LHC@home

Well I have finally got some work on my Mac with our new MacOS executable,
built on OS X 10.10.5 Yosemite .
Please report to me eric.mcintosh@cern.ch or to this thread if you get some worki
and there are problems. Eric.

---

The executable works with OS X 10.9.5 Mavericks and a Core 2 Duo CPU:
https://lhcathome.cern.ch/lhcathome/result.php?resultid=113186656

Thanks for your effort.

Thanks for the feedback. Good news for me. I am sorry it took so
long to get this executable installed. Eric.

---

I'm getting them on my iMac with 10.8.5, though the validation is only about 50/50 at this point:

https://lhcathome.cern.ch/lhcathome/results.php?hostid=10143259

---

Click here to see My Detailed BOINC Stats

All 3 of these inconclusives are paired with wingmen using the SixTrack v453.00
x86_64-apple-darwin app: Validation inconclusive tasks for computer 9631414. Same computer currently has 14 valids with a variety of computer types including Windows, Linux, and some Apple machines.

Hmm, interesting: all 8 tasks (from 20 Jan 2017, 0:27:54 UTC) have the same computer as wingmen, host 10372338. A windows machine with SixTrack v451.07 (pni)windows_intelx86 and client 7.6.22

And the 3 computed task are all inconclusives.

https://lhcathome.cern.ch/lhcathome/results.php?hostid=10262927

I have tried it on MAC "El Capitan" and "Sierra"
Allmost 50% of WU's are "Completed, validation inconclusive"
That is a lot more than usually, compared to a Windows Machine.

MEA CULPA; the MacOS version is incompatible with the Linux/Windows.
Version 4530 has a minor bug fix in the RIPPLE code but it gives different
results. Hope to consolidate in a week or so with new updated versions
everywhere. The current 4530 MacOS is withdrawn. Eric.

---

The current 4530 MacOS is withdrawn.

I think I have recent inconclusive results where my wingman's task was sent out after this post (e.g., Workunit 56122830).

Sorry the MacOS executable is withdrawn temporarily.
Hope to have it back in 1 to 2 weeks. Eric.

---

Sorry the MacOS executable is withdrawn temporarily.

There are hosts out there still returning results using the v453 apple-darwin app (e.g., 10417292, 10415063, & 10411359). I know it might be a lesser problem right now, but are in progress tasks for that app not going to be canceled?

FYI: The macOS results (from SixTrack version 4.5.30) are physically valid since none of the currently running inputs use the removed RIPP feature, however they may differ slightly from the Linux and Windows version (from SixTrack 4.5.17) as in version 4.5.25 we fixed a small bug in how the input files were read. This means that when it compares results from MacOS with Windows or Linux, it fails... We're really sorry for this mess.

The good news is that version 4.6.x is in the final stages of testing, introducing lots of new features for the researchers running the simulations, and versions for Linux (64-bit, we can probably do 32-bit as well if there is demand for it), Mac OS, Windows (64- and 32-bit), FreeBSD and possibly NetBSD. AVX is also supported, and we've also acquired an ARM server to test on...

Further we think we've figured out why it is sometimes driving the I/O crazy (overzealous flushing of output files) - fixing this should speed things up quite a bit. So there may be another release not so long after the next one.

Now why it hasn't stopped sending SixTrack jobs to Mac hosts I unfortunately cannot answer; my work is mainly on the SixTrack application programming side, not on the job management code. However I know it is being looked into.

It is possible to follow the development at our GitHub page: https://github.com/SixTrack/SixTrack
Please let us know if you see something strange :) There is also a Google summer of code internship coming up...

Thanks a lot for your computing time and patience, it is really appreciated!

-- Kyrre Sjobak, SixTrack developer @ CERN

I should also add that if anybody has any requests for other platforms in addition to the above list (other than android/GPUs) please make yourself known and I'll see what I can do.

kyrsjo wrote:

FYI: The macOS results (from SixTrack version 4.5.30) are physically valid since none of the currently running inputs use the removed RIPP feature, however they may differ slightly from the Linux and Windows version (from SixTrack 4.5.17) as in version 4.5.25 we fixed a small bug in how the input files were read. This means that when it compares results from MacOS with Windows or Linux, it fails... We're really sorry for this mess...

Thanks for the feedback, Kyrre! If MacOS results are actually valid from a scientific perspective, what's the plan to deal with the "Validation inconclusive" results?

The problem is when BOINC tries to validate results from SixTrack 4.5.30 (Mac OS) with results from 4.5.17 (everything else); they are not always *exactly* the same due to numerical roundoff, so BOINC will mark them as invalid.

I don't know what can be done about those WUs... As I said, they might have passed if compared to another mac result. How does this affect your scores (I'm new to BOINC...)?

From the scientific perspective, it is very good that if they DO pass the validation test, the physics is OK, so we don't need to go hunt down every WU that was done and validated on OS X and then resubmit these.

As far as I know, the main thing remaining before we can release 4.6 is to fix the testing code on Windows, so that we can validate that our Windows executable produces the correct results. Then we need to run the tests, and hopefully that's it...

Again, thanks a lot to those of you who pointed this out!

Great to see some SixTrack development :)

I know people asked for GPU in the past, I'm not sure how much sense, effort & benefit from your perspective.

I felt the addition of AVX wasn't too much effort and should give good boost in speed for enabled systems.

kyrsjo wrote:

I don't know what can be done about those WUs... As I said, they might have passed if compared to another mac result. How does this affect your scores (I'm new to BOINC...)?

How it affects volunteers scores/credit depends on which tasks sent out for a workunit get validated. For me, I suspect that a third task returned by a Windows or Linux host will validate with me and I'll get credit; if that third task is from a Mac host, then it will likely validate with the other Mac host. In the latter case, my task will get marked as invalid and get reduced or no credit**. I'm not sure which because I don't remember ever having an invalid task here. Does that make sense?

** Of course, it's the science that matters... ;-)

I suggest someone look carefully at All tasks for computer 10357822 before deciding how to resolve the validation vs. scientifically correct results issue. This computer is currently showing that is has returned over 25,000 Sixtrack results and approximately 24,500 of them are inconclusive. I believe the computer is most likely malfunctioning in some way and returning unreliable results - as evidenced by its 345 invalids and 5 errors. But, surprisingly, it also has returned 75 valids. Therefore, I suspect a substantial number of its inconclusives are truly invalid and not just "rounding errors".

Thanks Stick; this is indeed our MacOS executable problem.
See other messages from Kyrre. Eric.

---

Why not set 'Homogeneous redundancy' for sixtrack, until the new version is completed?
https://boinc.berkeley.edu/trac/wiki/HomogeneousRedundancy
You need to wait until there are no jobs in progress to do this :-(