Observation of CreditNew Impact

Message boards : Number crunching : Observation of CreditNew Impact
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 15 · Next

AuthorMessage
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1379646 - Posted: 11 Jun 2013, 1:24:08 UTC - in response to Message 1379645.  

TBar, IMO all the data needs to be collected or it is not as valid as it could be.
Your arbitrary editing of the incoming data screws it up to some small degree.

Look around. It's the Credit Guesser that is being arbitrary. You need to re-evaluate your opinion.
ID: 1379646 · Report as offensive
Profile Gatekeeper
Avatar

Send message
Joined: 14 Jul 04
Posts: 887
Credit: 176,479,616
RAC: 0
United States
Message 1379650 - Posted: 11 Jun 2013, 2:04:01 UTC - in response to Message 1379646.  
Last modified: 11 Jun 2013, 2:05:04 UTC

Look around. It's the Credit Guesser that is being arbitrary. You need to re-evaluate your opinion.


Normally, I let these type of posts go right by me. But I can't let your line about re-evaluating opinions go by.

First let me make it clear. I agree completely that Creditnew is an unmitigated disaster. But, it's what we have for the moment. So, deal with it, or switch to another project.

As to your proposal to start cherry-picking AP workunits, do yourself a big favor. Go to your account preferences and untick both APv6 and "accept work of other types when no other work available". Doing otherwise is unfair to the science, and to those of us who believe in the science, as opposed to being in it for the credits.
ID: 1379650 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1379655 - Posted: 11 Jun 2013, 2:23:26 UTC - in response to Message 1379650.  

Look around. It's the Credit Guesser that is being arbitrary. You need to re-evaluate your opinion.


Normally, I let these type of posts go right by me. But I can't let your line about re-evaluating opinions go by.

First let me make it clear. I agree completely that Creditnew is an unmitigated disaster. But, it's what we have for the moment. So, deal with it, or switch to another project.

As to your proposal to start cherry-picking AP workunits, do yourself a big favor. Go to your account preferences and untick both APv6 and "accept work of other types when no other work available". Doing otherwise is unfair to the science, and to those of us who believe in the science, as opposed to being in it for the credits.

I'm pointing out a big problem that is causing other people to leave. I'm offering a solution, other than just whining as others are doing. Go back a few months and look at a conversation I had with someone complaining about them maybe receiving a few less credits because someone else was finding a way to keep their hardware working. I'm not in it for credits, but when I see a big problem that is causing people leave, I feel I must say something. Try and find an example of me Cherry-Picking tasks. Just because I point out an example, is not saying it's being implemented. You did see the suggestion...didn't you? About having the scheduler assign the tasks? Here's another suggestion, try and find one example of me Cherry-Picking tasks that I chose to run. I hope you aren't implying that I shouldn't have a choice of which tasks I run. You aren't doing that are you?
ID: 1379655 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19402
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1379658 - Posted: 11 Jun 2013, 2:33:16 UTC
Last modified: 11 Jun 2013, 2:55:11 UTC

The way CreditNew is supposed to work,
it should recognise that it is granting low credits for some tasks
and then increase the the multiplication factor.

If you delete the tasks that get low credits then CreditNew will think it is giving the correct credit and NOT increase the multiplication factor.

Therefore if you wish the average granted credits for an application to be increased you should delete those tasks which claim High credits.

edit] Except of course CreditNew will just deduce you are working shorter hour/day and dock your pay, in credits, accordingly.
ID: 1379658 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1379661 - Posted: 11 Jun 2013, 2:52:40 UTC - in response to Message 1379658.  

The way CreditNew is supposed to work,
it should recognise that it is granting low credits for some tasks
and then increase the the multiplication factor.

If you delete the tasks that get low credits then CreditNew will think it is giving the correct credit and NOT increase the multiplication factor.

Therefore if you wish the average granted credits for an application to be increased you should delete those tasks which claim High credits.

The problem is the Credit Guesser is comparing the times of the Fastest hardware and assuming that is the standard for the task. I think someone else commented about that a few days ago. Remember that? It would straighten out after the slower hardware results came in? I think they later said that the problem should have corrected itself by now, it hasn't. Since it hasn't corrected itself, Force It. Match the Fastest Hardware with the Slowest Hardware. That is basically what was suggested by a few people a few days ago. That's what I'm suggesting now. I don't know why I'm meeting with such resistance by suggesting what was suggested by a few important people around here.
Shrugs...
ID: 1379661 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19402
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1379664 - Posted: 11 Jun 2013, 3:04:02 UTC - in response to Message 1379661.  

I think the only way to prove your case, is to go and read the description of CreditNew and then publish here a very clear detailed explanation of how it supposed to work.
Then once you have done that you should be able to argue your case convincingly.

I have tried to read the description, but failed to understand exactly how it works and more importantly I failed to understand how it was better than Eric K's flop counting.

I can only assume, from personal experience, that it was not an idea thought up by the great(?) Dr. A therefore it must be crap.
ID: 1379664 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1379666 - Posted: 11 Jun 2013, 3:16:34 UTC - in response to Message 1379664.  

I think the only way to prove your case, is to go and read the description of CreditNew and then publish here a very clear detailed explanation of how it supposed to work.
Then once you have done that you should be able to argue your case convincingly.

I have tried to read the description, but failed to understand exactly how it works and more importantly I failed to understand how it was better than Eric K's flop counting.

I can only assume, from personal experience, that it was not an idea thought up by the great(?) Dr. A therefore it must be crap.

I'm not going to waste anymore time on this. The observations agree with the reasoning put forth by the admin. When two fast GPUs are matched, the likely outcome is a low score. Match one of the same fast GPUs with a CPU and the score is normal. Why should I argue the observations? Maybe some people around here should spend some time making observations. Go ahead, look around and see for yourself. I'm not going to look up the post, I think it was Eric that made the post about slower hardware results correcting the issue. I could be wrong...about it being Eric...
ID: 1379666 · Report as offensive
Profile Gatekeeper
Avatar

Send message
Joined: 14 Jul 04
Posts: 887
Credit: 176,479,616
RAC: 0
United States
Message 1379669 - Posted: 11 Jun 2013, 3:24:28 UTC - in response to Message 1379655.  
Last modified: 11 Jun 2013, 3:25:16 UTC

You did see the suggestion...didn't you? About having the scheduler assign the tasks? Here's another suggestion, try and find one example of me Cherry-Picking tasks that I chose to run. I hope you aren't implying that I shouldn't have a choice of which tasks I run. You aren't doing that are you?


I did see the suggestion. However, AFAIK, creditnew is part of BOINC. The schedulers, splitters, etc. are part of seti. Why should the seti staff have to rework code to address a problem with another program entity over which they have (again, AFAIK) no control? And then, if Dr A changes creditnew, there's more program work for seti staff to change again.

As to having task choices, you can run whatever you like. I was running AP myself since the rollout, after a long hiatus waiting for the new installer. I've already stopped requesting AP's because of an issue (possibly BOINC-related) of forced termination at EOJ. When that issue is resolved, I'll decide to restart AP or not. I only suggested you could do the same with the creditnew issue.
ID: 1379669 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1379684 - Posted: 11 Jun 2013, 4:17:14 UTC - in response to Message 1379669.  

You did see the suggestion...didn't you? About having the scheduler assign the tasks? Here's another suggestion, try and find one example of me Cherry-Picking tasks that I chose to run. I hope you aren't implying that I shouldn't have a choice of which tasks I run. You aren't doing that are you?


I did see the suggestion. However, AFAIK, creditnew is part of BOINC. The schedulers, splitters, etc. are part of seti. Why should the seti staff have to rework code to address a problem with another program entity over which they have (again, AFAIK) no control? And then, if Dr A changes creditnew, there's more program work for seti staff to change again.

As to having task choices, you can run whatever you like. I was running AP myself since the rollout, after a long hiatus waiting for the new installer. I've already stopped requesting AP's because of an issue (possibly BOINC-related) of forced termination at EOJ. When that issue is resolved, I'll decide to restart AP or not. I only suggested you could do the same with the creditnew issue.

I wouldn't think it would be that difficult to have the scheduler include a CPU with each task, for a little while. It's does appear that might help the situation. I've got all my CPUs working APs, so, I'm doing my share.

Did you see the thread about the AP issue? Did you see all the time I spent on that issue? I put in months on that problem. I did find a work around that should work, it works for me. You might want to try the workaround before you give up. I haven't had one of those errors in weeks. Also, it only affects GPUs and more people don't have the issue than do. Even Mark only has it affecting 3 of his 9 machines. A large number of people will never have that problem...
ID: 1379684 · Report as offensive
W-K 666 Project Donor
Volunteer tester

Send message
Joined: 18 May 99
Posts: 19402
Credit: 40,757,560
RAC: 67
United Kingdom
Message 1379691 - Posted: 11 Jun 2013, 4:34:47 UTC - in response to Message 1379666.  
Last modified: 11 Jun 2013, 4:35:29 UTC

I am not defending CreditNew, I personally think it is a system that needs to be got rid of asap.

I would like to see a separation of BOINC and Seti at Berkeley, so that Seti can choose it's preferred method of granting credits like all other projects. And from what we have seen before,
1. Cobblestones - basically, benchmark * time, which fails because same computer different OS produced benchmarks so different it was unbelievable.
2. Flop counting - a quick rough way of estimating work done, this produced claimed credits for each task, by each computer doing that task, that were within 0.1 credits in 99.999999% of the time. The reason we have the "Flopcounter: 18023829946978.504000" in the results file.
3. CreditNew - not understood, probably not programmed correctly the worst method so far.

My vote goes for 2.

But at the moment it is what we have and it depends on RAC. RAC is based on a time of 7 days and is an exponential curve and you will only see the full results of the change after about five (5) weeks at a minimum, if it continues to fall.
(~63% change in 1st week, 63% of 37 (100 - 63) in 2nd week, ... total of 99.9% at end of week 5)

And as I posted elsewhere the average host only does about 2 tasks/week and the APR calculation for each app requires ten 100% completed validated tasks before it is brought into the equation, it could be months before we know the final outcome.

Unless the credit system is changed.
ID: 1379691 · Report as offensive
Profile betreger Project Donor
Avatar

Send message
Joined: 29 Jun 99
Posts: 11416
Credit: 29,581,041
RAC: 66
United States
Message 1379695 - Posted: 11 Jun 2013, 4:52:13 UTC - in response to Message 1379646.  
Last modified: 11 Jun 2013, 4:52:48 UTC

TBar, IMO all the data needs to be collected or it is not as valid as it could be.
Your arbitrary editing of the incoming data screws it up to some small degree.

Look around. It's the Credit Guesser that is being arbitrary. You need to re-evaluate your opinion.

TBar, all the following discussions have been about creditnew when we were talking about the database. With out the best database we can create what ever credit system is used the credit scores are BS.
ID: 1379695 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1379696 - Posted: 11 Jun 2013, 5:23:23 UTC - in response to Message 1379695.  

TBar, IMO all the data needs to be collected or it is not as valid as it could be.
Your arbitrary editing of the incoming data screws it up to some small degree.

Look around. It's the Credit Guesser that is being arbitrary. You need to re-evaluate your opinion.

TBar, all the following discussions have been about creditnew when we were talking about the database. With out the best database we can create what ever credit system is used the credit scores are BS.

If the database keeps collecting a larger percentage of data on fast GPUs against fast GPUs it is going to begin to 'assume' that such a case should be awarded a low score. If you don't correct that assumption, the problem may take a very long time to correct itself. If it ever corrects itself. This is not a new theory, it was thought the results of the slower machines would correct the problem. Why do you think matching a GPU against a CPU is a bad idea? It happens all the time. It just takes longer to sort out when you start from scratch and the first results are from GPU vs GPU.
ID: 1379696 · Report as offensive
kittyman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Jul 00
Posts: 51478
Credit: 1,018,363,574
RAC: 1,004
United States
Message 1379722 - Posted: 11 Jun 2013, 6:49:45 UTC
Last modified: 11 Jun 2013, 6:51:25 UTC

I do hope Dr. Anderson is losing a bit of sleep these days wondering why his CreditScrewed disaster is not working. Especially on Boinc, his pet project.

And perhaps wondering where in the devil he went wrong.

I understand this was a major change in the work and applications being used on this project. But his algorithms should have been better prepared for an event like this, perhaps.
"Time is simply the mechanism that keeps everything from happening all at once."

ID: 1379722 · Report as offensive
Darth Beaver Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 20 Aug 99
Posts: 6728
Credit: 21,443,075
RAC: 3
Australia
Message 1379879 - Posted: 11 Jun 2013, 13:40:46 UTC

7 days no change then respect to the users no . users respect to setting will begin

seti classic 3mill users 2.5 permantly connected

Boinc Seti@home 2.5mil users who freaking know how many permantly connect anymore

Credit not important !! 30,000 hrs seti classic 1,338 lost not counted

no respect to users

slowest system laptop , credit will be 90 now

2 other system money wasted !!

time to do real F scince save planet improve mans knowledge

OR SHOW RESPECT AND THANKS TO USERS
ID: 1379879 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1379901 - Posted: 11 Jun 2013, 14:36:08 UTC - in response to Message 1379722.  
Last modified: 11 Jun 2013, 14:36:27 UTC

I do hope Dr. Anderson is losing a bit of sleep these days wondering why his CreditScrewed disaster is not working. Especially on Boinc, his pet project.

And perhaps wondering where in the devil he went wrong.

I understand this was a major change in the work and applications being used on this project. But his algorithms should have been better prepared for an event like this, perhaps.

Trying to track how much computational work a machine has done in a day is a fine idea. However when the result of the tracking mechanism shows a machine, that processes more or less 24/7, changing as much as 40-50% a day. Then it appears as though something is wrong with the mechanism.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1379901 · Report as offensive
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 1379911 - Posted: 11 Jun 2013, 15:07:47 UTC
Last modified: 11 Jun 2013, 15:10:30 UTC

The change from v6 to V7 was perfect, all the builds works, all the optimized builds and installer works, beside and easy to compreend small bugs like the scheduler inhability to choose the right cuda version fast and the choose to sending Vlars to Nvidias, nothing that could obscure the shine of this new bright sun and to finish the process with a golden key then comes Creditnew... and put all down!

It´s a classic example of faulty planning, if Creditnew problem was old known, why don take some action BEFORE?

I´m with you Glenn, is not about credits, it´s abour user respects. Our hosts still crunching and doing science, even some who in the past runs stock apps, are running faster now due the optimized version, so teoricaly, beacause they are more, in the end more WU/day are doing now.

We could produce less WU/day (not 50% like the credit shows the real one was about 15-20% if you take the Vlars out of the equation the loss could be even less, math don´t lies) mostly because the autocorrelation new subroutine, but the resoults stills valid and now with some more scientific acuracy, that i belive anyone of us are ready to accept because is clearely good for the project, so we lost nothing, just our credits, and credits value nothing...

NO credits have a big value to us users. Credits is the only way we "simple users" could measure our daily production and feel we are doing our part. All Human not scientic concepts, but WE USERS are humans not host numbers... I belive that is the real part the admins "forget".

Could be not a scientific way to see the things since the same (or even more) science is still doing, but It´s human way to see the things, our insatisfaction is fully understable, is like someone take something from you without asking.

The simple answers, "i don´k know what is happening" "I hope that will be fixed on 4 weeks" "The problem is Creditnew" ETC was simply a way to try explaing something without no explanation.
ID: 1379911 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1379914 - Posted: 11 Jun 2013, 15:14:54 UTC

Here's another fine example;
Valid AstroPulse v6 tasks for computer 6991309
    3034421519 1261794911 10 Jun 2013, 10:50:10 UTC 10 Jun 2013, 22:45:39 UTC Completed and validated 6,047.62 940.14 60.97 AstroPulse v6 v6.04 (opencl_ati_100)
    3034175782 1261685412 10 Jun 2013, 4:25:49 UTC 10 Jun 2013, 15:32:14 UTC Completed and validated 5,684.67 642.85 137.87 AstroPulse v6 v6.04 (opencl_ati_100)
    3033863586 1261545686 9 Jun 2013, 21:00:17 UTC 10 Jun 2013, 10:50:10 UTC Completed and validated 3,600.30 255.01 182.67 AstroPulse v6 v6.04 (opencl_ati_100)
    3033566823 1261412259 9 Jun 2013, 14:23:13 UTC 10 Jun 2013, 10:50:10 UTC Completed and validated 8,436.70 364.68 372.00 AstroPulse v6 v6.04 (opencl_ati_100)
    3027654022 1257371134 4 Jun 2013, 8:51:15 UTC 4 Jun 2013, 12:06:43 UTC Completed and validated 5,762.35 696.26 431.68 AstroPulse v6 v6.04 (opencl_ati_100)
    2992477201 1243102165 11 May 2013, 19:50:58 UTC 12 May 2013, 14:06:03 UTC Completed and validated 10,967.15 5,885.53 760.20 AstroPulse v6 v6.04 (opencl_ati_100)


At that rate, his next score may be negative. There's something seriously wrong with those scores.

ID: 1379914 · Report as offensive
Profile betreger Project Donor
Avatar

Send message
Joined: 29 Jun 99
Posts: 11416
Credit: 29,581,041
RAC: 66
United States
Message 1379940 - Posted: 11 Jun 2013, 18:09:38 UTC - in response to Message 1379696.  
Last modified: 11 Jun 2013, 18:43:36 UTC

If the database keeps collecting a larger percentage of data on fast GPUs against fast GPUs it is going to begin to 'assume' that such a case should be awarded a low score. If you don't correct that assumption, the problem may take a very long time to correct itself. If it ever corrects itself. This is not a new theory, it was thought the results of the slower machines would correct the problem. Why do you think matching a GPU against a CPU is a bad idea? It happens all the time. It just takes longer to sort out when you start from scratch and the first results are from GPU vs GPU.

TBar, if fast GPUs against other fast GPUs is happening then the assumption is correct and trying to bias the database to pretend it is not and trying to bend it in order to get the different results is not good science.
As an aside I feel credit new is an abomination.
ID: 1379940 · Report as offensive
bill

Send message
Joined: 16 Jun 99
Posts: 861
Credit: 29,352,955
RAC: 0
United States
Message 1379966 - Posted: 11 Jun 2013, 18:44:57 UTC - in response to Message 1379722.  

I do hope Dr. Anderson is losing a bit of sleep these days wondering why his CreditScrewed disaster is not working. Especially on Boinc, his pet project


Given that people have been complaining for years
about Credinew I seriously doubt whether he has lost
any sleep at all. He just doesn't care.
ID: 1379966 · Report as offensive
Rolf

Send message
Joined: 16 Jun 09
Posts: 114
Credit: 7,817,146
RAC: 0
Switzerland
Message 1380013 - Posted: 11 Jun 2013, 20:23:15 UTC - in response to Message 1379966.  

I do hope Dr. Anderson is losing a bit of sleep these days wondering why his CreditScrewed disaster is not working. Especially on Boinc, his pet project


Given that people have been complaining for years
about Credinew I seriously doubt whether he has lost
any sleep at all. He just doesn't care.

+1
Does he even know what "taking care" means?
ID: 1380013 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 15 · Next

Message boards : Number crunching : Observation of CreditNew Impact


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.