| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: BOINC Points post in Member News |
| No member browsing this thread |
|
Thread Status: Locked Total posts in this thread: 277
|
|
| Author |
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Sorry, I can't give you direct contact information. WCG will deal with your inquiry promptly, but please bear in mind that credit issues do not rate highly on their list of priorities. It is more important to do the science, and keep the vast majority of members satisfied. WCG cannot spend a disproportionate amount of time catering for the needs of supercrunchers. They simply lack the resources. I trust you understand this. I'm not asking for a "disporportionate" amount of their time. I'm asking for some information. What is wrong with that? Every time that someone brings up a credit issue the canned response is that the people don't have the time to answer as they are tied up with the science end. When will you stop using this excuse to evade problems. When people finally understand that to have a sucessful DC app you also have to have a fair and equitable credit system those apps will do much better and without all this dialogue. Look at F@H and D2OL, no issues on these points at all. OK, now lets get back on topic. You don't have the information I need so don't feel any need to reply to my posts on this issue. I'll wait until I hear from them or until someone else from WCG with more info comes online. Thanks.  [Edit 1 times, last edit by Movieman at Nov 13, 2006 7:38:49 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
So they have the time for the whingers who didn`t like opti`s but no one else......
----------------------------------------Be careful Didy you`re one step away ! JohnMurphy I`d go along with that and I bet it would provide a much fairer system than at present and far far simpler ! [Edit 1 times, last edit by Former Member at Nov 13, 2006 6:23:00 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Oh, I have it, I'm just not able to give it. You will have to make do with the normal channels. I believe WCG read all emails on the day they are received, or the working day after. They are then passed to the person best able to deal with the query, or to the project scientists for project related queries.
Have patience. I imagine they are still dealing with the weekend backlog. WCG have already spent considerable time recently dealing with the credit issue. They want it to be fair just as much as you and me. |
||
|
|
Movieman
Veteran Cruncher Joined: Sep 9, 2006 Post Count: 1042 Status: Offline |
Oh, I have it, I'm just not able to give it. You will have to make do with the normal channels. I believe WCG read all emails on the day they are received, or the working day after. They are then passed to the person best able to deal with the query, or to the project scientists for project related queries. Have patience. I imagine they are still dealing with the weekend backlog. WCG have already spent considerable time recently dealing with the credit issue. They want it to be fair just as much as you and me. Now this is an answer I can understand. Let's get one thing straight, we are not enemies. At least we should not be so.I know you have preconceived views of me and I can live with that. Just understand that what you and I both want is the same thing: To bring in more machinery to WCG to help them do their work. This is my way of trying to do just that. The better the credit system, the more machinery I can get devouted to it. Yes, it is that simple an equation. Thank you. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi Whl. The points would be determined by the PPH of the benchmark machine. This could either be what that machine actually benchmarked or a figure determined to be fair by the project. it doesn't matter Like I said I'm no Mathematician. what I have attempted to do is create a set of ratios like a gearbox or transmission. If your CPU represents 1st gear, you will travel a certain distance/hour/1000rpm. 2nd gear a different but equally constant ratio. and so on. I'm very uncertain of my math and if someone could correct it or make it clearer I'd be pleased. I'm sure that the concept is sound. If the benchmark machine is a 1.5Ghz Pentium 4, just as an example, and it crunches a FAAH WU in 8 Hours and claims 80 points. It is producing 80/8 = 10 PPH for this work and that PPH will be a constant for that machine on FAAH WUs. An AMD blah blah may do the same WU in 5 Hours and claim 65 Points or 65/5 = 11PPH. But the Benchmark machine tells us that the work is worth 80 points so we know that the AMD should be rated at 8/5*100 = 160% That percentage would be a constant for FAAH work With the given example the AMD works for 5 hours to receive 10*5*160% = 80 points not 65 Had it worked for 6 hours, on a different WU, we'd have 10*6*160 or 96 points and we know that it would have taken the benchmark machine 9.6 hours to do the same work because they are a fixed ratio. So the work unit itself doesn't need to be benchmarked and longer and shorter WUs wont effect the ratio. Also, because we now know the constant of the AMD V's the Intel we can use either of them as a virtual benchmark machine against the C2D or Athlon64, celeron998 whatever... Cheers. ozylynx.  edit for clarity. lol Hi ozylynx, I like some of what you are saying there, but I cant see that working with the HPF2 project, as each job sent has a different random seed number, making the jobs vary in complexity. So in effect you could have a very fast machine taking 8 hours to complete a job and a very slow machine taking 2 hours, which would completely bugger it all up mate. ;-) Something similar to what you say, I could imagine working with HDC and FAAH though, as long as a fair benchmark could be established. It would take some tweaking to get that right. The way the quorum handles points at the moment would need to be scrapped for it to have any chance of working. (I do understand though that the quorum is necessary for the validation of the science on these 2 projects and understand why it is'nt necessary with HPF2 peeps). It would be a lot easier ozzy mate, if we were all using the same hardware and OS.  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi ozylynx, I like some of what you are saying there, but I cant see that working with the HPF2 project, as each job sent has a different random seed number, making the jobs vary in complexity. So in effect you could have a very fast machine taking 8 hours to complete a job and a very slow machine taking 2 hours, which would completely bugger it all up mate. ;-) Actually it is just that situation that it is designed to work best in. I've found another problem that may kill the idea entirely though (amazing what an hours nap will do) It wont work for overclocking as it stands. If a machine is overclocked after performance factoring it will score low. If clock speed is reduced it will score high. This might be manageable with a dynamic comparison via a Quorum but would probably lead to cheating in non quorum projects. Unless someone has a way to fix this it looks like it's in the bin. Cheers ozylynx |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Actually it is just that situation that it is designed to work best in. If you say so Ozzy. There is a bit more to think about when it comes to HPF though IMHO. ;-) |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
OK it wont work because of overclocking concerns. There may be other reasons and I'd like to know so I don't fall into the same trap again.
Here's why the concept should work in HPF: The benchmarking is done against another, predetermined but live machine. This effectively gives it an anchor to reality. The benchmark is also carried out in a real life situation doing exactly the type of work required of the machine, nothing more, nothing less. Its performance factor should remain constant in comparison to the benchmark machine. Uther than overclocking concerns, this means that the only influence on the points is the CPU time taken to perform the WU. That's why I stated that wildly varying times shouldn't have any effect, other than the desired one. If it takes a long time you get more points, shorter time less points. That is exactly the way that the BOINC benchmark should work but unfortunately doesn't. The only real difference is that it is 'anchored'. A subroutine to check for changes in FSB or clock multipliers since last benchmark might work if it could be implimented. Any ideas? Cheers. ozylynx |
||
|
|
knreed
Former World Community Grid Tech Joined: Nov 8, 2004 Post Count: 4504 Status: Offline Project Badges: 
|
I would like to thank everyone for shifting the discussion to something more useful - i.e. the benchmark itself which is still a problem.
It is interesting that zombie67 proposed the solution he did. I sent a slightly broader request to Dr David Anderson at BOINC today. I have not heard back and as we are one of many BOINC projects I will have to wait and see what happens. Fundementally, this problem is a BOINC problem and the best solution for it is that BOINC develop and release to the solution to all projects rather then all projects spending their time and energy developing their own solutions. Here is my proposal: Each application that runs (for most BOINC projects this is just one but for us it is multiple) projects will identify a 'Benchmark workunit'. This should be a short workunit but still reflect the how the application will run. This will be used in a two step process: 1) It will be run on all computers. The results from this will be used to create an average points for the benchmark workunit. This will be the net aggregate result from all computers. Since the goal of credit is to award the same amount of credit for the same amount of processing done. The average amount of credit for this 'benchmark workunit' will thus be the fair credit for that workunit. 2) All computers will periodically have the benchmark workunit sent to them to run. Based upon the average computed in step 1 and the length of time it takes to run the benchmark workunit the server will be able to compute the 'fair' points per second to award to the computer for results returned. This should have two effects: 1) The claimed scores will be vastly more consistent within each workunit. 2) The benchmark itself is based on the actual performance of the computer doing the work in question. Thus it will be much more accurate then any artificial benchmark could be. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Per app benchmarks I think have a lot of promise for correcting assessing contribution. Hopefully they will listen.
|
||
|
|
|