New credit system

We are currently experimenting with a new credit system for Docking@Home. This system will grant fixed credits based on server-side flops estimate for a workunit. To stay in line with other projects, we are currently thinking of granting 3.75e-12 credits per flop. In the case of our current target (workunit) this means 56.25 credits are granted for every result finished and validated (15e12*3.75e-12=56.25).

Also we are thinking about adding some extra credit for long term volunteers. For example: if somebody would be actively connected to the project for a 100 days, we would grant 100*0.01=1 credit to all validated results. This number would slowly increase on a daily basis of course :-) In the example above this would mean the volunteer receives a total of 57.25 credits for his/her finished and validated workunit.

For us this seems the fairest way of granting credit for work done, while at the same time keeping the competitive element, because faster computers will be able to complete their results faster and thus collect more credit per hour/day/month/year (yes, buying a faster machine will get you higher in the ranks, even with fixed credit :-)

We would appreciate your comments and feedback on this proposal. But whatever the credit system will look like in the future, always remember that science and curing diseases will come first!

Thanks
The D@H Team
____________
D@H the greatest project in the world... a while from now!

Sounds good to me in the case of Linux based hosts as they will get more appropriate credit than they do with the benchmark system. For Windows however, it may result in you losing participants unless the windows app can be optimized.

Put it this way, most Windows hosts running the stock Berkeley BOINC client currently claim about 70 to 85 credits per work unit because of the amount of CPU time it takes. Reducing their credit without reducing the processing time will mean this project grants the lowest credit of all.

I agree with your proposal, I just think the figure needs a bit more work.

____________


Join the #1 Aussie Alliance on Docking@Home

Agreed. We will also work on optimization of the windows app at the same time to get it more in line with the linux and mac running times. This is probably going to be our biggest challenge though since Charmm was never meant to run on windows. But we'll figure out something, I am sure :-)

AK

Sounds good to me in the case of Linux based hosts as they will get more appropriate credit than they do with the benchmark system. For Windows however, it may result in you losing participants unless the windows app can be optimized.

Put it this way, most Windows hosts running the stock Berkeley BOINC client currently claim about 70 to 85 credits per work unit because of the amount of CPU time it takes. Reducing their credit without reducing the processing time will mean this project grants the lowest credit of all.

I agree with your proposal, I just think the figure needs a bit more work.

Agreed. We will also work on optimization of the windows app at the same time to get it more in line with the linux and mac running times. This is probably going to be our biggest challenge though since Charmm was never meant to run on windows. But we'll figure out something, I am sure :-)

Richard will be working on the diskwriting problem the next two weeks. His analysis will hopefully pinpoint the problem.

Thanks
Andre

Agreed. We will also work on optimization of the windows app at the same time to get it more in line with the linux and mac running times. This is probably going to be our biggest challenge though since Charmm was never meant to run on windows. But we'll figure out something, I am sure :-)

As noted in other threads, the disk writes (debug info) may well be part of the poor performance on Windows. Maybe once that's reduced/turned off, things will improve a lot.

We are currently experimenting with a new credit system for Docking@Home. This system will grant fixed credits based on server-side flops estimate for a workunit. To stay in line with other projects, we are currently thinking of granting 3.75e-12 credits per flop. In the case of our current target (workunit) this means 56.25 credits are granted for every result finished and validated (15e12*3.75e-12=56.25).

Also we are thinking about adding some extra credit for long term volunteers. For example: if somebody would be actively connected to the project for a 100 days, we would grant 100*0.01=1 credit to all validated results. This number would slowly increase on a daily basis of course :-) In the example above this would mean the volunteer receives a total of 57.25 credits for his/her finished and validated workunit.

But whatever the credit system will look like in the future, always remember that science and curing diseases will come first!

Thanks
The D@H Team

We are currently experimenting with a new credit system for Docking@Home. This system will grant fixed credits based on server-side flops estimate for a workunit. To stay in line with other projects, we are currently thinking of granting 3.75e-12 credits per flop. In the case of our current target (workunit) this means 56.25 credits are granted for every result finished and validated (15e12*3.75e-12=56.25).

Also we are thinking about adding some extra credit for long term volunteers. For example: if somebody would be actively connected to the project for a 100 days, we would grant 100*0.01=1 credit to all validated results. This number would slowly increase on a daily basis of course :-) In the example above this would mean the volunteer receives a total of 57.25 credits for his/her finished and validated workunit.

For us this seems the fairest way of granting credit for work done, while at the same time keeping the competitive element, because faster computers will be able to complete their results faster and thus collect more credit per hour/day/month/year (yes, buying a faster machine will get you higher in the ranks, even with fixed credit :-)

We would appreciate your comments and feedback on this proposal. But whatever the credit system will look like in the future, always remember that science and curing diseases will come first!

Thanks
The D@H Team

We will document the results of this experiment as soon as they come available and have been analyzed. As a matter of fact, we have just implemented a new HR class for PII's and III's on the Windows platform. This should get rid of quite some invalid results problems.

Thanks
Andre

Is there a reasonable way to share the results of the WUs beginning at 19724 broken out by host(CPU, OS)?

Thank you.

Hello,

For us this seems the fairest way of granting credit for work done, while at the same time keeping the competitive element, because faster computers will be able to complete their results faster and thus collect more credit per hour/day/month/year (yes, buying a faster machine will get you higher in the ranks, even with fixed credit :-)

But whatever the credit system will look like in the future, always remember that science and curing diseases will come first!

That credit system sounds good to me too, but as Yoda said it only works in line with optimisations for windows. Good luck with that!

I like the idea of the extra credit too :)

Managed to get the windows running time down from 8 to 4.5 hours (3.5 on the same linux box). Not bad :-) We're getting there...

Thanks!
Andre

That credit system sounds good to me too, but as Yoda said it only works in line with optimisations for windows. Good luck with that!

I like the idea of the extra credit too :)

Managed to get the windows running time down from 8 to 4.5 hours (3.5 on the same linux box). Not bad :-) We're getting there...

Thanks!
Andre

That credit system sounds good to me too, but as Yoda said it only works in line with optimisations for windows. Good luck with that!

I like the idea of the extra credit too :)

Managed to get the windows running time down from 8 to 4.5 hours (3.5 on the same linux box). Not bad :-) We're getting there...

Thanks!
Andre

No changes to the code; just optimization parameters of the compiler. That's not as easy as it sounds though, because too much optimization will introduce more divergencies. Have it set to O2 and included some floating point precision stuff now and tested on all the platforms that we have in the lab; up to know things look good. We have to shave of another hour or so though to get it in line with Linux/Mac workunits.

Michela/Richard are working on the disk write problem and hope to have something ready in 2 weeks or so.

Thanks
Andre

Managed to get the windows running time down from 8 to 4.5 hours (3.5 on the same linux box). Not bad :-) We're getting there...

Thanks!
Andre

Hi Andre,

I was just wondering if this was with an AMD or Intel CPU? Were the changes all to the code writing the debug file? Async writes, maybe?

Thanks,

-- David

If you go to fixed credits i might have to bring my Core 2 Duo here to crunch for a bit. The credits before were so bad i went and crunched RCN lol. if anyone knows RCN the credits there are quorum of 1 so on a Linux box absolute garbage.
____________

If you go to fixed credits i might have to bring my Core 2 Duo here to crunch for a bit. The credits before were so bad i went and crunched RCN lol. if anyone knows RCN the credits there are quorum of 1 so on a Linux box absolute garbage.

Any hints as to when this might come in? I run all Linux machines (plus 1 part time Windows box). I have a Core 2 Duo 2.13Ghz and a Celery 2Ghz that will be taking work on 20 seconds after the credits are fixed. I like the science here but the competitive side of me doesn't like the fact the credits are so poor. Fix that and i have an almost ideal project.
____________

We'll implement the fixed credits as soon as we have the crunching times on the several platforms more in line with each other. This shouldn't be too long from now; maybe two weeks.

Thanks,
Andre

Any hints as to when this might come in? I run all Linux machines (plus 1 part time Windows box). I have a Core 2 Duo 2.13Ghz and a Celery 2Ghz that will be taking work on 20 seconds after the credits are fixed. I like the science here but the competitive side of me doesn't like the fact the credits are so poor. Fix that and i have an almost ideal project.

> Points granted between Linux and Windows (don't know about Mac) currently favour Windows by quite a bit, even with the longer processing times.
I did a little experiment to see what the difference will be with the current WU times and the new point system (yes I know, some people say I have too much time on my hands).
Using 11 results spread over about a month to see if any processing times have changed (I have used 11 as 1 computer only has 11 successful results).

Intel P4 2.53 @2.75 GHz running Windows.
AMD X2 4800+ (core speed 2.4 GHz) running Windows.
AMD Opteron 275 (core speed 2.2 GHz) running Linux.
AMD Opteron 285 (core speed 2.6 GHz) running Linux.

>> Intel P4 WUs from 22/11/06 to 1/1/07
Ave processing time = 31,767.91 seconds (8.82 hours)
Ave points awarded = 83.11
Total points awarded for 11 results = 914.22
Ave Points/hour = 9.42

>> AMD X2 4800+ WUs from 6/12/06 to 5/1/07
Ave processing time = 21,301.70 seconds (5.92 hours)
Ave points awarded = 81.78
Total points awarded for 11 results = 899.61
Ave points/hour = 13.83

>> AMD Opteron 275 WUs from 19/12/06 to 9/1/07
Ave processing time = 10,283.96 seconds (2.86 hours)
Ave points awarded = 28.14
Total points awarded for 11 results = 309.51
Ave points/hour = 9.84

>> AMD Opteron 285 WUs from 13/12/06 to 9/1/07
Ave processing time = 8,692.57 seconds (2.42 hours)
Ave points awarded = 27.55
Total points awarded for 11 results = 303.07
Ave points/hour = 11.38

Now if the new system (without any bonus points) is applied then
Ave points awarded = 56.25 (for all computers)
Total points awarded for any 11 results = 56.25x11 = 618.75 (for all computers)
Ave points/hour for Intel P4 2.53 = 56.25 points/8.83 hours = 6.37 (from 9.42)
Ave points/hour for AMD X2 4800+ = 56.25 points/5.92 hours = 9.50 (from 13.83)
Ave points/hour for Opteron 275 = 56.25 points/2.86 hours = 19.67 (from 9.84)
Ave points/hour for Opteron 285 = 56.25 points/2.42 hours = 23.24 (from 11.38)

So from this Linux credit will about double, 303 to 618 points (about time) and Windows credit will drop by a third 900 to 618 points.
Credit per hour drops 32% on the Intel P4
Credit per hour drops 31% on the AMD 4800+
Credit per hour increases 100% on the Opteron 275
Credit per hour increases 104% on the Opteron 285

So I am in favour of the nem system (makes it about equal to Einstein for credit per hour) but Windows needs to be sorted out (I don't know what chance you have as Bill Gates has tried for over 10 years and he still can't sort Windows out).

But you can see that if people are only doing a small number of Linux jobs and the same number of Windows jobs they will get more points with Windows even though it takes longer to do a job.

Andre I have not noticed any difference in the processing times of my Windows work units since November to now. Linux also has stayed the same.

I know this has been long winded but it shows to you 4 different computers and 2 operating systems and shows the differences between them.

P.S. Sorry if my maths is out a bit but I think it is right.
____________

Hello,

So I am in favour of the nem system (makes it about equal to Einstein for credit per hour) but Windows needs to be sorted out (I don't know what chance you have as Bill Gates has tried for over 10 years and he still can't sort Windows out).

But you can see that if people are only doing a small number of Linux jobs and the same number of Windows jobs they will get more points with Windows even though it takes longer to do a job.

Just out of interest Conan what version of BOINC are you using? Thats very good credit and speed you get out of your Linux systems. Congratz on the speed you get though my C2D is jealous lol.
I find a WU on my Kubuntu 6.10 Core 2 Duo E6400 takes just under 2hrs 50mins and get awarded 16-20 credits (although ive got as high as 25 and as low as 11). Im lucky to get 6-6.5 cr/hr/core on a machine that regularly gets 19cr/hr/core on CPDN and Einstein and around 24 cr/hr/core on Seti optimized. I think i even get better credit on Malaria control as i quorum against windoze machines which really drags my granted credit up.

____________

Thanks for this detailed analysis Conan; I'm glad that people have a lot of time on their hands. We are very close to getting the windows crunching time similar to the linux crunching time on the same host, which will address your concern. If testing goes without problems, we might possibly release the new windows app today. Then we'll closely observe the new situation for a while and in case times are similar for most platform types, we'll go to fixed credits.

Cheers
AK

> Points granted between Linux and Windows (don't know about Mac) currently favour Windows by quite a bit, even with the longer processing times.
I did a little experiment to see what the difference will be with the current WU times and the new point system (yes I know, some people say I have too much time on my hands).
Using 11 results spread over about a month to see if any processing times have changed (I have used 11 as 1 computer only has 11 successful results).

Intel P4 2.53 @2.75 GHz running Windows.
AMD X2 4800+ (core speed 2.4 GHz) running Windows.
AMD Opteron 275 (core speed 2.2 GHz) running Linux.
AMD Opteron 285 (core speed 2.6 GHz) running Linux.

>> Intel P4 WUs from 22/11/06 to 1/1/07
Ave processing time = 31,767.91 seconds (8.82 hours)
Ave points awarded = 83.11
Total points awarded for 11 results = 914.22
Ave Points/hour = 9.42

>> AMD X2 4800+ WUs from 6/12/06 to 5/1/07
Ave processing time = 21,301.70 seconds (5.92 hours)
Ave points awarded = 81.78
Total points awarded for 11 results = 899.61
Ave points/hour = 13.83

>> AMD Opteron 275 WUs from 19/12/06 to 9/1/07
Ave processing time = 10,283.96 seconds (2.86 hours)
Ave points awarded = 28.14
Total points awarded for 11 results = 309.51
Ave points/hour = 9.84

>> AMD Opteron 285 WUs from 13/12/06 to 9/1/07
Ave processing time = 8,692.57 seconds (2.42 hours)
Ave points awarded = 27.55
Total points awarded for 11 results = 303.07
Ave points/hour = 11.38

Now if the new system (without any bonus points) is applied then
Ave points awarded = 56.25 (for all computers)
Total points awarded for any 11 results = 56.25x11 = 618.75 (for all computers)
Ave points/hour for Intel P4 2.53 = 56.25 points/8.83 hours = 6.37 (from 9.42)
Ave points/hour for AMD X2 4800+ = 56.25 points/5.92 hours = 9.50 (from 13.83)
Ave points/hour for Opteron 275 = 56.25 points/2.86 hours = 19.67 (from 9.84)
Ave points/hour for Opteron 285 = 56.25 points/2.42 hours = 23.24 (from 11.38)

So from this Linux credit will about double, 303 to 618 points (about time) and Windows credit will drop by a third 900 to 618 points.
Credit per hour drops 32% on the Intel P4
Credit per hour drops 31% on the AMD 4800+
Credit per hour increases 100% on the Opteron 275
Credit per hour increases 104% on the Opteron 285

So I am in favour of the nem system (makes it about equal to Einstein for credit per hour) but Windows needs to be sorted out (I don't know what chance you have as Bill Gates has tried for over 10 years and he still can't sort Windows out).

But you can see that if people are only doing a small number of Linux jobs and the same number of Windows jobs they will get more points with Windows even though it takes longer to do a job.

Andre I have not noticed any difference in the processing times of my Windows work units since November to now. Linux also has stayed the same.

I know this has been long winded but it shows to you 4 different computers and 2 operating systems and shows the differences between them.

P.S. Sorry if my maths is out a bit but I think it is right.

I was about to say the same thing...

Charmm 5.4 for windows has been released. Let's see how this works out on platforms we can't test in the lab. Maybe Conan can perform another one of his analyses in a couple of days :-)

Thanks
Andre
____________
D@H the greatest project in the world... a while from now!

Just out of interest Conan what version of BOINC are you using? Thats very good credit and speed you get out of your Linux systems. Congratz on the speed you get though my C2D is jealous lol.
I find a WU on my Kubuntu 6.10 Core 2 Duo E6400 takes just under 2hrs 50mins and get awarded 16-20 credits (although ive got as high as 25 and as low as 11). Im lucky to get 6-6.5 cr/hr/core on a machine that regularly gets 19cr/hr/core on CPDN and Einstein and around 24 cr/hr/core on Seti optimized. I think i even get better credit on Malaria control as i quorum against windoze machines which really drags my granted credit up.

Hello,

> Hello Clownius,
I am using Boinc version 5.2.13 on the Linux machines since about April '06.

Fair enough. My computers often claimed 10-15 credits per WU but i am using 5.4.11 which seems to be one of the worst Linux benchmarks ever lol. Hopefully once i get my next computer i can run this one headless and reduce my run times but at the moment it is my desktop so its doing everything lol.
What if anything do you do to optimize your Linux boxes? Custom kernels or anything/overclock? I want to try and get my machines running as fast as comparable windoze machines before to long.
Also a huge congratulations for single handedly holding off BOINC@AUSTRALIA for so long we just cant quite get past you. An outstanding performance.
____________

Linux BOINC 5.8.1 has better benchmarks on my P4/3.4 if I run with one thread - only about 5% lower than Windows. But If I run with HT enabled and using both virtual cores it's still awful. I haven't tested it on any other computers as this is the only Linux box I've got, but my guess is that at least on single core processors it would be close to Windows.

And yes, congratulations for holding us off for so long. Imagine where we'd be if we were a combined Aussie team (hmm, we'd probably be ranked 4th!)

It could be that AMD chips have better Floating Point ability but I am unsure if Docking uses Floating Point or Integer.

We rely on almost only floating point operations (let's say 99.5%). The app doesn't use any cpu-specific features (like for example SSE2/3) though. At least not yet; we would only be able to optimize per platform if we go fully HR which we might do later on.

We rely on almost only floating point operations (let's say 99.5%). The app doesn't use any cpu-specific features (like for example SSE2/3) though. At least not yet; we would only be able to optimize per platform if we go fully HR which we might do later on.

HR?

> Thanks Clownius and Webmaster Yoda, BOINC@AUSTRALIA is certainly pushing me and making me work hard (I am sure Docking@home does not mind in the least), If there were more members in my team....ah now there's a thought, web page, members???

Clownius my computers are all standard bar my Intel which runs Windows (it is overclocked from 2.53 to 2.75 GHz, fastest speed but slowest computer), my ASUS overclocking motherboard on the 4800+ machine wont overclock it reboots all the time (took a year to work that out) and now runs perfectly at standard speeds. The Linux machines were built by me but I have done nothing else with them (I have done bugger all with them since April bar an attempt to rerun a CPDN model on the Opteron 275 machine a couple of months ago which stuffed everything, I did it wrong apparently, and a reinstall was required duplicating this host in 4 projects including Docking).
Don't really have an idea how compile anything let alone a Kernel.

Andre confirmed that Floating Point calculations are done in Docking which I have read favour the AMD chips particuarly Server types like my Opterons.

Just checked whilst writing this and the new 5.04 Docking version only took just 12,050.06 seconds (3h 20m 50s) on my Intel P4 2.53 machine compared to 31,747.91 seconds (8h 50m) on version 5.03. That is really impressive.
With the proposed 56.25 cobblestones for each WU = 56.25/3.34 = 16.84 c/h
(Compared to 9.42 average under the old system).
____________

Finally one of my hosts, which has Pentium M (Dothan) and 1GB RAM, completed crunching several workunits. Actually the time when it needed to crunch a task became almost half (from 8.23 hour to 3.93), and it became relatively close to one which linux through VMWare takes in crunching a task.

On Charmm 5.3, credits/hour is 8.68, and on Charmm 5.4 it will be 56.25/3.93 = 14.31. This seems to me higher than expected, isn't it.

suguruhirahara
____________

I'm a volunteer participant; my views are not necessarily those of Docking@Home or its participating institutions.

Hi Suguru,

The 56.25 is not set in stone yet. We are still experimenting with different numbers and methods. In the end we'd like to be as much in line with what other projects grant (anybody having any data on credit/h lying around?).

Thanks
Andre

Finally one of my hosts, which has Pentium M (Dothan) and 1GB RAM, completed crunching several workunits. Actually the time when it needed to crunch a task became almost half (from 8.23 hour to 3.93), and it became relatively close to one which linux through VMWare takes in crunching a task.

On Charmm 5.3, credits/hour is 8.68, and on Charmm 5.4 it will be 56.25/3.93 = 14.31. This seems to me higher than expected, isn't it.

suguruhirahara

Somebody posted an exellent comparison of the credits/hour of nearly all the project, but sadly the links to the charts expired.

That's why I planned to do one based on an average of the last few WUs of projects I recently contributed in.

I'm using an athlon 64 X2 4600+ running on windows xp sp2.

I'll post it as soon as I complete it.

____________

AMD Athlon 64 X2 4600+ Windows XP SP2


this picture won't expire (unless I delete it from my server)
____________

According to this table, our possible future 14.31 credits/h doesn't look so crazy :-)

Thanks a lot Tom!

Andre

AMD Athlon 64 X2 4600+ Windows XP SP2


this picture won't expire (unless I delete it from my server)

According to this table, our possible future 14.31 credits/h doesn't look so crazy :-)

> This is a rough analysis of my projects since November, it shows that credit does vary even on projects with fixed credits. The faster the machine the more credit per hour awarded. You can see that I seem to have a credit drop or a rise in credits depending on project or host.
> What computer are you using to base your credit per hour of 14.31 on?

The below credit per hour correlates with granted credit in a Quorum or standardised credit system.

Project____| Computer____________|Date_______|Cred/Hr_______|Date_______|Cred/Hr
Rosetta____| Intel 2.40GHz Cel_____|08/11/06___| 7.37_________|02/01/07___| 6.85
Rosetta____| Intel 2.53GHz P4______|08/11/06___| 9.04_________|02/01/07___| 8.58
Rosetta____| AMD 4800+ 2.4GHz___|08/11/06___|14.43_________|02/01/07___|14.13
Rosetta____| Opteron 275 2.2GHz___|08/11/06___|14.81_________|02/01/07___|15.00
Rosetta____| Opteron 285 2.6GHz___|08/11/06___|15.43_________|02/01/07___|16.09
Docking____| Intel 2.53GHz P4______|09/01/07___|9.42 (v5.03)____|13/01/07___| 9.18 (v5.04)
Docking____| AMD 4800+ 2.4GHz___|08/11/06___|______________|09/01/07___|13.83 (v5.03)
Docking____| Opteron 275 2.2GHz___|08/11/06___|11.09_________|09/01/07___| 9.84
Docking____| Opteron 285 2.6GHz___|08/11/06___|10.42_________|09/01/07___|11.38
Einstein____| Intel 2.53GHz P4______|08/11/06___|3.5(long wu)____|13/01/07___|14.81 (short wu)
Einstein____| AMD 4800+ 2.4GHz___|08/11/06___|20.88_________|13/01/07___|21.68 (short wu)
Einstein____| Opteron 275 2.2GHz___|08/11/06___|19.55_________|13/01/07___|20.43 (short wu)
Einstein____| Opteron 285 2.6GHz___|08/11/06___|22.86_________|13/01/07___|23.96 (short wu)
CPDN_____| Opteron 275 2.2GHz___|08/11/06___|17.82_________|13/01/07___|17.64
CPDN_____| Opteron 285 2.6GHz___|08/11/06___|______________|13/01/07___|18.43
QMC______| AMD 4800+ 2.4GHz___|08/11/06___|13.74_________|___________|
QMC______| Opteron 275 2.2GHz___|08/11/06___|18.26_________|___________|
QMC______| Opteron 285 2.6GHz___|08/11/06___|21.55_________|___________|
LHC______| Opteron 285 2.6GHz____|08/11/06___|10.22_________|13/01/07___|13.34
Ralph_____| AMD 4800+ 2.4GHz____|08/11/06___|14.37_________|13/01/07___|17.53
Ralph_____| Opteron 275 2.2GHz____|08/11/06___|15.38_________|13/01/07___|25.39
Ralph_____| Opteron 285 2.6GHz____|08/11/06___|15.85_________|13/01/07___|25.20
Seti_______| Intel 2.53GHz P4______|08/11/06___|12.47_________|13/01/07___|11.77
Seti______| AMD 4800+ 2.4GHz______|08/11/06___|10.05_________|___________|

____________

With 56.25 credits/WU and the runtime/WU (windows 5.04) wouldn't change anymore, we would have on my 4600+ 24.37 credits/h.

That is, even though we would attract many stats addicted people, far too much.

22.11 is the average of the projects granting the highest amount of credits/h (Einstein, SIMAP and TANPAKU).

Converting this into credits/WU on Docking@home would be 51.04 credits/WU. In my opinion a value around this would be best. It's in line with other projects and the moderately high amount of credit granted attracts many credit hunters.

____________

Converting this into credits/WU on Docking@home would be 51.04 credits/WU. In my opinion a value around this would be best. It's in line with other projects and the moderately high amount of credit granted attracts many credit hunters.

> More Statistics.
Follow up after 5.03 finished and 5.04 started.

DATE____________AVE TIME(H)__ AVE POINTS__AVE POINTS/HOUR
............................. GRANTED/WU
Intel P4 2.53 @2.75 GHz
Up to 28/12/06--8.81 hours----89.00-------10.10 v5.03 (ave 5 wu's)
28/12...9/1/07--8.83 hours----83.11-------9.88 v5.03 (ave 11 wu's)
10/1...20/1/07--3.33 hours----30.94-------9.28 v5.04 (ave 11 wu's)

AMD X2 4800+
Up to 28/12/06--6.24 hours----82.28-------13.18 v5.03
28/12...9/1/07--5.92 hours----81.78-------13.83 v5.03
10/1...20/1/07--2.38 hours----33.44-------14.06 v5.04

AMD Opteron 275
Up to 28/12/06--2.84 hours----27.43-------9.67
28/12...9/1/07--2.86 hours----28.14-------9.84
10/1...20/1/07--2.84 hours----28.25-------9.95

AMD Opteron 285
Up to 28/12/06--2.41 hours----32.38-------13.43
28/12...9/1/07--2.42 hours----27.55-------11.38
10/1...20/1/07--2.41 hours----28.64-------11.88

If you use 1st proposal from Docking of 56.25 per WU then

P4 2.53 computer gives 56.25 per wu / 3.33 H = 16.89 cr/h (E@H=14.81)
AMD 4800+ comp gives 56.25 per wu / 2.38 H = 23.64 cr/h (E@H=21.68)
AMD Opteron 275 gives 56.25 per wu / 2.84 H = 19.81 cr/h (E@H=20.43)
AMD Opteron 285 gives 56.25 per wu / 2.41 H = 23.34 cr/h (E@H=23.96)

With Tom's suggestion of 22.11 per Hour or 51.04 per WU for a 4600+ AMD computer (same speed as my 4800+ just a smaller level 2 cache, I think Tom's is faster than mine, as I don't get 22.11)

P4 2.53 computer gives 51.04 per wu / 3.33 H = 15.33 cr/h
AMD 4800+ comp gives 51.04 per wu / 2.38 H = 21.45 cr/h
AMD Opteron 275 gives 51.04 per wu / 2.84 H = 17.97 cr/h
AMD Opteron 285 gives 51.04 per wu / 2.41 H = 21.18 cr/h

With Webmaster Yoda's 40 to 45 credits per WU

P4 2.53 computer gives 40.00 per wu / 3.33 H = 12.01 cr/h
---------------------- 45.00 per wu / 3.33 H = 13.51 cr/h
AMD 4800+ comp gives 40.00 per wu / 2.38 H = 16.81 cr/h
---------------------- 45.00 per wu / 2.38 H = 18.91 cr/h
AMD Opteron 275 gives 40.00 per wu / 2.84 H = 14.09 cr/h
---------------------- 45.00 per wu / 2.84 H = 15.85 cr/h
AMD Opteron 285 gives 40.00 per wu / 2.41 H = 16.60 cr/h
---------------------- 45.00 per wu / 2.41 H = 18.67 cr/h

> Getting a reasonable match to the higher crediting projects I think I prefer about 50.00 points per WU, giving 15.02 points P4 2.53, 21.00 points AMD 4800+, 17.61 points Opteron 275 and 20.75 points Opteron 285.
This is close to, but less than, what Einstein gives.

The value given should also be based on the science that is done as I don't believe just giving any old cobblestone value gives credit to the amount work done by individual computers.


____________

Of course the credit will be different for every workunit we'll distribute; More complex protein-ligand combinations will get more credit (based on the est. FLOPS of a wu).

Thanks for the analysis!
Andre

The value given should also be based on the science that is done as I don't believe just giving any old cobblestone value gives credit to the amount work done by individual computers.

With Tom's suggestion of 22.11 per Hour or 51.04 per WU for a 4600+ AMD computer (same speed as my 4800+ just a smaller level 2 cache, I think Tom's is faster than mine, as I don't get 22.11)

Starting with workunit 25683 we are using a fixed credit amount. For the current test workunit (1tng) this is 49.5. Please let us know how much credit/hour this comes out to on your machine. See the news for more info.

Thanks
The D@H Team
____________
D@H the greatest project in the world... a while from now!

Starting with workunit 25683 we are using a fixed credit amount. For the current test workunit (1tng) this is 49.5. Please let us know how much credit/hour this comes out to on your machine. See the news for more info.

Thanks
The D@H Team

This is something Memo can work on as part of the shared memory script he is building to show the HR classes ready to be distributed.

Thanks for the idea!
Andre

Starting with workunit 25683 we are using a fixed credit amount. For the current test workunit (1tng) this is 49.5. Please let us know how much credit/hour this comes out to on your machine. See the news for more info.

Thanks
The D@H Team

Can you add the low and high result numbers that are "ready to send" on the Server Status Page to the right of the "results ready to send" number?

Thank you for brining in fixed credits :). Expect some Core 2 Duo love from Feb 15th when AA5 is over and i move my computers off ABC a bit. I may even throw a little Celeron or two into the mix as well but methinks the poor 700 may struggle but the 2Ghz should do OK.
____________

Starting with workunit 25683 we are using a fixed credit amount. For the current test workunit (1tng) this is 49.5. Please let us know how much credit/hour this comes out to on your machine.

Thanks for collecting all the great data!

Btw, I'll be offline for the next weeks. See you later!

Alexander thanks for the analysis. I think that it looks good as only the really old machines will be affected but those are very rare this days specially crunching for a project.

Does anyone has any comments with the gain/loss of credit per hour?

Alexander thanks for the analysis

Does anyone has any comments with the gain/loss of credit per hour?

Alexander thanks for the analysis. I think that it looks good as only the really old machines will be affected but those are very rare this days specially crunching for a project.

Does anyone has any comments with the gain/loss of credit per hour?

Starting with workunit 25683 we are using a fixed credit amount. For the current test workunit (1tng) this is 49.5. Please let us know how much credit/hour this comes out to on your machine. See the news for more info.

Thanks
The D@H Team

I updated my small comparison of the projects with my A64 X2 4600+ :

____________

<snip>... looks good and quite reasonable to me:

<snip>...
Athlon XP 2,0 GHz / Linux w/ BOINC 5.4.x (8 results):

average time per workunit: 3 hrs 39 min
average claimed credits: 6.46 per hour
average granted credits: 7.24 per hour
claimed credits w/ new fixed credit scheme would be: 13.54 per hour ( + 110 % )

Athlon XP 2,0 GHz / Linux w/ BOINC 5.8.x (57 results):

average time per workunit: 3 hrs 40 min
average claimed credits: 9.02 per hour
average granted credits: 8.66 per hour
claimed credits w/ new fixed credit scheme would be: 13.50 per hour ( + 50 % )
<snip>...

Please let us know how much credit/hour this comes out to on your machine. See the news for more info.

I guess that everybody is getting more credit, at the same time it seems that we are at about the same as the other projects.

Now I am interested in computers that have no big gain, no gain or even loss of credits/hr

I guess very old machines will suffer from this but I wonder how old a machine needs to be in order to lose credit. Anyone with old machines?

I guess that everybody is getting more credit, at the same time it seems that we are at about the same as the other projects.

Now I am interested in computers that have no big gain, no gain or even loss of credits/hr

I guess very old machines will suffer from this but I wonder how old a machine needs to be in order to lose credit. Anyone with old machines?

we are at about the same as the other projects.

The other projects are sending a variety of wu's so the standard deviations of their results are much higher then Docking. In some cases wildly more.

Understand totally, I was simply pointing out that although crunched on the same machine, the statistical significance of the results is not as black and white as it may first appear.

Just working it out in my head, the small SIMAP sample had an average claim of 8.73, but an SD of over 5!

@Memo, My oldest machine on this project is an Intel P4 2.53 @2.66 GHz

Intel P4 2.53 @ 2.66 GHz Windows XP

Docking Ave = 9.27 cr/h (old system)
Docking Ave = 14.25 cr/h (new system) +53.72%
Rosetta Ave = 8.29 cr/h
Einstein Ave = 13.73 cr/h

AMD 4800+ Windows XP

Docking Ave = 14.65 cr/h (old system)
Docking Ave = 20.76 cr/h (new system) +41.71%
Rosetta Ave = 13.96 cr/h
Ralph Ave = 14.74 cr/h
Einstein Ave = 23.09 cr/h
QMC Ave = 26.44 cr/h (New Fixed credit system)

AMD Opteron 275 Linux Fedora Core 3

Docking Ave = 11.73 cr/h (old system)
Docking Ave = 17.43 cr/h (new system) +48.59%
Rosetta Ave = 14.30 cr/h
Ralph Ave = 16.51 cr/h
Einstein Ave = 19.75 cr/h
QMC Ave = 25.34 cr/h (New Fixed credit system)
CPDN Ave = 17.76 cr/h
Predictor Ave= 10.30 cr/h

AMD Opteron 285 Linux Fedora Core 3

Docking Ave = 12.50 cr/h (old system)
Docking Ave = 20.54 cr/h (new system) +64.32%
Rosetta Ave = 17.01 cr/h
Ralph Ave = 19.10 cr/h
Einstein Ave = 23.21 cr/h
QMC Ave = 24.61 cr/h (New Fixed credit system)
CPDN Ave = 18.31 cr/h
LHC Ave = 15.39 cr/h

Going on these results I find the new Docking system to be very good and fair, on a par with most projects.
I do not possess a slower machine to see if the credits go negative or only very small increases.
It would be nice if Predictor upped their cobblestone reward system to match others, it is now my most miserly project.

____________

Thanks for the detailed data Conan. This shows us that we are probably doing the right thing going to fixed credits. The next thing we will work on is assign extra varariable credit based on how long the volunteer has been supporting D@H. That will give some new twist to credit collecting :-)

Andre

@Memo, My oldest machine on this project is an Intel P4 2.53 @2.66 GHz

Intel P4 2.53 @ 2.66 GHz Windows XP

Docking Ave = 9.27 cr/h (old system)
Docking Ave = 14.25 cr/h (new system) +53.72%
Rosetta Ave = 8.29 cr/h
Einstein Ave = 13.73 cr/h

AMD 4800+ Windows XP

Docking Ave = 14.65 cr/h (old system)
Docking Ave = 20.76 cr/h (new system) +41.71%
Rosetta Ave = 13.96 cr/h
Ralph Ave = 14.74 cr/h
Einstein Ave = 23.09 cr/h
QMC Ave = 26.44 cr/h (New Fixed credit system)

AMD Opteron 275 Linux Fedora Core 3

Docking Ave = 11.73 cr/h (old system)
Docking Ave = 17.43 cr/h (new system) +48.59%
Rosetta Ave = 14.30 cr/h
Ralph Ave = 16.51 cr/h
Einstein Ave = 19.75 cr/h
QMC Ave = 25.34 cr/h (New Fixed credit system)
CPDN Ave = 17.76 cr/h
Predictor Ave= 10.30 cr/h

AMD Opteron 285 Linux Fedora Core 3

Docking Ave = 12.50 cr/h (old system)
Docking Ave = 20.54 cr/h (new system) +64.32%
Rosetta Ave = 17.01 cr/h
Ralph Ave = 19.10 cr/h
Einstein Ave = 23.21 cr/h
QMC Ave = 24.61 cr/h (New Fixed credit system)
CPDN Ave = 18.31 cr/h
LHC Ave = 15.39 cr/h

Going on these results I find the new Docking system to be very good and fair, on a par with most projects.
I do not possess a slower machine to see if the credits go negative or only very small increases.
It would be nice if Predictor upped their cobblestone reward system to match others, it is now my most miserly project.

Thanks for the detailed data Conan. This shows us that we are probably doing the right thing going to fixed credits. The next thing we will work on is assign extra varariable credit based on how long the volunteer has been supporting D@H. That will give some new twist to credit collecting :-)

Andre

@Memo, My oldest machine on this project is an Intel P4 2.53 @2.66 GHz

Intel P4 2.53 @ 2.66 GHz Windows XP

Docking Ave = 9.27 cr/h (old system)
Docking Ave = 14.25 cr/h (new system) +53.72%
Rosetta Ave = 8.29 cr/h
Einstein Ave = 13.73 cr/h

AMD 4800+ Windows XP

Docking Ave = 14.65 cr/h (old system)
Docking Ave = 20.76 cr/h (new system) +41.71%
Rosetta Ave = 13.96 cr/h
Ralph Ave = 14.74 cr/h
Einstein Ave = 23.09 cr/h
QMC Ave = 26.44 cr/h (New Fixed credit system)

AMD Opteron 275 Linux Fedora Core 3

Docking Ave = 11.73 cr/h (old system)
Docking Ave = 17.43 cr/h (new system) +48.59%
Rosetta Ave = 14.30 cr/h
Ralph Ave = 16.51 cr/h
Einstein Ave = 19.75 cr/h
QMC Ave = 25.34 cr/h (New Fixed credit system)
CPDN Ave = 17.76 cr/h
Predictor Ave= 10.30 cr/h

AMD Opteron 285 Linux Fedora Core 3

Docking Ave = 12.50 cr/h (old system)
Docking Ave = 20.54 cr/h (new system) +64.32%
Rosetta Ave = 17.01 cr/h
Ralph Ave = 19.10 cr/h
Einstein Ave = 23.21 cr/h
QMC Ave = 24.61 cr/h (New Fixed credit system)
CPDN Ave = 18.31 cr/h
LHC Ave = 15.39 cr/h

Going on these results I find the new Docking system to be very good and fair, on a par with most projects.
I do not possess a slower machine to see if the credits go negative or only very small increases.
It would be nice if Predictor upped their cobblestone reward system to match others, it is now my most miserly project.

Frequent Cruncher Credits.....