The big credits discussion.

Message boards : BOINC client : The big credits discussion.
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 14351
Netherlands
Message 12986 - Posted: 10 Oct 2007, 20:34:56 UTC
Last modified: 10 Oct 2007, 23:05:31 UTC

The big credit discussion. Moved here from the Cosmology forums, on request of the project administrator there. For he doesn't see the use of having a big heated discussion on his project boards on things he cannot change on other projects anyway and we agree with that. So for everyone who wants to discuss credits on any project, please go ahead. Use this thread.

Challenges on how things can be improved;
which project gives out too much credit, which doesn't give enough credit;
will the amount of credits given decide where people go to to crunch?;
are you willing to crunch for nowt, if a project stood up that wouldn't give out credit for their science?;
Opinions on anything credit related are welcome.

I'm also requesting that people who know of discussions on project forums point to this thread. And vice versa, point to the thread you originated from, by means of quotes and links.


Although we are more relaxed on heated discussions, this thread will be moderated. Not as heavy as when it would take place on any project forums, but please when we ask to calm down on the inflammations, do heed the request.

If you want to discuss in this thread and you just came here for the first time, do know that you need to register separately at these forums.

Go on then.
ID: 12986 · Report as offensive
Nicolas

Send message
Joined: 19 Jan 07
Posts: 1174
Argentina
Message 12987 - Posted: 10 Oct 2007, 21:02:17 UTC - in response to Message 12986.  

Just a little comment from me. I have seen ideas in many places about giving credit for recruiting people, for donating hardware, for helping on the forums, or for having a computer that is more efficient with the specific project app. With all that, I think trying to keep cross-project parity is a hard (if not impossible) task.

But it's also hard to drop cross-project parity, what would we do with all the multi-project stats sites?
ID: 12987 · Report as offensive
JRenkar

Send message
Joined: 10 Oct 07
Posts: 1
Message 12988 - Posted: 10 Oct 2007, 21:18:09 UTC

Thank Ageless for getting this discussion over here :) Need a thread at the origination project page to link here along with Kecks question.

I found it interesting that a lot of people would not join the RND project when they were multiplying the Boinc benchmark credits by 100 because they didn't want to be known as cheaters. I am for the record FOR cross-project parity as much as is humanly possible. More ranting later as the discussion heats up :D~
ID: 12988 · Report as offensive
Nicolas

Send message
Joined: 19 Jan 07
Posts: 1174
Argentina
Message 12989 - Posted: 10 Oct 2007, 21:48:39 UTC - in response to Message 12988.  

I am for the record FOR cross-project parity as much as is humanly possible.

I'm also FOR cross-project parity, but I think it's not very humanly possible.

SETI gives fixed credits. The admins probably used the average credits being given using benchmarks to decide how many credits to give - in order to keep it matching other what projects give. If somebody uses an optimized SETI app, he will be really computing more workunits a day, so it's OK that he gets more credits than with the default app. But in that case, SETI+optimized app would be giving more credits for that user than any other project. How can cross-project parity be kept in that case?

Or, suppose somebody makes actual improvements to the application performance by changing the way it calculates, not just by optimizing it for a specific processor. If it's used "officially" (ie. sent by the project), and credits per workunit aren't changed, SETI would be giving more credits than other projects. If credits are changed to match, they're now giving less credits for the same work.
ID: 12989 · Report as offensive
Nicolas

Send message
Joined: 19 Jan 07
Posts: 1174
Argentina
Message 12990 - Posted: 10 Oct 2007, 22:01:34 UTC - in response to Message 12986.  
Last modified: 10 Oct 2007, 22:03:56 UTC

irc://irc.freenode.net/boinc
xcamel: btw: abc@home seems to be working nicely
PovAddict: yeah giving fake credits too, as far as I know
c0dewarrior: as in more, or less?
xcamel: "fake" ??
PovAddict: admin's words: they are "exponential"
xcamel: meaning?
PovAddict: twice as long workunit gives *more* than twice credits
xcamel: ah..
* c0dewarrior quickly signs up for abc to increase his credit yield
PovAddict: ^^ see, that's exactly what they end up causing
PovAddict: people choosing projects over how many credits they give; *that's* why they should all give the same
c0dewarrior: was just kidding, signing up for stupidity@home instead
xcamel: "all giving same" is a communist tenant.
c0dewarrior: so is "same giving all"
PovAddict: 1. project B starts giving more credits than project A
PovAddict: 2. people join to B just because of credits
PovAddict: 3. A increases credits to "get its users back"
PovAddict: 4. loop repeats until it's a total chaos
PovAddict: doesn't seem so unlikely to happen
xcamel: it ignores human nature... the more an admin wants a project to get done, the more credits it should give
PovAddict: what's to stop an admin from giving a million per CPU minute?
PovAddict: all admins would want it "asap"

EDIT: removed [pre] as it adds stupid extra spacing. And yes I was exaggerating on my "total chaos" apocalyptic prediction.
ID: 12990 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 14351
Netherlands
Message 12992 - Posted: 10 Oct 2007, 22:10:31 UTC - in response to Message 12989.  
Last modified: 10 Oct 2007, 22:14:08 UTC

I think that as long as there is no set way to give out credit, that you will always have that problem.

Now there are still projects that give credit based on the (benchmark*run time)/86400 method, there are those projects that use the Fpops method (but with what multiplier exactly? Is that standardized somewhere? Should it be?) and then there are projects that give credit set by the project on the server. Anything I missed?

The first one is easy to break, just artificially increase your benchmarks and you'll be on your merry way.
The second one is harder to break, but for when the project has their application software as Open Source. Adjust the multiplier and that's that.
While the third one is fairly unbreakable, unless you manage to hack the server. Good luck with that. :-)

I've been through some discussion threads on other forums and always seen people say they get this, that and that at other projects. Some are even keeping scores (can someone lure Tony towards here? ;-)). Edit: one example of that is here.

But what of a real solution? Should all projects use one standardized method, and if so, what would that be? How much credit is enough, considering that the credits were invented to give slow computers the same amount of credit as fast computers. (?? is this correct??)
ID: 12992 · Report as offensive
Nicolas

Send message
Joined: 19 Jan 07
Posts: 1174
Argentina
Message 12993 - Posted: 10 Oct 2007, 23:00:24 UTC - in response to Message 12992.  

considering that the credits were invented to give slow computers the same amount of credit as fast computers.

Same credit per workunit. Credits/day is (and should be) lower for slower computers.
ID: 12993 · Report as offensive
Nicolas

Send message
Joined: 19 Jan 07
Posts: 1174
Argentina
Message 13044 - Posted: 12 Oct 2007, 23:52:21 UTC - in response to Message 12987.  

There's more issues with non-CPU-intensive projects (which I'll call NCI). I think it breaks a bit the concept of the total credit. Users can run multiple nci projects at the same time as normal projects. Users running nci projects get more credits per day than users with the same CPU specs who run only normal projects.
ID: 13044 · Report as offensive
zombie67
Avatar

Send message
Joined: 14 Feb 06
Posts: 129
United States
Message 13046 - Posted: 13 Oct 2007, 0:03:27 UTC - in response to Message 13044.  

There's more issues with non-CPU-intensive projects (which I'll call NCI). I think it breaks a bit the concept of the total credit. Users can run multiple nci projects at the same time as normal projects. Users running nci projects get more credits per day than users with the same CPU specs who run only normal projects.

Good point! I can run up to 10 DepSpid WUs per machine, in addition to the 1 normal WU per thread.
Reno, NV
Team: SETI.USA
ID: 13046 · Report as offensive
SekeRob

Send message
Joined: 25 Aug 06
Posts: 1596
Message 13049 - Posted: 13 Oct 2007, 6:49:03 UTC - in response to Message 12987.  

Just a little comment from me. I have seen ideas in many places about giving credit for recruiting people, for donating hardware, for helping on the forums, or for having a computer that is more efficient with the specific project app. With all that, I think trying to keep cross-project parity is a hard (if not impossible) task.

But it's also hard to drop cross-project parity, what would we do with all the multi-project stats sites?

My thinking is that endurance bonuses for finishing long jobs or other honorary tokens like for the quoted should remain separate from the BOINC scoring. Let a project do the praise for that at the project.

What I'd like to see is a resource weighing, not just the CPU.... e.g. the volume of disk I/O 'outside' of VM, the peak RAM/VM use. Drop processing time entirely for the purpose of scoring and measure according a 1 Tera-Ops, not flops or iops = 100,000 cobblestones, after the fact.

Coelum Non Animum Mutant, Qui Trans Mare Currunt
ID: 13049 · Report as offensive
W-K ID 666

Send message
Joined: 30 Dec 05
Posts: 314
United Kingdom
Message 13092 - Posted: 16 Oct 2007, 3:22:53 UTC

Taking the opposite tack, because I like throwing the spanner in the works.

Should projects that have open source applications lower there credits/time when their application is made more efficient, by the efforts of volunteer optimisers?
ID: 13092 · Report as offensive

Message boards : BOINC client : The big credits discussion.

Copyright © 2020 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.