Advanced search

Message boards : News : New Student and QMML Project

Author Message
Dominik
Send message
Joined: 15 Dec 17
Posts: 9
Credit: 0
RAC: 0
Level

Scientific publications
wat
Message 48605 - Posted: 4 Jan 2018 | 13:44:42 UTC

Hello everyone and happy new year!

My name is Dominik and I am a masters student from Germany. I studied biochemistry in the city of Bayreuth, but I always focused my work more towards biophysics and bioinformatics. I'm in the group since October, where we have since then started to initiate a quantum chemical machine learning project.
We soon came to the conclusion that we need a quantum chemical database in order to drive research forward in this field.

This is where we need your help!

Some of you have already seen that I released a small amount of 50.000 WUs yesterday evening and there is even more. The WUs contain a certain amount of small molecules for which quantum chemical simulations will be performed. Depending on the size of the molecule the simulation (and the WU) will take longer. We did initial benchmarks to estimate how to design the WUs, but everything turned out a bit different with the very long WUs I prepared for Toni which you were fighting with over the last month. Thanks to everyone who was helping and sticking through this! I will try to adjust the WUs as best as I can, but for now the current DOMINIK WUs that are already on the server are fixed.
Regarding the windows based QM tasks I cannot give a clear answer about this yet, but it's definitely one of our goals.

It is a long term goal of ours to keep running the quantum chemistry tasks on GPUGRID and I try my best to ensure that there are always tasks available.

So thanks again to everyone who is participating and helping out. I really appreciate it very much.

Have a great start into 2018!

[CSF] Thomas H.V. DUPONT
Send message
Joined: 20 Jul 14
Posts: 732
Credit: 115,270,366
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 48614 - Posted: 4 Jan 2018 | 16:20:38 UTC

Welcome on board, Dominik, and Die besten Wünsche für dieses neue Jahr.
____________
[CSF] Thomas H.V. Dupont
Founder of the team CRUNCHERS SANS FRONTIERES 2.0
www.crunchersansfrontieres

Profile Logan Carr
Send message
Joined: 12 Aug 15
Posts: 240
Credit: 64,069,811
RAC: 0
Level
Thr
Scientific publications
watwatwatwat
Message 48620 - Posted: 4 Jan 2018 | 20:01:11 UTC

Thank you for the heads up and info. Always good to get an update.

My name is Logan. Pleasure to meet you!
____________
Cruncher/Learner in progress.

3de64piB5uZAS6SUNt1GFDU9d...
Avatar
Send message
Joined: 20 Apr 15
Posts: 285
Credit: 1,102,216,607
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwat
Message 48621 - Posted: 4 Jan 2018 | 20:43:59 UTC

Welcome... herzliche Grüße auch aus Österreich! :)
____________
I would love to see HCF1 protein folding and interaction simulations to help my little boy... someday.

kain
Send message
Joined: 3 Sep 14
Posts: 152
Credit: 705,632,245
RAC: 1,970,066
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 48622 - Posted: 4 Jan 2018 | 22:57:56 UTC

That's a very, very good news :)
Best wishes from Poland :)

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1332
Credit: 7,100,167,459
RAC: 15,030,489
Level
Tyr
Scientific publications
watwatwatwatwat
Message 48625 - Posted: 5 Jan 2018 | 3:14:39 UTC

Hi Dominik. Very surprised to see the very large number of tasks you made available. I am very new to the project and my first couple of weeks, I wondered if ANY work was going to be made available. See that has changed.

Now on to my question. First one of your comments in your introduction post has me intrigued.


Depending on the size of the molecule the simulation (and the WU) will take longer. We did initial benchmarks to estimate how to design the WUs, but everything turned out a bit different with the very long WUs I prepared for Toni which you were fighting with over the last month.


Can you say what the distribution is in the 50,000 tasks you posted is? Are they all the same for expected compute load? Are some of them very simple calculations with quick completion times expected? Are some of them very long calculations?

I am trying to figure out why my first ten or so tasks got 150-160 credits for 3500 seconds of compute time and the next batch got 20 credits for the same amount of compute time. The batch today got 6 or 7 credits for 2000 seconds of compute time. The credit awarded does not scale with compute time.

Can you explain this discrepancy?

Jim1348
Send message
Joined: 28 Jul 12
Posts: 819
Credit: 1,591,285,971
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 48626 - Posted: 5 Jan 2018 | 5:20:34 UTC - in response to Message 48625.

I am trying to figure out why my first ten or so tasks got 150-160 credits for 3500 seconds of compute time and the next batch got 20 credits for the same amount of compute time. The batch today got 6 or 7 credits for 2000 seconds of compute time. The credit awarded does not scale with compute time.

Can you explain this discrepancy?

I will jump in and just offer the observation that a lot of people on a lot of projects ask the same question. It is related to how BOINC calculates credits, which I don't claim to understand. But the results are almost random, and there is probably nothing that GPUGrid can do about it.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1332
Credit: 7,100,167,459
RAC: 15,030,489
Level
Tyr
Scientific publications
watwatwatwatwat
Message 48627 - Posted: 5 Jan 2018 | 6:17:46 UTC - in response to Message 48626.

Well I am quite familiar with what we "setizens" refer to as "CreditScrew" but I was under the impression that GPUGrid.net was one of the projects that didn't utilize the BOINC CreditNew algorithm like Einstein@Home for instance.

Toni
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 9 Dec 08
Posts: 1006
Credit: 5,068,599
RAC: 0
Level
Ser
Scientific publications
watwatwatwat
Message 48630 - Posted: 5 Jan 2018 | 9:27:49 UTC - in response to Message 48627.

Hi all, and of course welcome to Dominik!

Credits for CPU tasks follow the "old" default algorithm for Boinc. I don't fully understand it. Credits algorithm have generated endless controversy to the point that honestly I don't think a solution exists. So we are just keeping the default.

Dominik
Send message
Joined: 15 Dec 17
Posts: 9
Credit: 0
RAC: 0
Level

Scientific publications
wat
Message 48631 - Posted: 5 Jan 2018 | 9:38:00 UTC
Last modified: 5 Jan 2018 | 9:40:36 UTC

Hi Dominik. Very surprised to see the very large number of tasks you made available. I am very new to the project and my first couple of weeks, I wondered if ANY work was going to be made available. See that has changed.


We wanted to set everything up as smooth as possible because we knew the workload that was coming. So as Toni always said those WUs were experimental, wherefore we didn't want to release too much and instead try to adapt to the problems in the first place.


Now to your question:

The kind of quantum chemical simulation that is performed is always the same. The only thing that differs is the molecule species that is put in. The bigger the molecule, the longer the simulation takes.
There are currently 150 different molecules with ascending size distributed in the 50.000 WUs, so I expect the last ones to take longer than the first ones i.e.
The actual time is obviously heavily dependent on your system and threads, for the current ones the computation time seems to be on average about 25min. For some later ones in the distribution this time will scale up.
Regarding the credit issue you and maybe others are experiencing, I am sorry that this is happening to you, but there is currently nothing I can do from my side and as Jim said I'm not sure if this is something we can target at all.
Thanks for reporting this issue and for coming back to give it another try!

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1332
Credit: 7,100,167,459
RAC: 15,030,489
Level
Tyr
Scientific publications
watwatwatwatwat
Message 48640 - Posted: 5 Jan 2018 | 21:13:00 UTC - in response to Message 48630.

Hi Toni and Dominik. Thanks for the info about the project using the "old" credit calculation mechanism. That is a GOOD THING!

FYI, you can read about CreditNew here

There has been much discussion about the failures of CreditNew to account for newer and faster hardware because it is tied to antiquated application APR's.

David Anderson has expressed no interest whatsoever in changing the code for the algortithm. It is what it is. Just accept it.

A majority of SETI participants have lobbied to return to the "old" Credit mechanism to no avail.

Thanks for the explanation about the nature of the tasks Dominik. My batch last night had one that reverted to the middle ground of credit I have seen so far so that must have been a bigger molecule.

Is still will get work here as necessary for when SETI@home is having issues.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1332
Credit: 7,100,167,459
RAC: 15,030,489
Level
Tyr
Scientific publications
watwatwatwatwat
Message 48641 - Posted: 5 Jan 2018 | 21:13:01 UTC - in response to Message 48630.
Last modified: 5 Jan 2018 | 21:44:03 UTC

Hi Toni and Dominik. Thanks for the info about the project using the "old" credit calculation mechanism. That is a GOOD THING!

FYI, you can read about CreditNew here

There has been much discussion about the failures of CreditNew to account for newer and faster hardware because it is tied to antiquated application APR's.

David Anderson has expressed no interest whatsoever in changing the code for the algorithm. It is what it is. Just accept it.

A majority of SETI participants have lobbied to return to the "old" Credit mechanism to no avail.

Thanks for the explanation about the nature of the tasks Dominik. My batch last night had one that reverted to the middle ground of credit I have seen so far so that must have been a bigger molecule.

I still will get work here as necessary for when SETI@home is having issues.

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1617
Credit: 8,119,044,351
RAC: 17,916,004
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 48644 - Posted: 5 Jan 2018 | 23:08:34 UTC - in response to Message 48641.

Be careful of your terms. 'CreditNew' has been the default BOINC mechanism since 2010. I suspect this is what GPUGrid is using for these tasks: the support mechanisms for 'even older credit' have been removed from the codebase.

David has expressed an interest in exploring credit mechanisms: I am the only person (so far) to have responded to Issue #2132 in over 3 months.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1332
Credit: 7,100,167,459
RAC: 15,030,489
Level
Tyr
Scientific publications
watwatwatwatwat
Message 48645 - Posted: 6 Jan 2018 | 2:16:38 UTC - in response to Message 48644.

Hi Richard, thanks for the input. I just joined so don't know the history of GPUGrid.net.

I was just going on the post by Toni and Dominik saying they were using the "old" algorithm.

Wasn't aware the code for the earlier credit algorithms had been removed from the BOINC source.

I have never seen any post on the project forums from DA saying he was interested in revisiting the code for credit. I just recently started lurking on the developer site.

mmonnin
Send message
Joined: 2 Jul 16
Posts: 337
Credit: 7,211,226,065
RAC: 14,043,787
Level
Tyr
Scientific publications
watwatwatwatwat
Message 48648 - Posted: 6 Jan 2018 | 14:24:43 UTC

Credit for the CPU tasks here is even pretty low for creditnew.

Post to thread

Message boards : News : New Student and QMML Project

//