Boinc Pentathlon 2024

[StefanR5R]

MilkyWay@home:

I caught some which take 1...3 days when running singlethreaded on my lower clocked server CPUs.

Are those task actually progressing in the progress bar? I still have a few stragglers that seem to stall. Goes to X% then doesn't seem to advance any.

My long running tasks keep progressing according to boincmgr's tasks tab.

Some time ago on the MW@H message board, I have read reports of tasks which seemingly get stuck. IIRC the reports were from Windows hosts. I never got the problem myself; I only have Linux hosts which may or may not be related.
"Number crunching : Some tasks seem never-ending." (Edit, there are also some older partially related threads.)

One member reported back there that allowing BOINC to run all of the time resolved the problem for them (computing preferences, "Use at most 100 % CPU time").

--------

By the way, if you have single-threaded tasks but want them to switch to multi-threaded after the fact, you can do so with an app_config.xml (in the BOINC data directory, projects\milkyway.cs.rpi.edu_milkyway\ subdirectory):

<app_config>
    <app_version>
        <app_name>milkyway_nbody</app_name>
        <plan_class>mt</plan_class>
        <avg_ncpus>4</avg_ncpus>
        <cmdline>--nthreads 4</cmdline>
    </app_version>
    <app_version>
        <app_name>milkyway_nbody_orbit_fitting</app_name>
        <plan_class>mt</plan_class>
        <avg_ncpus>4</avg_ncpus>
        <cmdline>--nthreads 4</cmdline>
    </app_version>
    <app_version>
        <app_name>milkyway_nbody_orbit_fitting</app_name>
        <!-- for the a priori single-threaded version without plan_class -->
        <avg_ncpus>4</avg_ncpus>
        <cmdline>--nthreads 4</cmdline>
    </app_version>
</app_config>

Once this file is added (or edited), best is to shutdown and restart the BOINC client to put it to full effect.

I presume that you can also use a respectively edited app_config.xml to switch multithreaded tasks to singlethreaded after the fact, but I haven't tested this.

EDIT: If you prefer to stick with the web preferences to control multithreading, I think this only affects tasks which are downloaded after the web prefs were modified.

 

[Ken g6]

Thank you @StefanR5R! I had one that was going to take another day to finish, but now it should take less time than it's already been running.

 

[StefanR5R]

You can also thank the admin who recently performance-tested both the single-threaded binary and the MT binary in 1t mode, found that the latter was at least as good in 1t as the former, and therefore installed the MT binary as 1t app_version. That's why the --nthreads cmdline switch is now also effective at the 1t app_version.

After Kevin Roux took over as new admin, AFAIK without any help from his predecessors available to him, things were sometimes shaky at MW@h at first. But since then there has been a steady string of fixes and improvements, pretty good communication of what's happening, and response to user issues. I for one like it!

(Edit: The big downside at MW@h of course is that total-credit ranking of members and teams is almost frozen, ever since the GPU accelerated "Separation" project finished.)

 

[Orange Kid]

The Final Javelin Throw is announced.

Start: 17 May 2024 00:00 UTC
End: 18 May 2024 00:00 UTC

 

[StefanR5R]

So, about TeAm AnandTech's javelin from yesterday — is the footage already in now which it took when it passed the dark side of the moon?

 

[StefanR5R]

if you have single-threaded tasks but want them to switch to multi-threaded after the fact, you can do so with an app_config.xml

Just now I noticed that the results tables on the MW@h server show such tasks wrongly with run time = CPU time. But stderr.txt shows that these tasks ran multithreaded.

 

[Orange Kid]

30% Bonus day in the Steeplechase.
Start: 16 May 2024 00:00 UTC
End: 17 May 2024 00:00 UTC

 

[Skillz]

So, about TeAm AnandTech's javelin from yesterday — is the footage already in now which it took when it passed the dark side of the moon?

Not sure if we hit our target, but it's out there. LOL

 

[Orange Kid]

20% Bonus day in the Steeplechase
Start: 18 May 2024 00:00 UTC
End: 19 May 2024 00:00 UTC

 

[Orange Kid]

The final Javelin Throw is done.
Congrats to all, an exciting throw it was.
The medalists are...

1
TeAm AnandTech
17 571 752

2
Planet 3DNow!
10 675 648

3
SETI.Germany
10 376 816

 

[StefanR5R]

I am applauding the Javelin Throwers among us, for all their well planned and well executed work.

Also kudos to NFS@home's admin Greg who agreed to host this particular discipline of the Pentathlon and watched over the server operations to let it survive and perform during those peak loads.

 

[Skillz]

May I make a suggestion to all of those who are only running Milkyway@home for the Pentathlon and will stop working on it when it's over.

Please set your resource share to 0 on the project. What this does is only sends you enough work for as many cores/threads as you have on your system. This will lower the number of downloaded (ready-to-run) tasks in your queue and only give you tasks which you can run immediately.

So when you do abruptly stop running the project at the end of the Pent instead of having ~1000 tasks that will be left in a time-out stage, you'll only have as many as you have threads on the host.

The setting for this is located on the project web site.
https://milkyway.cs.rpi.edu/milkyway/prefs.php?subset=project


The --- next to resource share on my page indicates that I have 0 entered in that field. Do this for all locations if you have more than just the default location.

Additionally to that, it's super helpful if you abort all the tasks you downloaded but wont complete at the end of the Pent as well. This release them from their deadline and allow the tasks to get sent back out immediately to those who are still running the project. Make sure you report the aborted tasks before quitting.

If you are running PG then I suggest doing the same if you did download a huge cache then just abort the tasks so they're sent back to the server.

 

[Orange Kid]

The Final Two Disciplines have ended. Quite the Pentathlon it was.
Congrats to all the Teams that participated and congrats to all the participants. Many well fought battles .
The medalists in the Marathon...

1
TeAm AnandTech
163 853 545

2
Planet 3DNow!
113 830 007

3
SETI.Germany
73 381 429


The medalists in the Steeplechase...

1
TeAm AnandTech
25 885 981

2
Planet 3DNow!
17 549 746

3
SETI. Germany
13 776 535


...and the overall standing medalists are...

TeAm AnandTech
500

2

Planet 3DNow!
450

3

SETI. Germany
383

ONLY 350 DAYS TILL THE NEXT PENTATHLON

 

[Skillz]

Congrats Team. That's 3 consecutive Pentathlon championships in the last three years.

It was fun, but now it's time to power down and let my place cool off. It's far too hot here right now.

 

[StefanR5R]

Yesterday pschoefer posted some comments on the organization and project selection: original post in German/ machine-translated

The machine translation is of uneven quality, but I have more plans for today and won't attempt an own translation.

The decision to replace Obstacle Run by its predecessor, Marathon (and in turn move the "obstacles" over to Steeplechase which replaced Cross Country) went hand in hand with the choice of PrimeGrid for the Marathon. PrimeGrid's "fast proof" validation would not have been a good fit with "obstacle" days. MilkyWay@home, which thankfully jumped in for the iron-curtain'd SiDock on short notice, has got a validation scheme which isn't quite perfect for "obstacles" either, but that turned out to not be much of an… obstacle. (MilkyWay@home was included in the very 1st Pentathlon of 2010 and then never again until now. Previously lacking responsiveness of the project owners/ admins was one of the deciding reasons for absence of this project from Pentathlons during all these years.)

NFS@home's validator and transitioner kept up well with the load on Javelin days, and pschoefer points out that big bunker releases were thankfully (a) happening mostly early in a UTC day, and (b) were spread out into a mostly American peak starting 00:00UTC and another mostly European peak several hours later. (I believe that another factor — not just this year but also during several past Pentathlons — is that individuals who concentrate abnormously huge bunkers tend to know what happens if bunkers are emptied too sudden, nevermind too late. Plus, that you don't need to wait with file uploads until the day on which validation is desired.)

In one of the less well machine-translated parts, pschoefer says that a small number of participants evidently spent less time thinking about the real problems with project selection and the compromises involved, than they spent time to leave certain comments in the shoutbox or in some forums. (At least as far as I was following the shoutbox, there thankfully were only few loudmouths and they didn't manage to set the prevailing tone in the shoutbox.)

 

[StefanR5R]

From the shoutbox:

Congrats GPUUG for securing six sevens!

the last team that managed to take #7 everywhere later went on to win three Pentathlons 😉

Congrats and good luck to GPU Users Group!

(The previous 7/7/7/7/7/7 finish happened in 2018.)

 

[Skillz]

In one of the less well machine-translated parts, pschoefer says that a small number of participants evidently spent less time thinking about the real problems with project selection and the compromises involved, than they spent time to leave certain comments in the shoutbox or in some forums. (At least as far as I was following the shoutbox, there thankfully were only few loudmouths and they didn't manage to set the prevailing tone in the shoutbox.)

I wish they'd do away with the ability to post anonymously. That would shut a lot of people up I think.

 

[Ken g6]

I wish they'd do away with the ability to post anonymously. That would shut a lot of people up I think.

Maybe, but signing up isn't hard. I'm just surprised the Shoutbox isn't filled with spam.

 

[Skillz]

Maybe, but signing up isn't hard. I'm just surprised the Shoutbox isn't filled with spam.

You're right, its not hard; but people will be less likely to argue and complain about dumb crap if people know who they are. Sure, they can create a fake account on the SG forums to post, but SG can easily just ban them or prevent them from posting immediately until their account is verified in some way.

Most of the complaining comes from people being able to hide behind a fake name, monitor and keyboard. Remove the fake name, it'll be a lot less.