LHC@home

"2025-09-26 15:27:27 (19450): called boinc_finish(0)" is not a "similar log" but a success log.

The other looks like an error log, but what is the final status of the task + does it get credited ?

This is just a cosmetic issue related to timestamp generation for the message. It will hopefully be fixed in the next release.

Edit: The jobs are successful even if this message appears.

Dear Volunteers,

It’s a pleasure to meet you! I’m a postdoctoral researcher in the HL-LHC team, working with Frederik on the development and use of Xboinc. (You may have noticed me as “camontan” in the names of these first Xtrack batches.)

First of all, thank you for your trust and patience during these crucial early stages of deployment. Rolling out scientific applications is never straightforward, and surprises are always around the corner. We truly appreciate having so many enthusiastic volunteers helping us smooth out the rough edges.

We are now gradually scaling up the upload of the many particle-tracking jobs we need. Going forward, WorkUnits should take at most 10 hours (on paper), with a more realistic average runtime of 1 to 4 hours on a modern CPU.

Regarding the issues you’ve reported:

• Progress bar not updating correctly: this seems linked to how BOINC internally handles job completion fractions. I’m investigating and will improve this for the next release.
  
• boinc_finish(0) in stderr: as Laurence said, it is a cosmetic issue in the BOINC internal code, we expect to have it fixed with the next app release.
  

And now, a bit about what you’re actually tracking!

The particles are being simulated across many different configurations of a realistic High Luminosity LHC lattice, including complex effects such as beam-beam interactions (the influence each particle beam feels from the other). Running these simulations across multiple configurations is both time consuming and computationally expensive, but it’s essential for identifying the best machine settings.

In this campaign, we’re also building a large dataset of scenarios to train machine learning surrogate models, which will dramatically speed up the exploration of optimal configurations. Over the next few days, I’ll be uploading many more of these jobs. I’ll also be presenting the project at the upcoming HL-LHC collaboration meeting, where I’m sure we’ll attract even more colleagues eager to run exciting simulations with your help, all for the advancement of accelerator physics :D

Once again, thank you for your contributions and support. This project wouldn’t be possible without you, and every bit of computing power you share is deeply appreciated.

Cheers,
Carlo

Carlo, thanks for your information.

---

Supporting BOINC, a great concept !

Thank you Carlo for the information and welcome to the LHC forum and to the project..

---

Thank you for all the info and looking forward to chipping in with the few computers I have.

Regarding runtime estimates there are some massive differences between different scans and/or hosts.

This Intel Ultra 5 125H, running Win11, shows 1m22s for 7 different scans. The first tasks have been running for over 5 hours (over 4 hours of actual CPU time), probably on the efficiency cores. I had a MilkyWay task hogging 8 cores (possibly the performance ones) but now that's out of the way.

This Intel i7-1165G7, also running Win11, shows both 1m50s and 2h12m47s estimates for tasks from the same scan (#16). 2 tasks have been finished and validated after running more than 4 hours (>3 hour CPU time).

Last but not least, this old Intel i5-6200U, running Linux Mint 22.1, shows either 2m0s or 1h43m12s estimates for different scans. 8 tasks have been finished, with runtimes ranging from just under 2 hours to a little over 3 (CPU times similar, this being the only computer that is not throttled to reduce overheating/fan noise).

Hope this helps a bit.

---

Thanks, Carlo, sounds good.

---

I have a bunch now, all went to 100% quick again and are still going, do we think these will complete in a reasonable time?

I have a bunch now, all went to 100% quick again and are still going, do we think these will complete in a reasonable time?

I think so, got 4 finished WU with run-time 1 h. Carlo said normal runtime 1 to 4 hours, max. 10 (on paper).

---

Cool, mine are sat at 2 hours 30 at the moment, if they go much over 5 hours ill cancel them, 9950x so they should be finishing on the quicker end of the spectrum, I would imagine.

You have 27 valid WUs. I wouldn't cancel the stragglers. Nice machine you have there. :)

---

WU arrives and get crunched successfully here, seems everything is working fine. Seems to last around 2 hours and half/3 hours on my PC. And very very low RAM usage.

Note: ignore the %, it currently is useless.

---

My 9950x had a 2:40.

Longest so far is 7:50 on i9-13700T and shortest is 32 min on the 9950X

Looks like about 3 - 4 hours to completion here.
Would be nice if the remaining estimate could be updated, its at 49 seconds now so the machine downloads hundreds of units, ill have to abort a lot of them before they even start

Looks like about 3 - 4 hours to completion here.
Would be nice if the remaining estimate could be updated, its at 49 seconds now so the machine downloads hundreds of units, ill have to abort a lot of them before they even start

Same here, for completion times; 27 validated.
99.9999% after 25 minutes, then the slow slog until done. Will keep an eye on it but will may end up aborting as well.

as an aside, personally don't care about lost beta time because that part of the deal. if anything, I'd be freaking happy that anything that failed on my comps would help solve a problem!

---

What do you have in the work amount settings (store at least xx days and up to an additional xx days of work) ?

@Richie_unstable

0,1 days and an additional 0,0.

---

@Richie_unstable

0,1 days and an additional 0,0.

I initially cut it to 0,01/0 but then raised it back to 0,05/0 which yields quite a big chunk of tasks on all rigs.

---

Using prefs with home, school or work and
setting Tasks to max. 8 and Cpu to 1.

Had also at beginning of XTrack hundreds of Tasks in Boinc,
when there is no limit setting. Win11pro with Boinc 8.2.4.

Each Wu seems to have a runtime that is 300 times longer, assuming an average runtime of 4 hours. With a working cache of one day, the setting would have to be reduced from 1 day to 0.0033 days. I have no idea whether the BOINC client can calculate with such values, but I assume so.

---