Message boards : Number crunching : New credit system now being tested at RALPH@home
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · 8 · Next
Author | Message |
---|---|
Whl. Send message Joined: 29 Dec 05 Posts: 203 Credit: 275,802 RAC: 0 |
The only info needed is the average run time for each target on a particular spec machine....... Hi Danny, That sounds very good, but what happens when an AMD64 @ over 2.6Ghz is reported as a Sempron 3100+ as happened to me, until I changed the OS ? Or some of the massively overclocked systems on some teams that are reported as standard speed ? Would probably cause another huge row, at a cost to the project. I dont mean to have a go at you though Danny by saying that. [Edit] BTW, even after I changed the OS, Rosetta/BOINC still did'nt report the true speed the 64 was running at. It just reported the model, which would'nt really give a true reflection of the work it has been doing.[/Edit] |
Jack Shaftoe Send message Joined: 30 Apr 06 Posts: 115 Credit: 1,307,916 RAC: 0 |
Going back to the zeroing out credit discussion and after hearing both positives and negatives to it wouldn't it be easier to just have 2 databases? That's a great idea in my opinion. I'm sure we can agree that everyone would get what they wanted if they could see both sides of the coin - just the new system on one set of graphs and combined new/old on another set. Not sure if it's easy/difficult to implement though. |
Jose Send message Joined: 28 Mar 06 Posts: 820 Credit: 48,297 RAC: 0 |
I am reaching a very high level of frustration with this type of thread. I have been working with a friend on a statistical analysis tool that could help pointing out the most egergious deviations in credit claims across the type of client being used, the actual work per credit claimed and the ammount of time crunched per credit and per model. Given the situation here , I have decided it is not worth continuing the testing. To put it mildly , I am having serious doubts about the undersatnding of statistical methods by some people here and I am worried at what is being passed as statistical analysis. To be even more blunt: I am having doubts about the intelectual honesty and the motivation of some of the main posters in this thread. I am tired of the erroneous comparison of the nature aof the work units in the different projects. THEY ARE NOT COMPARABLE as the projects themselves are not comparable. That is the fallacy that Boinc perpetuates: that differnt in nature projects can be measured by the same standards. I am tired of the whole optimized vs non optimized clients thing. specially the continuous use of the word cheating. This is the last time you are going to hear me say this: AS LONG AS THE USE OF OPTIMIZED CLIENTS IS NOT EXPLICETLY PROHIBITED IN A PROJECT , THEIR USE IS NOT CHEATING. Specially since the open nature of the BOINC code allows for changes and modifications in the code. I am tired on the editorializing done in some of the statistics to back up claims that some teams, what the hell, a specific team is cheating. It is a known fact that I am a member of that team. I KNOW that the cheater tag that some of the people here are trying to tag on us is unfair and borders on the slanderous. I am giving fair warning: Attempt to flame my team and my team members at your own peril... Flame and you will be roasted. If you flame us, do not expect us to take your slanders lying down. I AM NOT THE MOST GENTLE OF PEOPLE , AND I AM NOT THE MOST SUAVE OF PEOPLE... this BS against my team has to STOP. Fair warning again: Flame at your own risk. If you dont like the fact that in my team we crunch for Rosetta as if Rosetta were our wife, lover or best friend and that we only know one speed: full blast and that level dedication (including the investment in state of the art cpus, mobos and memory ) yields the numbers we are getting and THAT WE WILL GET REGARDLESS OF THE CREDIT SYSTEM used; I have news for you: tough noogies.... our team is commited to Rosetta because we believe in the project worth so if you thing we are going to fade out because your baiting and jealousy driven harping ..think again...We will not go away. As to the tracking of cheats..the real ones... we are still doing it... So if you are offended by what I have said... I really dont care. I am going back to work in what it ismportant: the crunching of models. But I have given fair warning...dont act surprised if you ontinue to question the integrity of my team (or for what it is worth the top teams in this project) if you dont get a cool reaction. We have tried reason: obviously your agenda in attacking the top teams doesnt allow for reason. Okies ...flame back at will :P |
suguruhirahara Send message Joined: 7 Mar 06 Posts: 27 Credit: 125,766 RAC: 456 |
I am tired of the whole optimized vs non optimized clients thing. specially the continuous use of the word cheating. This is the last time you are going to hear me say this: AS LONG AS THE USE OF OPTIMIZED CLIENTS IS NOT EXPLICETLY PROHIBITED IN A PROJECT , THEIR USE IS NOT CHEATING. Specially since the open nature of the BOINC code allows for changes and modifications in the code. So why don't project leaders prohibit the use of the optimised client? According to posts on seti@home forum, at least the source code of the 5.5.0 was not available, which means that no one except the code writer cannot know what the application does. Even more without publishing source code distributing the software violates GPL. I'm not sure whether even using the software violate it or not, though. |
Mats Petersson Send message Joined: 29 Sep 05 Posts: 225 Credit: 951,788 RAC: 0 |
I am tired of the whole optimized vs non optimized clients thing. specially the continuous use of the word cheating. This is the last time you are going to hear me say this: AS LONG AS THE USE OF OPTIMIZED CLIENTS IS NOT EXPLICETLY PROHIBITED IN A PROJECT , THEIR USE IS NOT CHEATING. Specially since the open nature of the BOINC code allows for changes and modifications in the code. Using software produced by GPL-breakages is not illegal - but probably immoral. Producing code without supplying source-code is definitely a breach of contract, the producer of such code is breaking the law, even if he/she/it/they are not gaining anything from doing so, since part of the GPL is that the source code must be made available to anyone, and it must also be available for another three years ... -- Mats |
kevint Send message Joined: 8 Oct 05 Posts: 84 Credit: 2,530,451 RAC: 0 |
Ok first of all I really dont like seeing SETI and Rosetta in the same sentance. They are 2 different projects and point systems so leave the sepprate. EXCACLY my point - you appear to have started Rosetta a few months ago, have a total RAC of 120, and what - a total of 80 credits - Yes, you would not care - - you are what we call a ZERO RACer. You have not spent not even a single dollar, euro, etc.. on the project. you have not investment in time nor money. Therefore you would not care if you are zeroed out. And as for doing the credits again.. I don't WANT to do them again- that is the most stupid thing I have heard in a long time. On the other hand - there are those that have invested time and money into the project and to just toss out those accomplishments are not only absurd but could very well be project damning. As I know that I would not be back, and there are others, many others that would just leave and go crunch for SIMAP or QMC or something else. As far as SETI and Rosetta being different projects I agree - but there MUST be a cross project equalization of credit systems. That is one of BOINC's features, the abilty to crunch several projects and be granted similar credit for these project. Don't be negative towards other project team, if SETI.USA were to turn focus to Rosetta - it would stomp and in no time be the number 1 team here. People crunch what interests them. I am interested in many things so I crunch many projects. And a fair cross project equalized credit system is important. Are you new to DC ? I have been crunching DC since May of 1999 - SETI.USA |
Jack Shaftoe Send message Joined: 30 Apr 06 Posts: 115 Credit: 1,307,916 RAC: 0 |
So if you are offended by what I have said... I really dont care. Not offended, just annoyed to see interesting ideas buried in more long angry posts full of noise. Argh... To get back to point, resetting everyone to zero is not an option. Too many people don't like that idea and threaten to leave. I don't support that anymore. What about the 2 graphs idea? One for new system only, one for total in both systems? If there are objections to this - what are they and why? I would like this. Can it be done? |
suguruhirahara Send message Joined: 7 Mar 06 Posts: 27 Credit: 125,766 RAC: 456 |
there MUST be a cross project equalization of credit systems. That is one of BOINC's features, the abilty to crunch several projects and be granted similar credit for these project. Same here. A few adjustments can go, but the completely different system cannot as far as this project applys BOINC framework at least, since the new system could change the relationship between crunching and 1 credit granted. just annoyed to see interesting ideas buried in more long angry posts full of noise. Argh... Me too. Moderators should do something here... |
[B^S] thierry@home Send message Joined: 17 Sep 05 Posts: 182 Credit: 281,902 RAC: 0 |
The word 'cheaters' was used in a lot of threads and most of the time misused. I think many people want to point out the fact that if some use an optimized client and others not, they cannot compare their score to others (especialy when they are low in credits ;-)). To summarize: - The cheaters is the guy how do something somewhere to multiply his credits by a few zillions. - The optimized clients are there, available, free and the projects don't care if you use it or not, so everybody is free to use them (or not). - Projects like CPDN, Seti, Seti Beta, Einstein, Rieselsieve have a fixed and standard credits system, so no problem with optimized client - Projects like Predictor, Sztaki, LHC, Simap, Primegrid, ... have a quorum of three results (sometime two), so no problem either. - Rosetta project team is working on a new credit system. - Finally QMC is the only project where the optimized client use can be discussed. So it's maybe the time to stop all those silly discussions/lists/... Or we can start new flame threads in the QMC forum for those who are interested by flaming (I'm sure there are some). |
suguruhirahara Send message Joined: 7 Mar 06 Posts: 27 Credit: 125,766 RAC: 456 |
Cudgel your brains about the new credit system, visit RALPH's forum, and write your ideas down on that thread. Rosetta project team offers you all to post suggestions here. edit: not only thierry@home but others should go there |
[B^S] thierry@home Send message Joined: 17 Sep 05 Posts: 182 Credit: 281,902 RAC: 0 |
Thank you. I already knew it and I have already made it. |
njkid32 Send message Joined: 27 Mar 06 Posts: 15 Credit: 80 RAC: 0 |
Ok first of all I really dont like seeing SETI and Rosetta in the same sentance. They are 2 different projects and point systems so leave the sepprate. Ok I asked the rest of my team to stop coming here and fighting with you children and I am going against my own word. But, if your going to talk to me this way I just cant resist.... Let me just tell you that what RAC that you see there was a long time ago and I crunch for VNS. I am the one with the Kentsfield at 3.5ghz and Conroe at 4ghz. https://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=270793 I am not new to DC and if you want to show whos toys are better lets go right ahead cuz YOU WONT WIN!! If you think that SETI guys want to come to Rosetta and stomp all over XS man, your crazy... But, if you think you can do it bring it kid! So dont sit here and tell me that I am nobody cuz from the 30th page where your comps are keep clicking the back button till you get to page one then look at the top. Thats why my team call him Clark Kent. |
suguruhirahara Send message Joined: 7 Mar 06 Posts: 27 Credit: 125,766 RAC: 456 |
@njkid32 calmdown. |
Scribe Send message Joined: 2 Nov 05 Posts: 284 Credit: 157,359 RAC: 0 |
Gawd here we go again, back to the old pissin contest! Why don't you all just wait and see what Dr Baker's choice is, instead of trying to outguess what they will eventually choose to do. |
senatoralex85 Send message Joined: 27 Sep 05 Posts: 66 Credit: 169,644 RAC: 0 |
Hmm, this thread is getting way off topic. May I ask that we focus on the new credit system as best we can so that when the Baker lab reads this post they can better address our concerns. Thanks mmciastro for your information. Keep us posted on what gooes on RALPH as I do not have time to keep up with it and like periodic updates! Jose, I would suggest talking with the Rosetta @ home staff about your statistical tool. I would agree with you that most of the users on the boards here (including myself) do not have the necessary tools to perform some of the statistical analysis that you seem to be eluding to. I have a hard enough time applying standard deviation. Its a great idea and if the staff is willing to work with you on it, I think it would be an asset to this project. What does everyone else think about using statistics to catch dishonest crunchers? If so, do you have any suggestions on a system so that honest people do not get upset when they are mislabled as dishonest? |
dcdc Send message Joined: 3 Nov 05 Posts: 1830 Credit: 119,208,549 RAC: 2,517 |
The only info needed is the average run time for each target on a particular spec machine....... Yeah - its a good point, but I don't think we'd need to know specifically what speed the computers run at - just that their config hasn't been changed during the time they've been running rosetta. It might be possible to calibrate the old jobs from when they were run on the inhouse computers, although I'm sure a fairly large sample of computers could be put together that have been running from very early on which would give a fairly accurate measure. I'd have a go myself, but i've not got the spare time at the moment. I think it'd just be a case of taking a pool of (say 20) computers, and then for each job type that's been released we just need to know the number of models produced and the time taken so the average time per model can be calculated for each computer. The new jobs can then be run on these computers and the average time taken per model calculated again. Then it's just a case of working the following out for each computer: A new job New job total time = 13900 New job number of decoys = 32 Average CPU time per decoy (Xnew) = 434.4 Credits awarded per decoy (CA) = 2.00 calculating the credit for an old job Job ABC123 total time = 14400s Job ABC123 number of decoys = 32 Average CPU time per decoy (Xold) = 450 Credits awarded per decoy = (CA / Xnew * Xold) = 2.07 Obviously you need to take an average over, say, 20 computers, but from this we can (even now) award credits fairly accurately right back to the start of rosetta. |
BennyRop Send message Joined: 17 Dec 05 Posts: 555 Credit: 140,800 RAC: 0 |
If those that are claiming 10 times or more the RAC for that cpu - are being flagged by Jose, and their models/hour is equal to the standard for that cpu - then I don't see any problem with presenting that quietly to Dr. Baker and letting the Rosetta team deal with it. They've got the records for each of the major WUs and can tell what the "average" credit/model for each of the major WUs is; and see how out of line potential problem systems are reporting. But it should be kept quiet and not the public lynching that has been going on for the past few months. Regardless, it's a short term problem - since it'll disappear when we move to the new credit system. Since the Rosetta labs has mentioned several times that the various optimized boinc clients were okay to use - it's a little late to normalize their scores for running the client with Rosetta. Once you've handed out a bowl of crisps/chips and given the crowd permission to eat them - once they've eaten the bowl of crisps/chips you can't ask for them back. (Although it's been pointed out that some here might actually attempt that..) It's time to forget about the war of the benchmarks and move on to the upcoming War of Model Production. The most interesting bit will be finding which cpus produce better than they benchmarked with the old system.. in which case, if we have the benchmarks being done by the best benchmarking cpus under the old system, some cpus will get a really nice boost in scores if they can outproduce the benchmarking cpus. Take a look at the credit system thread, and post your suggestions for possible improvements. |
gomeyer Send message Joined: 11 Jun 06 Posts: 5 Credit: 181,323 RAC: 0 |
Just a couple of general thoughts about a new credit system from someone new to R@H, if I may. My theme will be Equity. Although no credit system is likely to make everyone happy achieving some level of Credit Equity within the Rosetta community would be very beneficial. Judging by some of the views expressed in this thread I’d say such an attempt is appropriate. However the most important potential gain to be made by a new credit system would be with one that attempted to create better Cross-Project Equity. BoincStats and similar sites make competition across projects inevitable. People are less likely to join any project that is not at least on a par with others and we all know how important more participation is here. Also, just a guess but I’m thinking some current Rosetta participants might consider increasing their share if points were awarded with something approaching equity. I for one would. My point is that I hope that any new credit system would specifically address these two issues. |
XS_Vietnam_Soldiers Send message Joined: 11 Jan 06 Posts: 240 Credit: 2,880,653 RAC: 0 |
After reading through all the comments and ideas, I'd be interested in knowing exactly how many WU are actually getting "crunched" on the average day/week,etc.. Not points, but actual work units.. Is there a moderator or someone who can find this info for us? Might be interesting to see the breakdown. Thanks for your time, Movieman |
Hoelder1in Send message Joined: 30 Sep 05 Posts: 169 Credit: 3,915,947 RAC: 0 |
After reading through all the comments and ideas, I'd be interested in knowing exactly how many WU are actually getting "crunched" on the average day/week,etc.. I think the number you are looking for is actually listed on the homepage under server status: Successes last 24h: 132,647. If I am not mistaken these would be the number of WUs crunched in the last 24 hours, but please note that a WU takes anywhere from 3 to 24 hours to complete, depending on the user's preferences. Perhaps structures per day would be a more meaningful number. If David Kim's estimate of 2 credits/decoy that he uses for his Ralph tests is approximately right than the structures per day would be roughly 'credits last 24 hours' divided by 2 or about 2,000,000 structures per day. I hope this helps, -H. Team betterhumans.com - discuss and celebrate the future - hoelder1in.org |
Message boards :
Number crunching :
New credit system now being tested at RALPH@home
©2024 University of Washington
https://www.bakerlab.org