day?

[SkillzTA]

2 minutes ago, Captainmarino said:

What do we think about bunkering YoYo in case tomorrow is bonus day?

 

Just make sure you edit your cc_config.xml file so it doesn't report too many tasks at once. Otherwise, you'll get an error about reporting tasks and screw up the instance with getting results reported which would eventually prevent you from getting more results to crunch unless you reset/detach the project and try again. (There is a way to edit a file to fix the issue, but it's not easy, takes a lot of time and IMHO not worth it since I made the mistake on one of my hosts on the 50% bonus day and just deleted the project and started over on it. )

 

This is what you need to add:

 

<cc_config>
<options>
    <max_tasks_reported>10</max_tasks_reported>
    <report_results_immediately>1</report_results_immediately>
  </options>
</cc_config>

If you already have a cc_config.xml file, then just copy the stuff either between the <options> </options> tags and place it between the same tags (anywhere) in your cc_config or if yours doesn't contain the <options> </options> tags (It might just have the <log_flags> tags) then copy all of the stuff between including the <options> and </options> and place it before the </cc_config> flag at the bottom.

 

Then issue read config files in the BOINC Manager (under the options window in advanced mode).

[leadeater]

7 minutes ago, Captainmarino said:

What do we think about bunkering YoYo in case tomorrow is bonus day?

That would be best, should be able to get enough tasks to hold out until bonus day.

[leadeater]

6 minutes ago, SkillzTA said:

(There is a way to edit a file to fix the issue, but it's not easy, takes a lot of time and IMHO not worth it since I made the mistake on one of my hosts on the 50% bonus day and just deleted the project and started over on it. )

I got nuked by this also and tried editing the files and gave up, was far too much work since it was like all my systems. I did find another easier way to fix it, change to report only 1 tasks and keep clicking update project until all tasks report or the queue ready to report stops reducing then detach/re-add project. I've been quite lucky that last time it happened I was able to just clear it out doing the report 1 tasks without having to detach.

 

But stuff like this is what makes the Pentathlon fun and challenging.

 

LHC@Home and virtual box a few years back still gives me nightmares

[SkillzTA]

So here is the thing. The reason why P3D is doing so well with Einstein@home right now is mostly because their team consists of a lot of AMD GPUs. AMD GPUs are pretty good at Einstein@home (the special app doesnt work with it) so their GPU numbers are able to do a lot. This is why they did so poorly on Primegrid because AMD cards do very poorly on PrimeGrid compared to what Nvidia GPUs do.

However, with Nvidia cards, utilizing the special app with the amount of GPUs you guys have could really, really do some damage. You just need to install Linux and the special app. 

 

I would recommend installing Linux Mint 20.3. It's the easiest to learn, has a built in GPU Drivers GUI utility that allows you to easily install GPU drivers without having to learn the CLI. Additionally, you don't even need to install it on your SSD (or HDD, I sure hope no one here has an OS on an HDD though). You can download the Live version of Linux Mint, write it on a USB stick and boot directly into Mint from the USB stick on your computer. Then go from there. I do not recommend using this permanently as all the writes to the drive will eventually kill it, but for 3 days it would be fine.

 

I think there is a GUI package manager that will find and install BOINC for you, but if not then simply type in a terminal

 

sudo apt install boinc boinctui -y

Let it install and then you can look for the BOINC manager in the programs tab. Otherwise, you can run boinctui in the terminal to open a primitive CLI BOINC Manager to get Einstein@home attached and running. 

[Captainmarino]

16 minutes ago, SkillzTA said:

 

Just make sure you edit your cc_config.xml file so it doesn't report too many tasks at once. Otherwise, you'll get an error about reporting tasks and screw up the instance with getting results reported which would eventually prevent you from getting more results to crunch unless you reset/detach the project and try again. (There is a way to edit a file to fix the issue, but it's not easy, takes a lot of time and IMHO not worth it since I made the mistake on one of my hosts on the 50% bonus day and just deleted the project and started over on it. )

 

This is what you need to add:

 

<cc_config>
<options>
    <max_tasks_reported>10</max_tasks_reported>
    <report_results_immediately>1</report_results_immediately>
  </options>
</cc_config>

If you already have a cc_config.xml file, then just copy the stuff either between the <options> </options> tags and place it between the same tags (anywhere) in your cc_config or if yours doesn't contain the <options> </options> tags (It might just have the <log_flags> tags) then copy all of the stuff between including the <options> and </options> and place it before the </cc_config> flag at the bottom.

 

Then issue read config files in the BOINC Manager (under the options window in advanced mode).

That's interesting. I don't think I've run into that problem yet... and now I don't want to! Thanks

[leadeater]

Well that was a bust for the low ram VMs, need 3GB/thread and they only have 1GB/thread. Will put them back on Yoyo ECM.

[nadoteg]

1 minute ago, leadeater said:

Well that was a bust for the low ram VMs, need 3GB/thread and they only have 1GB/thread. Will put them back on Yoyo ECM.

Idk from where I sit, that is a lot of ram depending on the machine.

[SkillzTA]

20 minutes ago, leadeater said:

Well that was a bust for the low ram VMs, need 3GB/thread and they only have 1GB/thread. Will put them back on Yoyo ECM.

Limit the amount of tasks that run concurrently for Einstein@home.

 

<app_config>
   [<project_max_concurrent>N</project_max_concurrent>]
   [<report_results_immediately/>]
</app_config>

 

Where N is enter the number of tasks that VM/Host can run without running into memory constraints. 

 

N would be N = (Total RAM - 4GB) / 3GB

Where 4GB is reserved for the OS and whatever else is running. Adjust that depending on how much RAM the system is using without running BOINC.

 

Example: 32GB System Memory

 

(32GB - 4GB) = 28GB

28GB / 3GB = 9.33

N = 9

 

This file belongs in the

Windows:

• C:\ProgramData\BOINC\projects\einstein.phys.uwm.edu\ directory 
• (Unless you are running a different data directory)

 

Linux:

• /var/lib/boinc-client/projects/einstein.phys.uwm.edu/ directory
• (Unless you are using a different data directory)

This file probably does NOT exist. So create an app_config.xml file, paste that into it, save it. 

 

Then you can either restart BOINC or re-read the config files.

 

Then make sure BOINC is able to use up to 99% of the RAM.

Also make sure it wont lower the RAM use % when system is in use.

Also make sure it wont suspend the tasks when system is in use or Non-BOINC CPU % reaches a certain threshold. (Unless the system is being used for something else that does sometimes or all the time require CPU usage, then adjust that accordingly) 

-Make note that you'll be running a lot less cores/threads this way so CPU usage probably wont be that high for BOINC use.

[leadeater]

16 minutes ago, SkillzTA said:

Limit the amount of tasks that run concurrently for Einstein@home.

My counter argument is "but I'm lazy". Still have plenty others running the CPU tasks so not a big deal.

 

17 minutes ago, nadoteg said:

Idk from where I sit, that is a lot of ram depending on the machine.

Just 3x 24 vCPU 32GB and 3x 32 vCPU 32GB. Most of the others are 386GB-1TB ram, ~2 with 256GB.

 

I use the above method to limit ecmP2 tasks on the 256GB systems that have too many threads to handle many of those.

[SkillzTA]

Just now, leadeater said:

My counter argument is "but I'm lazy". Still have plenty others running the CPU tasks so not a big deal.

 

Just 3x 24 vCPU and 3x 32 vCPU 32GB. Most of the others are 386GB-1TB ram, ~2 with 256GB.

 

Well you'll run into the same issue with ECM and ECM P2.

 

ECM uses around 4GB.

ECM P2 uses around 8GB.

 

It just doesn't use this amount all the time. It'll use as little as 8MB, then jump to 4GB, then back to 8MB for example. When this happens, the RAM will get sent to the page file, the task will get suspended until RAM is free and thus you'll only be running a few tasks at a time when this happens. Its probably why tasks are taking so long to run for those who hit this limit. 

 

You'll want to use the same formula above for an app_config.xml file inside the yoyo@home data directory only you'll want the math to be for 4GB of RAM per task.

[leadeater]

4 minutes ago, SkillzTA said:

Well you'll run into the same issue with ECM and ECM P2.

Sorry I edited in this just after I posted it.

 

7 minutes ago, leadeater said:

I use the above method to limit ecmP2 tasks on the 256GB systems that have too many threads to handle many of those.

 

Those smaller VMs just have a different preference/location set to only ecm.

[Blqckqut]

12 hours ago, Mxyzptlk said:

My passion is folding.... Nvidia is where it is at if you want to get the most bang for the buck in folding. 

whats folding? like folding clothes?

[adm0n]

The special app on Linux is wicked fast. I can finish a single task in 90 seconds. But now lots of them don't get validated anymore. Are there too many completed tasks coming in?

[leadeater]

56 minutes ago, adm0n said:

The special app on Linux is wicked fast. I can finish a single task in 90 seconds. But now lots of them don't get validated anymore. Are there too many completed tasks coming in?

If that is happening use v0.95 instead

[adm0n]

24 minutes ago, leadeater said:

If that is happening use v0.95 instead

It started happening on windows as well, as soon as I switched over to the gamma ray task. As far as I can see, it always uploads to parts per task (so something like a filename_0 and filename_1). And they both show up in the task list, but only one gets validated. Is that how it is supposed to work?

[Idiot_icLoser55]

6 hours ago, SkillzTA said:

I would recommend installing Linux Mint 20.3. It's the easiest to learn, has a built in GPU Drivers GUI utility that allows you to easily install GPU drivers without having to learn the CLI. Additionally, you don't even need to install it on your SSD (or HDD, I sure hope no one here has an OS on an HDD though). You can download the Live version of Linux Mint, write it on a USB stick and boot directly into Mint from the USB stick on your computer. Then go from there. I do not recommend using this permanently as all the writes to the drive will eventually kill it, but for 3 days it would be fine.

Dumb suggestion, and have no idea whether this would work or what the performance affect would be. But what about running in a docker container on windows instead of the live USB. Those are already running linux.

[adm0n]

23 minutes ago, Idiot_icLoser55 said:

Dumb suggestion, and have no idea whether this would work or what the performance affect would be. But what about running in a docker container on windows instead of the live USB. Those are already running linux.

Docker utilizes the host OS, in this case windows, to run programs in an virtual environment, without actually running another OS. So if you use docker on windows, you can only run other Windows programs. If you want to run Linux programs, you'd need to first set up a VM, on which you then can run a docker container. At that point, you might as well run everything on the VM anyway.

I've never used GPU Passthrough with WSL, but it would be interesting, if that would be possible for this.

[Favebook]

4 hours ago, Blqckqut said:

whats folding? like folding clothes?

I don't know if you are joking or not. If you are - that's a good one!

 

If you are not, Folding is similar to BOINC but more user-friendly and less configurable way of helping battle diseases:
https://foldingathome.org/

 

[SkillzTA]

4 hours ago, adm0n said:

It started happening on windows as well, as soon as I switched over to the gamma ray task. As far as I can see, it always uploads to parts per task (so something like a filename_0 and filename_1). And they both show up in the task list, but only one gets validated. Is that how it is supposed to work?

 

Validation takes time. Your wingman has to complete the same task and return it. It'll take a little while for the valids and pendings to kind of level out. Stay on the Linux app.

[adm0n]

1 minute ago, SkillzTA said:

Validation takes time. Your wingman has to complete the same task and return it. It'll take a little while for the valids and pendings to kind of level out. Stay on the Linux app.

Oh, are all computations done twice in order to validate them? Thanks for telling me

[leadeater]

19 minutes ago, adm0n said:

Oh, are all computations done twice in order to validate them? Thanks for telling me

Not all, depends on the project and application/task. It's up to the project owner and their requirements.