FormulaBOINC

[Ken g6]

I've checked FB Sprint page and noticed that mw@h and Skynet POGS weren't crossed yet from the list. Is there any possibility of running a project twice in the sprint?

No, they're just slow to update the site. They don't usually post medals immediately either.

 

[[H]Coleslaw]

Looks like QCN's recent stats blunder has revoked all of the FB points earned this year. Hopefully Sebastien does something to restore the standings. QCN's admins have not chimed in yet...

 

[crashtech]

QCN stats page on Formula BOINC gives a stat of negative 5 million points, rather depressing!

 

[GLeeM]

I'm drawing a blank - this race used to end before December, when did it start going all year?

 

[StefanR5R]

The 2016 rules drew the line on November 30.
http://formula-boinc.org/regles.py?year=2016

 

[ao_ika_red]

No, they're just slow to update the site. They don't usually post medals immediately either.

Thanks, I guess because it's operated by volunteers, it depends on whether they have time to update things thoroughly.

 

[Pokey]

I have participated in Formula BOINC in the past but have not kept up with your comings and goings this year. I have hit a milestone in S@H as well as F@H and can pitch in now until the F@H challenge starts, sooooo………… if someone would point me at a project(s)……………….. I’ll formulate with you guys for a while.
Just point me where I can help, I am GPU heavy and light on cpu’s if that helps.

 

[crashtech]

Hi @Pokey ! I heard we needed help in Collatz Conjecture, but I don't know if that's still the case.

Last I heard, the invite code was spammerssuck, and you have to set up an account on their site before attaching the project.

 

[TennesseeTony]

PM sent Pokey.

Good news, Milkyway and POGS got scratched off the list of Sprints today, so hopefully that means tomorrow's announcement will go smooth, unlike that one race where that they forgot to update the site.

 

[Orange Kid]

.............and Yafu !!

 

[TennesseeTony]

Great. Someone please tell me there is a practical, useful application of the mathematical knowledge gained from this project? I'd much rather be curing some disease....

 

[crashtech]

It says that it tests the latest BOINC code for bugs, that can't be bad, can it?

I want to admit that although I do like to give generously to disease cures, etc., I have a love for pure science as well, especially stuff to do with space. So for me, it's almost all good...

 

[ao_ika_red]

Any reason why does yafu need invitation code to acquire new account?

 

[StefanR5R]

Any reason why does yafu need invitation code to acquire new account?

Message 780 - Posted: 17 Aug 2015, 15:51:58 UTC

New accounts can only be created with an invitation code now. The invitation code is stated on the project start page. This is introduced because of so many spam accounts which were created in the past.
Additional I removed all accounts which have no credits and no hosts assigned.

-yoyo-

The start page says the code is yafu.

 

[StefanR5R]

The server maintains a limit of 2 tasks in progress.
Runtimes are variable from seconds to hours AFAIR.
The progress percentage indicator is wrong and should be ignored.
Tasks can be multi-threaded but have single-threaded portions on them. E.g. the YAFU-4t application allocates 4 logical CPUs and uses them most of the time, but sometimes only 1 CPU, leaving 3 idle.
YAFU-16t can take more than 16 CPUs.

Edit: Quorum is 1.
Credits are given per standard BOINC "creditNew" scheme. With quorum of 1, this means that credits will be all over the place for some time after a host was added to YAFU.

 

[ao_ika_red]

Thanks for the brief explanation, Stefan.
I just don't understand why small project like Yafu could be attacked by so much spam and traumatized the admins. Seti has more popularity but there's no additional security measure to register new account there.

 

[Kiska]

Credits are given per standard BOINC "creditNew" scheme.

Did you mean CreditScrew?

 

[StefanR5R]

I mean the scheme where they added lots of knobs and dials and non-deterministic components which project admins no longer fully understand, giving malicious users an edge over the admins. /s

 

[Ken g6]

The first update has OCUK #1, us #2, and all other teams - in all other leagues - behind us!

 

[Orange Kid]

A look at the marathon stats.
In first with a good lead, but we need to keep it that way
Second in the sprint at the moment.


Yafu..........................Marathon

1 TeAm AnandTech ..................25 ...........4,902,480
2 Crunching@EVGA .................18 ...........2,619,137
3 meisterkuehler.de Team ........15 ...........2,137,259
4 OcUK - Overclockers UK ........12 ..........1,244,252
5 Overclock.net ..........................10 ..........1,208,199
6 Ars Technica ............................8 ...........1,007,561
7 The Scottish Boinc Team .........6 ..............602,777
8 AMD Users ...............................4 ..............347,691
9 Portugal@Home .......................2 ..............246,805
10 SwissTeam.NET .....................1 ................78,623

 

[StefanR5R]

Half of my currently running tasks have now ≈20 hours elapsed time, and of course there is no way to know how much longer they will go on.

Run times according to YAFU server status:

                                     Runtime of last 100
                                     tasks in hours:        Users in
Application   Unsent   In progress   average, min, max      last 24 hours
-------------------------------------------------------------------------
YAFU             5        1104       1.59  (0    -  44.84)       94
YAFU f.s.c.      3          88       0.4   (0.01 -  12.44)       44
YAFU-4t         99         613       3.75  (0.01 -  77.38)      105
YAFU-8t         27         484       2.56  (0.01 -  69.97)       82
YAFU-16t       244         204       4.43  (0.01 - 139.84)       46

Of course the "average" run time shown there is utterly deceiving. After all, our hosts are occupied mostly by the long running tasks, not by the short ones. More interesting would be a weighted average, with run time being the weight.

Oh, and if there was a 139.84 h = 5.8 days long task among the last 100 YAFU-16t tasks, then I rather don't want to know what the peak run time among the last 10,000 tasks was.

 

[StefanR5R]

The Sprint Stats which we will read on Sunday will give only a rough picture of the actual computing time expended by each team.

Here is a forum thread about how well CreditNew works with YAFU's multithreaded tasks. (Hint: Not too well actually.)
https://yafu.myfirewall.org/yafu/forum_thread.php?id=147
The thread is from 2014 but I suspect it is still applicable today.

Summary: CreditNew requires many days to self-adjust credits/seconds for tasks with a medium thread count. The time needed to converge is longer than this FB Sprint is running. Highly threaded tasks however do not appear to converge to a reasonably stable credits/seconds even after many days. (@[H]Coleslaw, correct me if I got this wrong.)

However, this is one of the aspects of Formula Boinc. It treats the occasional alpha project the same as any much more reliable project.

------------
Edit:
Heh, I just saw this (after 11:00 UTC):

Pos.          Team            Points   Total Credits
  1   OcUK - Overclockers UK    25        404,358
  2   TeAm AnandTech            18        404,313

The random number generator on YAFU's server which computes credits apparently ran out of entropy.

------------
Edit 2:
rechenkraft.net got >500,000 points yesterday. They came almost entirely from a single user, from a single host, from a single task!
https://yafu.myfirewall.org/yafu/result.php?resultid=1168229

     Sent   8 Sep 2017, 1:52:24 UTC
 Received   14 Sep 2017, 17:03:54 UTC
 Run time   5 days 19 hours 50 min 19 sec
 CPU time   1 days 3 hours 52 min 46 sec
   Credit   494,440.00

Check out the ratio of run time versus CPU time. It's a 16T task on a 8C/16T host.

 

[ao_ika_red]

It's certainly a very bad decision to do experiment on yafu with my old and dusty C2D notebook. It run a single task since early thursday but there's no sign of completion yet. I hope getting more than 100k credit for this one.

 

[StefanR5R]

In boincmgr's Tasks tab, select the running yafu task, then click Properties.
The properties window should among else show something like "Directory slots/0" (or slots/1 or whatever).
Then open the directory C:\ProgramData\BOINC\slots\0 (or slots\1 etc.) and you will see all the scratch files of the job.
Most interesting of them is "factor.log" AFAIK. It should show which of the different stages of a YAFU task the job has completed:

Message 1035 - Posted: 21 Jul 2017, 6:20:12 UTC
[...]
Yafu runs in different stages to factor the composite.

One of the first stage is to try to find a factor with ecm. There one ecm process is forked per core, you will see many ecm processes.

If this doesn't factorize the composite it has to go the long way and do a complete factorization with gnfs, you will see many gnfs proecesses.
Afterwards it checks if it has enough relations found. If not yafu proceeds with gnfs. You will also see a big growing nfs.dat file in the slot folder.

At the end yafu consolidates the gnfs results and does a linear algebra. Unfortunately this stage is mostly single threaded.

yoyo

This reply was to somebody who used lots of cores per task. With your Core2 Duo, its at most 2 ECM processes, and then 2 GNFS processes, unless you modify cc_config.options.ncpus.

 

[Orange Kid]

In second and gaining at the moment, but with the unpredictable work times on this project who knows. Could become quite the see-saw battle.
Only 7 teams with points in this race, I'm guessing it's not a favorite project for most.



Yafu..........................Sprint.............Day One
1 OcUK - Overclockers UK ..........25 .............684,215
2 TeAm AnandTech ...................18 .............655,914
3 Crunching@EVGA ....................15 ............382,135
4 Overclock.net ............................12 ............151,280
5 meisterkuehler.de Team ...........10 .............110,658
6 Portugal@Home ........................8 .................6,403
7 AMD Users ................................6 .....................78


Yafu..........................Marathon
1 TeAm AnandTech .................25 .............5,375,945
2 Crunching@EVGA .................18 ..............2,955,793
3 meisterkuehler.de Team .........15 .............2,235,528
4 OcUK - Overclockers UK ........12 .............1,724,048
5 Overclock.net ..........................10 ..............1,347,390
6 Ars Technica ............................8 ...............1,007,561
7 The Scottish Boinc Team .........6 ..................602,777
8 AMD Users ..............................4 ...................347,769
9 Portugal@Home ......................2 ...................251,964
10 SwissTeam.NET ....................1 .....................78,623