Because windows makes up 87% of the boxes crunching for einstien, and 88% for seti. Linux has 8% in both and mac 5/4% respectively. All the other architectures have even lower shares of the total. This means that windows systems have the overwelming majority of the weight in determining how many credits to award per WU. All other architectures will float against that number depending on the relative optimizations of the specific projects clients. With the most 4.2x apps the intel mac has the same assembly running the inner loop as windows and linux and should theorectially give about the same credit/hour as a win/linux pc with an identical chip (give or take a few percent for OS overheads). The 11% gap you see is large enough that apparently either the mac/intel app for einstien isn't as tightly optimized as the windows one, or the seti app is more tightly optimized. Does anyone have a such a system to compare against? WIth only ~800 project wide hunting randomly to try and find one would be easier said than done (150 random machines checked would only be a 50% chance of finding one).
The PPC in your older systems has a completely different internal design, and cannot be expected to perform at the same levels as an x86 chip across different projects. Sometimes, as was the case with the first half of S4 (mac client was ~3x faster than x86) the result will be to your benefit, other times it will not. If you're concerned about maximizing your credit earned, move your machines to whichever projects calculation needs most strongly maximize your architectures unique capabilities.
I'm talking about the client ---> BOINC
Application ---> einstein_S5R1_4.24_windows_intelx86.exe
Alright, but I don't understand what influence Boinc can have on the applications ?
It used to have with third part Boinc when credit correction was usefull. At least in S@H and E@H, it doesn't matter what Boinc you use as the credits are "inside" each wu. And as far as I know, there are no other optimized app for the rest of the projects.
Generally speaking, I agree with you about Boinc concern towards its users, but I doubt Bruce is 100% OK with all this (lack) of consideration.
Private : I'd like you have a look to BoincStudio here please : http://forum.boincstudio.boinc.fr
For the moment, we are waiting for a new version (current is still working fine), but we may be intereted in you for beta test. If you can register there, I may be able to contact you in PM for further details. Thanks. /Private
But we can argue as much as we like, the "projects" won't listen anyway to their users ;-(
Believe me, we DO listen. But sometimes there is honest disagreement about what is in the best interests of the project and the BOINC community. In such cases, I have to think carefully and then decide. This decision might not be what you would have chosen, but it doesn't mean that I am not listening. I am forever grateful to you and the other people who run Einstein@Home. Without your support, we couldn't do any of this analysis.
Here's some more data. I started collecting data Aug 28th. Below is a chart from my mobile AMD64 3700+ for the "Dates returned period" Aug 28-today. The lowest granted credit/hour are granted for work units which were "long ones", the higher correspond to the "short ones". The calculated "claimed credit" using the boinc benchmark is 13.33 credit/hour (standard Boinc client 5.6.0).
It appears users could "cherry pick" the short ones for 25% higher credit, and still place Einstein at the highest granted credit/hour for any project as seen here (note: the Einstein data shown on the "cross project comparison" is for OLD data using 4.02). This is what it looks like for those running standard clients, those running "third party" boinc core clients should be seeing higher "granted credits" on other projects which involve a quorum, since they'd be paired up with others using them at roughly the same rate as I do, except when they're paired up, they get the higher claims.
But we can argue as much as we like, the "projects" won't listen anyway to their users ;-(
Believe me, we DO listen. But sometimes there is honest disagreement about what is in the best interests of the project and the BOINC community. In such cases, I have to think carefully and then decide. This decision might not be what you would have chosen, but it doesn't mean that I am not listening. I am forever grateful to you and the other people who run Einstein@Home. Without your support, we couldn't do any of this analysis.
Corollary: assuming that other BOINC projects do the same, this will tend to make hosts move to the projects that they are best suited for.
Does this mean that you recommand E@H users to stop the project on hosts that would grant more credit/hour on other projects ?
Would this mean that I should keep my Opteron 170 on E@H and move my other boxes (XP 2400+, P-M 1.6, P4B 2.4 and P4C3.2) on, let's say, S@H ?
In my opinion, this is quite a contradiction with this :
Quote:
Rationale: this way, people will chose projects based on their scientific and other merits, and likelihood of success and impact, NOT for other reasons such as credit granted.
Because of you Bruce, better said, thanks to you Bruce, I feel great crunching for E@H. For the opposite reason, I don't feel like crunching again for a project I quit a few months ago. Because of this average stuff that doesn't average anything, I can not choose to attach to the project I like.
I think, there are 3 kinds of users in Boinc :
1/ those mainly concerned by science.
2/ those more concerned by credit than by science.
3/ those equally concerned by credit AND science.
An average stuff about credit/hour is ignored by group 1. It is a concern for groups 2 and 3. Again, if credit wasn't a known concern for a majority of Boinc users (groups 2 and 3), projects would not maintain top users, computers, teams rankings ;) Guys like Willy at Boincstats.com and otherrs would not maintain stats.
It is well known that E@H and S@H are the two best projects for grant credit. Strange, these two also attract (by far) the greatest numbers of users and hosts (figures available here http://www.boincstats.com/index.php).
I understand it is not easy to state : you have to deal with both E@H interests AND the Boinc community interests. The Boinc community is made of projects, but also made of users. So your decision has to meet the interests of E@H, other projects and users. As I said, user interest in credit/hour is an important factor for a majority of us. The average thing is an interest only for projects.
That's the problem.
I would suggest this solution :
- Grant credit "inside" wu should be mandatory for every boinc project (no more third part boinc credit correction)
- Credit/hour/cpu should be "averaged" among plateforms (ie cpu architecture) and not among hosts or projects.
The last point means : a netburst P4 should get an equally close approximate credit/hour/cpu whatever the project. Same thing for A64, XP's, P3 and Core architecture. This way, you avoid the present situation :
Concerned with credit ? => A64 on E@H, the rest on S@H, nothing on other projects.
Not concerned with credit ? => do what you want.
I know nothing in programming, so I don't know if this solution is possible, too complex or what. But I think it could solve the current probem.
To be honest I am not understanding the negativity about this attempt to equalise redit across the projects.
Historically for over a year the project with the highest credit/hr has been CPDN, but with the optimisation of Einstein application it has shot up the table and has taken the correct steps to decrease it.
Seti has slipped in the credits/hr but there has been an increase in the latest version in the Beta app.
At the moment on my Pent M I am seeing:
Einstein (7 long 112.74 credit units) 17.66/hr
SetiBeta (17 @ ar=0.4265) 16.58/hr
CPDN (sulphur) 16.94/hr
So from a purely personal view on one computer, I would say that Einstein is still awarding about 5% more than other projects.
I also agree, the Credits Einstein has been granted were quite a bit too high compared to other Projects I know.
Despite having absolutely equal shares here, EAH still easily scores the majority as seen in my Sig.
Not only does it affect us individual crunchers, it also can easily put the whole Project into a wrong light (already, there were discussion about an artificial Credits inflation to attract more users, which would put EAH into discredit which it truly doesn't deserve)
Ideally, it's really best when Users get about the same credit, regardless of which Projects they join.
Has credit been reduced once again (i.e. today)? Or did I get a slightly smaller WU? My Results for user shows the last reported unit claiming only 108.21 (vs. the most recent range of 112-114).
Has credit been reduced once again (i.e. today)? Or did I get a slightly smaller WU? My Results for user shows the last reported unit claiming only 108.21 (vs. the most recent range of 112-114).
Has credit been reduced once again (i.e. today)? Or did I get a slightly smaller WU? My Results for user shows the last reported unit claiming only 108.21 (vs. the most recent range of 112-114).
Yeah, they did it again.
Anyhow 112-114 sounds bad make it a round 100. or 90 of even 50.
This will increase my time to claim the No. one spot in my team any further...
But it's fair to fight credits claimed/granted using the old system vs. the credits i need to get granted now.
Shorten it further ...should keep me from reaching my personal goal another 14 weeks. Or even more ... make it 28 weeks :D
P.S.
@Mr. Castro/Tony post your crap stats somewhere else... and NOT as a reply to my post... or i'll freak out!
Because windows makes up 87%
)
Because windows makes up 87% of the boxes crunching for einstien, and 88% for seti. Linux has 8% in both and mac 5/4% respectively. All the other architectures have even lower shares of the total. This means that windows systems have the overwelming majority of the weight in determining how many credits to award per WU. All other architectures will float against that number depending on the relative optimizations of the specific projects clients. With the most 4.2x apps the intel mac has the same assembly running the inner loop as windows and linux and should theorectially give about the same credit/hour as a win/linux pc with an identical chip (give or take a few percent for OS overheads). The 11% gap you see is large enough that apparently either the mac/intel app for einstien isn't as tightly optimized as the windows one, or the seti app is more tightly optimized. Does anyone have a such a system to compare against? WIth only ~800 project wide hunting randomly to try and find one would be easier said than done (150 random machines checked would only be a 50% chance of finding one).
The PPC in your older systems has a completely different internal design, and cannot be expected to perform at the same levels as an x86 chip across different projects. Sometimes, as was the case with the first half of S4 (mac client was ~3x faster than x86) the result will be to your benefit, other times it will not. If you're concerned about maximizing your credit earned, move your machines to whichever projects calculation needs most strongly maximize your architectures unique capabilities.
RE: I'm talking about the
)
Alright, but I don't understand what influence Boinc can have on the applications ?
It used to have with third part Boinc when credit correction was usefull. At least in S@H and E@H, it doesn't matter what Boinc you use as the credits are "inside" each wu. And as far as I know, there are no other optimized app for the rest of the projects.
Generally speaking, I agree with you about Boinc concern towards its users, but I doubt Bruce is 100% OK with all this (lack) of consideration.
Private : I'd like you have a look to BoincStudio here please :
http://forum.boincstudio.boinc.fr
For the moment, we are waiting for a new version (current is still working fine), but we may be intereted in you for beta test. If you can register there, I may be able to contact you in PM for further details. Thanks. /Private
RE: But we can argue as
)
Believe me, we DO listen. But sometimes there is honest disagreement about what is in the best interests of the project and the BOINC community. In such cases, I have to think carefully and then decide. This decision might not be what you would have chosen, but it doesn't mean that I am not listening. I am forever grateful to you and the other people who run Einstein@Home. Without your support, we couldn't do any of this analysis.
Here is another post by me on the same topic.
Cheers,
Bruce
Director, Einstein@Home
Here's some more data. I
)
Here's some more data. I started collecting data Aug 28th. Below is a chart from my mobile AMD64 3700+ for the "Dates returned period" Aug 28-today. The lowest granted credit/hour are granted for work units which were "long ones", the higher correspond to the "short ones". The calculated "claimed credit" using the boinc benchmark is 13.33 credit/hour (standard Boinc client 5.6.0).
It appears users could "cherry pick" the short ones for 25% higher credit, and still place Einstein at the highest granted credit/hour for any project as seen here (note: the Einstein data shown on the "cross project comparison" is for OLD data using 4.02). This is what it looks like for those running standard clients, those running "third party" boinc core clients should be seeing higher "granted credits" on other projects which involve a quorum, since they'd be paired up with others using them at roughly the same rate as I do, except when they're paired up, they get the higher claims.
RE: RE: But we can argue
)
Hi Bruce
Let's see if I correctly understand this :
Does this mean that you recommand E@H users to stop the project on hosts that would grant more credit/hour on other projects ?
Would this mean that I should keep my Opteron 170 on E@H and move my other boxes (XP 2400+, P-M 1.6, P4B 2.4 and P4C3.2) on, let's say, S@H ?
In my opinion, this is quite a contradiction with this :
Because of you Bruce, better said, thanks to you Bruce, I feel great crunching for E@H. For the opposite reason, I don't feel like crunching again for a project I quit a few months ago. Because of this average stuff that doesn't average anything, I can not choose to attach to the project I like.
I think, there are 3 kinds of users in Boinc :
1/ those mainly concerned by science.
2/ those more concerned by credit than by science.
3/ those equally concerned by credit AND science.
An average stuff about credit/hour is ignored by group 1. It is a concern for groups 2 and 3. Again, if credit wasn't a known concern for a majority of Boinc users (groups 2 and 3), projects would not maintain top users, computers, teams rankings ;) Guys like Willy at Boincstats.com and otherrs would not maintain stats.
It is well known that E@H and S@H are the two best projects for grant credit. Strange, these two also attract (by far) the greatest numbers of users and hosts (figures available here http://www.boincstats.com/index.php).
I understand it is not easy to state : you have to deal with both E@H interests AND the Boinc community interests. The Boinc community is made of projects, but also made of users. So your decision has to meet the interests of E@H, other projects and users. As I said, user interest in credit/hour is an important factor for a majority of us. The average thing is an interest only for projects.
That's the problem.
I would suggest this solution :
- Grant credit "inside" wu should be mandatory for every boinc project (no more third part boinc credit correction)
- Credit/hour/cpu should be "averaged" among plateforms (ie cpu architecture) and not among hosts or projects.
The last point means : a netburst P4 should get an equally close approximate credit/hour/cpu whatever the project. Same thing for A64, XP's, P3 and Core architecture. This way, you avoid the present situation :
Concerned with credit ? => A64 on E@H, the rest on S@H, nothing on other projects.
Not concerned with credit ? => do what you want.
I know nothing in programming, so I don't know if this solution is possible, too complex or what. But I think it could solve the current probem.
Regards,
BHS
To be honest I am not
)
To be honest I am not understanding the negativity about this attempt to equalise redit across the projects.
Historically for over a year the project with the highest credit/hr has been CPDN, but with the optimisation of Einstein application it has shot up the table and has taken the correct steps to decrease it.
Seti has slipped in the credits/hr but there has been an increase in the latest version in the Beta app.
At the moment on my Pent M I am seeing:
Einstein (7 long 112.74 credit units) 17.66/hr
SetiBeta (17 @ ar=0.4265) 16.58/hr
CPDN (sulphur) 16.94/hr
So from a purely personal view on one computer, I would say that Einstein is still awarding about 5% more than other projects.
Andy
I also agree, the Credits
)
I also agree, the Credits Einstein has been granted were quite a bit too high compared to other Projects I know.
Despite having absolutely equal shares here, EAH still easily scores the majority as seen in my Sig.
Not only does it affect us individual crunchers, it also can easily put the whole Project into a wrong light (already, there were discussion about an artificial Credits inflation to attract more users, which would put EAH into discredit which it truly doesn't deserve)
Ideally, it's really best when Users get about the same credit, regardless of which Projects they join.
Has credit been reduced once
)
Has credit been reduced once again (i.e. today)? Or did I get a slightly smaller WU? My Results for user shows the last reported unit claiming only 108.21 (vs. the most recent range of 112-114).
RE: Has credit been reduced
)
Your cpu time should answere your question. ;-)
29,951.19 sec / 112.56 credits (l1_0637.5)
27,123.91 sec / 108.21 credits (l1_0637.5)
These differences even in the same series are quite usual(up to about 5%).
cu,
Michael
RE: Has credit been reduced
)
Yeah, they did it again.
Anyhow 112-114 sounds bad make it a round 100. or 90 of even 50.
This will increase my time to claim the No. one spot in my team any further...
But it's fair to fight credits claimed/granted using the old system vs. the credits i need to get granted now.
Shorten it further ...should keep me from reaching my personal goal another 14 weeks. Or even more ... make it 28 weeks :D
P.S.
@Mr. Castro/Tony post your crap stats somewhere else... and NOT as a reply to my post... or i'll freak out!