BOINC Pentathlon 2021

[Gorgon]

Whoops.

sorry about that.

[Gorgon]

4 minutes ago, Windows7ge said:

Eh, I've spent more on server grade SSD's. It'd be worth it to power a 3-phase rack in my house. :3

 

I'm just saying you can create 3 phase in your house. Obviously the power has to come from somewhere so the supply side is going to see a much higher current draw than any of the phases you create so expect to need a FAT 240V breaker in the panel.

I don’t think my 100A service and the Aluminum wiring could handle that.

[leadeater]

@Windows7ge Update on my NUMA node testing, however I'm not sure the RAC numbers have fully stabilized yet so could be throwing things off.

 

DL560 Gen8 systems with single 32 vCore VM:

• 4977.99
• 4569.95
• 4298.01
• 4119.44
• 4041.31

Average: 4401.34

 

DL560 Gen8 systems with four 8 vCore VMs:

• 1317.54
• 1219.09
• 1207.57
• 1159.39
• 1135.10
• 1133.48

Average: 1195.36

4x Average: 4781.44

 

Results in marginal performance increase using 4 VMs per host rather than 1 VM per host, on this system.

 

 

DL580 Gen9 with single 96 vCore VM:

• 18,863.07

DL580 Gen9 with three 24 vCore VM (another real customer 24 vCore VM on this host):

• 3462.86
• 3409.26
• 2970.84

Average: 3280.90

4x Average: 13123.95

 

Results inconclusive, other customer VM might be having impact on overall system performance or need to wait longer for RAC to increase to actual throughput. ~19000 vs ~13000 is too big of a difference.

[Windows7ge]

2 minutes ago, Gorgon said:

I don’t think my 100A service and the Aluminum wiring could handle that.

I know the neutral (120V) on our line coming in is also Aluminum but I think the two primary supply lines (240V) from the split phase transformer are Copper.

 

I can't remember if we have 100A service or 200A...gotta look at the big breaker at the top of the panel again. I do want to get 240V in the homelab though. Everything on UPS has to stay on 120V but everything else I'd love to step up to 240V.

 

@leadeater Good notes. So there could be some merit to what you said. Do you think you could test the 96C for a couple more days? Whatever ends up being better is the route we'll go.

 

My guess would be something kernel related if the same performance persists.

[leadeater]

3 minutes ago, Windows7ge said:

@leadeater Good notes. So there could be some merit to what you said. Do you think you could test the 96C for a couple more days? Whatever ends up being better is the route we'll go.

Well I don't plan on changing anything with those two hosts, unless that customer VM goes idle and I can shut it down and move it to different hardware. Can't live move it as it's on newer generation CPU to where I want to move it to.

[Windows7ge]

10 minutes ago, leadeater said:

unless that customer VM goes idle and I can shut it down

Or in the famous words of Bob Ross that VM could just experience a "happy accident".

[leadeater]

13 minutes ago, Windows7ge said:

Or in the famous words of Bob Ross that VM could just experience a "happy accident".

[Gorgon]

What kind of time are people seeing in SiDock?

 

Running the CurieMarieDock on BOINC tasks (<name>cmdock-boinc</name>)

i5-6400 @ 3.1GHz is showing ~ 7hrs/thread 15% through the tasks

e3-1231v3 @3.7GHz is showing ~ 9hrs/thread 50% through the tasks

 

and I can confirm that indeed the check-pointing is NOT yet working and it is unclear from their website if the "new" server is in production yet.

 

I've tried adding a few more hosts with the app limited to just a couple of threads for now but the Server is NOT sending tasks.

 

There is another project called rxdoc but the Project Status Page shows it is not active.

 

I'm using this app_config.xml file in the /var/lib/boinc-client/projects/www.sidock.si_sidockboinc-client/projects/www directory in Linux (C:\ProgramData\BOINC\projects\www.sidock.si_sidock in Windows)

<app_config>
  <app>
    <name>cmdock-boinc</name>
      <max_concurrent>2</max_concurrent>
      <gpu_versions>
        <gpu_usage>1.0</gpu_usage>
        <cpu_usage>1.0</cpu_usage>
      </gpu_versions>
  </app>
  <app>
    <name>rxdock-boinc</name>
      <max_concurrent>2</max_concurrent>
      <gpu_versions>
        <gpu_usage>1.0</gpu_usage>
        <cpu_usage>1.0</cpu_usage>
      </gpu_versions>
  </app>
  <project_max_concurrent>3</project_max_concurrent>
</app_config>

Note that the app name "rxdoc-boinc" is just a guess based on the app that is currently active.

 

This Thread at Hard Forms implies that tasks are/were only getting released twice a day.

[leadeater]

1 minute ago, Gorgon said:

Running the CurieMarieDock on BOINC tasks (<name>cmdock-boinc</name>)

i5-6400 @ 3.1GHz is showing ~ 7hrs/thread 15% through the tasks

e3-1231v3 @3.7GHz is showing ~ 9hrs/thread 50% through the tasks

Mostly around these times as well, fyi I think it jumps from around 60% to complete from what I've seen. Not been able to watch it well enough to truly confirm this as the jobs run for ages but I have seen multiple tasks go from 60% right to complete and the task run time match these.

[leadeater]

6 minutes ago, Gorgon said:

I've tried adding a few more hosts with the app limited to just a couple of threads for now but the Server is NOT sending tasks.

 

Sadly atm there is nothing to give. Not sure if by choice/design or they just actually ran out.

[Gorgon]

5 minutes ago, leadeater said:

Sadly atm there is nothing to give. Not sure if by choice/design or they just actually ran out.

See the Link to HardForms.

 

It could be they ran out or it could be in prep for the server cut-over.

 

Hopefully the prior as cutting over to a new server this close to the start of the Marathon could be a recipe for disaster.

7 minutes ago, leadeater said:

Mostly around these times as well, fyi I think it jumps from around 60% to complete from what I've seen. Not been able to watch it well enough to truly confirm this as the jobs run for ages but I have seen multiple tasks go from 60% right to complete and the task run time match these.

Thanks!

[leadeater]

3 minutes ago, Gorgon said:

See the Link to HardForms.

 

It could be they ran out or it could be in prep for the server cut-over.

 

Hopefully the prior as cutting over to a new server this close to the start of the Marathon could be a recipe for disaster.

When I was reading over the SiDock forum it sounded like it had already been done, but that could have been a prior server move. I forget what the post date of this was.

[leadeater]

@GorgonThere are ~6k tasks in the queue now, you should see them coming through to your systems now.

[Windows7ge]

9 hours ago, leadeater said:

Forgot the /s .

In unrelated news since I'm still sitting on this Supermicro H11SSL-i with a scratch on the corner that probably still works I'm thinking of turning it into a iSCSI server for iPXE.

 

Also, the eBay seller who sold me the previous EPYC 7601 that didn't work updated the listing. The title now includes the words "(Dell Only)".

 

...well ya don't say...

[leadeater]

1 hour ago, Windows7ge said:

The title now includes the words "(Dell Only)".

 

...well ya don't say...

At least you know what base server system to get if you want to get it again lol

[Windows7ge]

3 minutes ago, leadeater said:

At least you know what base server system to get if you want to get it again lol

I don't know what prices Dell is charging for 2nd hand EPYC gear. Willing to bet I'm getting away cheaper though with my setup.

[Gorgon]

Added Quick Start thread for setting up the Marathon:

Comments? Suggestions?

[Gorgon]

I'm running 2 SiDock tasks on 8 9 systems now and will ramp that up as OpenPandemics tasks finish.

[Gorgon]

Áchtung!

 

Per this post at the SiDock forms as of today the project deadline has been reduced from 3 to 2 days so if you plan on bunkering don't start too soon.

 

And per a prior post in this thread they are still in the process of moving to a new server.

[leadeater]

3 hours ago, Gorgon said:

if you plan on bunkering

Nope, too much effort for me as I still haven't gotten around to make a script for that. I spent way too much time manually doing that last year, not going to try this year.

[Windows7ge]

19 minutes ago, leadeater said:

Nope, too much effort for me as I still haven't gotten around to make a script for that. I spent way too much time manually doing that last year, not going to try this year.

Not gonna lie I was kind of secretly planning not to bunker either. Even with all my CPU it really isn't that many jobs. Plus the what? It's 2 days now? Job return time. I would barely have anything put together especially if each job takes up to or around 9 hours according to Gorgon.

[leadeater]

8 minutes ago, Windows7ge said:

Not gonna lie I was kind of secretly planning not to bunker either. Even with all my CPU it really isn't that many jobs. Plus the what? It's 2 days now? Job return time. I would barely have anything put together especially if each job takes up to or around 9 hours according to Gorgon.

Really only worth it for Javelin or Sprint however the fact that SiDock doesn't have checkpointing is going to throw in a nice extra complication. As much as I don't like it doesn't have it I also like it as well, competition wise it's going to have the affect that you either sacrifice used run time, switch early or switch late to another project.

 

We may see some teams this year focus less on marathon, maybe.

[Gorgon]

32 minutes ago, leadeater said:

Really only worth it for Javelin or Sprint however the fact that SiDock doesn't have checkpointing is going to throw in a nice extra complication. As much as I don't like it doesn't have it I also like it as well, competition wise it's going to have the affect that you either sacrifice used run time, switch early or switch late to another project.

 

We may see some teams this year focus less on marathon, maybe.

You don't lose the work in progress if you suspend though, it appears to remain in memory.

[Windows7ge]

35 minutes ago, leadeater said:

Really only worth it for Javelin or Sprint however the fact that SiDock doesn't have checkpointing is going to throw in a nice extra complication. As much as I don't like it doesn't have it I also like it as well, competition wise it's going to have the affect that you either sacrifice used run time, switch early or switch late to another project.

 

We may see some teams this year focus less on marathon, maybe.

I plan to do my usual unless asked otherwise. (Focus on the Marathon)

[leadeater]

59 minutes ago, Gorgon said:

You don't lose the work in progress if you suspend though, it appears to remain in memory.

Hmm I thought I tried that and it lost my progress 

 

Must have done something else, more than just suspend. Oh well I don't remember. Will give suspend a shot if I need to, will try it on something that doesn't have 6+ hours on it though.