Welcome to MilkyWay@home

I've had enough !!!

Message boards : Number crunching : I've had enough !!!
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5

AuthorMessage
AnRM

Send message
Joined: 6 Mar 08
Posts: 15
Credit: 3,006,602
RAC: 0
Message 4840 - Posted: 20 Aug 2008, 6:06:01 UTC

Thanks for taking the 'bull' by the horns, Travis. We take exception to statements saying 'most' people support the Cruncher. We certainly don't and appreciate your remedial action. We note with great interest that the most extreme negative comments about this are from members of C's team....funny that!
You have shown great patience answering some of these provocative statements and you have our respect for that. Looking forward to the next phase....Rog.
ID: 4840 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Palo M.
Avatar

Send message
Joined: 13 Jul 08
Posts: 8
Credit: 133,655,357
RAC: 0
Message 4841 - Posted: 20 Aug 2008, 7:40:00 UTC

Thank you for explaining the background, Travis.
Some my notes to whole topic (not addressed to specific person):

  • Although I'm technically oriented, in BOINC world I've learned that distributed computing is a lot about psychology. So Travis may have preference to do a science work for MilkyWay, but somebody from project should think about the cheating, as in the crowd there will always be somebody thinking about cheating. And if there will be real cheating, other persons will discover it and the flame war will start. And there could be false accusation or at least controversy... None of this is good for project (from psychologic point of view).
  • The optimization may be done for good motives. But optimizer may make unintentional mistake and the results will not be correct. Also, there could be some problem related to hardware - often there are unstable oveclocked machines used, which makes results incorrect. This also supports the opinion that quorum should be 2 instead of 1 and validator should eliminate incorrect results.
  • IMHO the whole credit system is wrong concept (or unfair, depending on point of view). Look at the definition of cobblestone - it's about amount of time spent on reference computer, without taking into consideration how the time was actually spent. If the employer would pay me for time spent in office without any other conditions, without asking whether I do something meaningful or useless, I would sleep in the office everyday.
  • Now the location of the source code is known. But now I do not want to do my own compilation, as at the moment it seems it would not help the project (see my 2nd remark above). I'll wait until project is ready for that... I hope the day will come soon.

ID: 4841 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Conan
Avatar

Send message
Joined: 2 Jan 08
Posts: 122
Credit: 69,476,030
RAC: 2,852
Message 4843 - Posted: 20 Aug 2008, 11:06:22 UTC - in response to Message 4840.  
Last modified: 20 Aug 2008, 11:07:12 UTC

Thanks for taking the 'bull' by the horns, Travis. We take exception to statements saying 'most' people support the Cruncher. We certainly don't and appreciate your remedial action. We note with great interest that the most extreme negative comments about this are from members of C's team....funny that!
You have shown great patience answering some of these provocative statements and you have our respect for that. Looking forward to the next phase....Rog.


@ AnRM,
From your posting it appears that you have adopted the "Royal 'We'" (as used by the Queen of England/Britian "We are not amused"), and are speaking for all.

"We" should be replaced by "I", as in I take exception, I certainly don't and I note, this then shows the comment came from you and not others.

Support from a person's team is what being in a team is all about.

@ all MilkyWay voluteers and project members,

I have been sitting on the fence watching all three sides of this issue;

Yes Poorboy has a valid point about how certain computers can get far and above the normal amount of points for a given amount of work, this had to be looked into and poorboy thanked for making the project aware of it.

Calling someone a cheat without being able to actually prove it and just going on the apparent results does not raise the esteem or respect one is held in.

Crunch3r should be thanked for the work he has put into improving applications, for this and other projects, he is one very talented person.

If the work done for this project is producing results that are far and above what should of been expected then perhaps the optimiser should of looked into the code more to make sure it is valid or at least the project should of given the ok that all was well before that optimised app was used.
From the optimiser's comments this was done but the project does not appear to have had the time yet to fully test out the new optimisations and so no one knows if the new code is valid or not.
The validator thinks so though.

I do wonder how a computer (no matter how overclocked), can take 4 minutes to do a WU that I take 5 1/2 hours to do, that is extreme optisation I must say.

Travis has done what he believes is in the best interest of the project without upsetting too many people and hopefully keep ALL the current volunteers, poorboy and Crunch3r included,as all are needed to help this project.

This is some of my dollars worth of comments (I'll save some change for other topics), about this unfortunate fall down in camaraderie between volunteers.
Hope all will settle down and we keep this project humming along.

Thanks, and keep smiling as it makes others wonder what you have been up to.
ID: 4843 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Conan
Avatar

Send message
Joined: 2 Jan 08
Posts: 122
Credit: 69,476,030
RAC: 2,852
Message 4844 - Posted: 20 Aug 2008, 11:20:52 UTC
Last modified: 20 Aug 2008, 11:21:49 UTC

I forgot to add that I hope both Crunch3r and poorboy keep participating at this project, don't let the current optimisation issue cloud your vision as to why you both came to this project in the first place.

The project is interesting and has scientific merit, it needs our support.

And you can have a good or a bad day, whatever you wish.
ID: 4844 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Palo M.
Avatar

Send message
Joined: 13 Jul 08
Posts: 8
Credit: 133,655,357
RAC: 0
Message 4845 - Posted: 20 Aug 2008, 12:25:01 UTC - in response to Message 4843.  

Conan wrote:
I do wonder how a computer (no matter how overclocked), can take 4 minutes to do a WU that I take 5 1/2 hours to do, that is extreme optisation I must say.

I do not wonder. It's the "magic" of optimization of code to speed. Try some and you will see yourself. If you do not want to try, just think about very simple case: different speed of different sorting algorithms.
ID: 4845 · Rating: 0 · rate: Rate + / Rate - Report as offensive
AnRM

Send message
Joined: 6 Mar 08
Posts: 15
Credit: 3,006,602
RAC: 0
Message 4846 - Posted: 20 Aug 2008, 12:26:05 UTC - in response to Message 4843.  
Last modified: 20 Aug 2008, 13:12:46 UTC

@Conan
[quote]From your posting it appears that you have adopted the "Royal 'We'" (as used by the Queen of England/Britian "We are not amused"), and are speaking for all.

"We" should be replaced by "I", as in I take exception, I certainly don't and I note, this then shows the comment came from you and not others.[quote]

LOL...Hardly the 'Royal We'. Surely you give us too much credit! (Maybe not a good phrase to use under the circumstances?) As our account name hopefully suggests, two people are involved, no more, no less....:)....Cheers, Rog.(Maybe I should add Annelies' name too??)
ID: 4846 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Emanuel

Send message
Joined: 18 Nov 07
Posts: 280
Credit: 2,442,757
RAC: 0
Message 4847 - Posted: 20 Aug 2008, 13:12:12 UTC - in response to Message 4845.  

I do not wonder. It's the "magic" of optimization of code to speed. Try some and you will see yourself. If you do not want to try, just think about very simple case: different speed of different sorting algorithms.


There is no magic in computer sciences. The greater the difference in speed between two applications getting the same result, the more radical the differences between those applications generally are. This might make it a lot more difficult to check, and Travis has a point about holding off on the optimizations until the new code is out. However, I do hope they can work together in the future and help us all contribute as much as possible to science.
ID: 4847 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Conan
Avatar

Send message
Joined: 2 Jan 08
Posts: 122
Credit: 69,476,030
RAC: 2,852
Message 4848 - Posted: 20 Aug 2008, 13:14:55 UTC

@ Palo M.

I would love to be able to give programming and optimisation a go, but as I gave up on a 'c' programming course as it did not make sense to me and I have yet to work out how to compile a Linux kernal to make it work better or even upgrade an old Linux version to a new one, I tend to give credit where credit is due.
In this case I gave credit to Crunch3r on his ability to programme (and apparently you as well it seems}.

@ AnRM,

No problem, all clear now, sorry for making an assumption without all the information (bit like what has happened in this thread).

Cheers to all.

ID: 4848 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Palo M.
Avatar

Send message
Joined: 13 Jul 08
Posts: 8
Credit: 133,655,357
RAC: 0
Message 4849 - Posted: 20 Aug 2008, 15:03:28 UTC - in response to Message 4847.  

Emanuel wrote:
I do not wonder. It's the "magic" of optimization of code to speed. Try some and you will see yourself. If you do not want to try, just think about very simple case: different speed of different sorting algorithms.


There is no magic in computer sciences. The greater the difference in speed between two applications getting the same result, the more radical the differences between those applications generally are. This might make it a lot more difficult to check, and Travis has a point about holding off on the optimizations until the new code is out. However, I do hope they can work together in the future and help us all contribute as much as possible to science.

I quoted the word magic. It shouldn't be taken literally.
And, I agree with you that optimized code could be quite different from original one.
But, there should be a way to check correctness of the code easily. NOT ONLY because of possible optimization, but mainly because of further evolution. After each update of your code you should re-run the tests to make sure that you accidentally did not spoil something.
I made some optimized applications for a few other BOINC projects (although mostly just re-compilation without any source code change, to produce faster binary for my processor)... The first thing I always did were the tests to make myself confident about correctness of the build. And I still closely tracked the first few results reported to project.
Of course, at the beginning of software development it's important to produce something which really runs. But then, after the first version, there should come further steps. Tests are important in software development generally, in BOINC even more. If there are no tests available to ensure correctness of application (which seems to be the case for MilkyWay at the moment), then I can say that the software is still in the pilot phase and there is still long way to go...
ID: 4849 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 4851 - Posted: 20 Aug 2008, 15:48:11 UTC - in response to Message 4845.  
Last modified: 20 Aug 2008, 15:49:03 UTC

Conan wrote:
I do wonder how a computer (no matter how overclocked), can take 4 minutes to do a WU that I take 5 1/2 hours to do, that is extreme optisation I must say.

I do not wonder. It's the "magic" of optimization of code to speed. Try some and you will see yourself. If you do not want to try, just think about very simple case: different speed of different sorting algorithms.


Given what our code is doing, honestly the only way i can see this kind of speed improvement being possible is through vectorizing most of the floating point operations in all the loops of our code. There's really no way to increase the algorithmic runtime of what we're doing (it's pretty simple -- just a sum of a lot of floating point ops).

anyways, given what was happening i basically had two things that could be done:

1. reduce the granted credit for every WU by a factor of 20-50 (which i'm sure wouldn't make anyone happy).

2. reduce the max possible granted credit over a period of time.

I went with 2 for obvious reasons. At any rate, the optimized code will still give an improved amount of credit, just not a very extreme amount.

When we release the next version of the app, we'll have test cases to help you guys make sure that it's calculating the right values, if you feel like optimizing your own version.
ID: 4851 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Crunch3r
Volunteer developer
Avatar

Send message
Joined: 17 Feb 08
Posts: 363
Credit: 258,227,990
RAC: 0
Message 4852 - Posted: 20 Aug 2008, 16:10:48 UTC - in response to Message 4851.  


Given what our code is doing, honestly the only way i can see this kind of speed improvement being possible is through vectorizing most of the floating point operations in all the loops of our code. There's really no way to increase the algorithmic runtime of what we're doing (it's pretty simple -- just a sum of a lot of floating point ops).


What i did was a complete replacement for calculating the integrals, the worse performance hog is the GAUS_LEGENDRE and the functions that use it.

And yes it's completely vectorized ( not through compiler... it's using hand coded asm routines)...

And to repeat again what i've said in another thread...
i did run tests for several days and all results did match the stock app.


When we release the next version of the app, we'll have test cases to help you guys make sure that it's calculating the right values, if you feel like optimizing your own version.


creating a test case is easy.
simply download/copy the following files:
stars_convoled.txt, stars.txt rename a WU file to parameters.txt put that in the same dir as the app and you're ready to go.
compare the "likelyhood" of both apps in the "out" file, they should match.
That's it.



Join Support science! Joinc Team BOINC United now!
ID: 4852 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Kevint
Avatar

Send message
Joined: 22 Nov 07
Posts: 285
Credit: 1,076,786,368
RAC: 0
Message 4853 - Posted: 20 Aug 2008, 17:48:06 UTC


Travis,

One of the things that I think you are missing here.

This project works on an fixed credit system. 1 WU = xxx amount of credit.

So if a host can produce 10000000 VALID(the key is valid) WU's in 1 hour, then the credit granted should be the same per WU produced. Doing it any other way is non-scientific and not proper. Placing an upper limit on the amount of credit any host gets granted is not the proper thing to do. What do you think that Bill Gates, or Steve Jobs, or Warren Buffet would say this this type of crazy thinking, limiting the amount someone can get paid just because they happen to be produce more than crazy Joe down the street is absolutely ridicules. Maybe you need to run for some government office and see if you can get these types of crazy socialistic idea's passed.

Of course, like I said, the key is producing a valid result.
ID: 4853 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile ChertseyAl
Avatar

Send message
Joined: 31 Aug 07
Posts: 66
Credit: 1,002,668
RAC: 0
Message 4855 - Posted: 20 Aug 2008, 18:43:03 UTC - in response to Message 4839.  

I've tried to stay out of this discussion, but I'm afraid I feel the need to comment now :(


There are two problems:

1. We haven't tested his app to make sure the results are correct.


Then ban the user. Don't send work, don't accept results. You are admitting that you are accepting results that you don't trust. Not very scientific is it?


2. He was getting something along the lines of 200+ credit for around 3-4 minutes of work.


I'm failing to see why this is a problem. The Kredit Kops might see it as a problem, but in real life who cares? Is he going to buy a Ferrari by doing this?

and if you don't think the second has anything fishy about it... well i don't know i can say to convince you otherwise

This is a very fishy statement. Is your problem that the Kredit Kops don't like high RACs, or is your problem that you don't like the official code made to look like it was written by a grad school code monkey? For the avoidance of doubt, I am not suggesting that the latter is the case, it's just a wild guess.

I'll (partially) show my hand here. I (actually, the company that I own) used to make serious money from optimising real-time embedded software. I'd generally think that a 20-times speed improvement was the minimum that we could deliver to a customer for the outrageous fees that we charged. Sometimes we could do 50-times. The fact is that a lot of progammers just convert a formula or algorithm into code. They transliterate, and no more. Experts are the guys that take something basic and make it something special. Or in many cases take something useless and make it into a product.


Personally, I think that's way more credit than any project should be awarding


For me, that's a killer blow to the project. You are not DA (bow down before him mere mortals). Maybe you are DA. I just think that's a crazy statement and once you start down that particular route you will end up like Cosmo.

I've avoided looking at the code of any project, or indeed the BOINC code itself, as I've retired now, and don't want to be sucked into working 24 hour days to get code running faster. Been there, done that, got the Ferrari ;)

I'm thinking that I should detach from this project, based on your post (and another which you made which appears to be full of cr4p), but I'm really keen on the science. But if I suspect that you are holding it back (and, frankly, that's what it sounds like to me), I'll use my CPU cycles elsewhere.

Well, that was a jolly good rant. A waste of time as nobody will read it, but there you go.

Al.
ID: 4855 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 4856 - Posted: 20 Aug 2008, 19:12:30 UTC - in response to Message 4853.  


Travis,

One of the things that I think you are missing here.

This project works on an fixed credit system. 1 WU = xxx amount of credit.

So if a host can produce 10000000 VALID(the key is valid) WU's in 1 hour, then the credit granted should be the same per WU produced. Doing it any other way is non-scientific and not proper. Placing an upper limit on the amount of credit any host gets granted is not the proper thing to do. What do you think that Bill Gates, or Steve Jobs, or Warren Buffet would say this this type of crazy thinking, limiting the amount someone can get paid just because they happen to be produce more than crazy Joe down the street is absolutely ridicules. Maybe you need to run for some government office and see if you can get these types of crazy socialistic idea's passed.

Of course, like I said, the key is producing a valid result.


While nice in concept, this project isn't going to be giving any user 200+ credit every couple minutes. The project now works on a fixed credit system up to a point. If you want it to be a true fixed system, to keep the max amount of credit we award to any user at a rate we want, i'd have to lower the granted credit for all workunits by a factor of ~30. That isn't going to happen.

On another note, because my main priority right now is to get the next version of the application and assimilator up and running, after that, i'll look into new credit schemes. After discussing some things with the professors behind this project, we might move to awarding credit on some kind of logarithmic scale -- faster will still be more credit, but there's still going to be a limit.

ID: 4856 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Travis
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Aug 07
Posts: 2046
Credit: 26,480
RAC: 0
Message 4857 - Posted: 20 Aug 2008, 19:25:25 UTC - in response to Message 4855.  



2. He was getting something along the lines of 200+ credit for around 3-4 minutes of work.


I'm failing to see why this is a problem. The Kredit Kops might see it as a problem, but in real life who cares? Is he going to buy a Ferrari by doing this?


He's not, but it makes our project look bad: "oh hey! milkyway@home is giving away insane amounts of credit to a couple users!" and it's discouraging to our average user.

I have no problem with people using optimized code to get good amounts of credit from our project, but i do have a problem when this starts looking outrageous. Either way if Crunch3r continues to run his optimized app he'll still be getting more credit/time than any other user on this project. It wont be as much as before, but it'll still be #1. Having checked over the WUs he's been reporting they seem to be in line, so thats not a problem.


After some discussions today with the other people on our project, we'll probably be moving to some kind of logarithmic credit awarding system; which will still benefit people with optimized apps -- just not as much as happened in the particular case here. Faster will always be more credit, there will just be diminishing returns and an upper limit. I'm hoping this will keep the optimizers and our average users happy. If not, there's really nothing else I can say about the matter.


Also, since this thread (and most discussions about the mythical "credit") seems to have a magic way of preventing people from keeping to constructive criticism and arguments, and bringing out personal attacks, I'm shutting down the thread. I really don't have any more to say about the matter than what's already been said, and it doesn't seem anyone else has anything constructive to say either.
ID: 4857 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Previous · 1 · 2 · 3 · 4 · 5

Message boards : Number crunching : I've had enough !!!

©2024 Astroinformatics Group