In a few weeks when we start doing analysis of LIGO/GEO S5 data, we plan to start widely distributing optimized applications containing Akos's modifications. We will also begin to award credit in a more uniform way: the credit awarded for 'the same length' workunits will be predetermined by the project and will be the same for everyone.
I'm sure it will some solve the credit issue if crunch time among different platforms will be similar. We went that way on CPDN - just couple of types of climate models (phew, not a simple thing) with fixed credit per model.
No need of bechamarks, cool :-)
I see that project will benefit widely from cooperation with akosf...very good.
We will also begin to award credit in a more uniform way: the credit awarded for 'the same length' workunits will be predetermined by the project and will be the same for everyone.
Good. The premise that point-of-use information should be part of the credit calculation makes some sense in the SETI case where some work units running on properly running machines essentially "error out" in seconds. As apparently the project can't figure this out in advance, only point of use information can adjust.
But for Einstein the consistency of behavior of the WU's seems to be very high, so having the project pre-calculate the relative effort for each WU and simply award pre-determined credit for success seems excellent to me.
One fine thing is that it preserves the incentive for any and all forms of optimization. Do more useful work, get more credit. No overclaiming, no underclaiming. No harm to quorum partners.
I like this much better than application flop counting!
In a few weeks when we start doing analysis of LIGO/GEO S5 data, we plan to start widely distributing optimized applications containing Akos's modifications. We will also begin to award credit in a more uniform way: the credit awarded for 'the same length' workunits will be predetermined by the project and will be the same for everyone.
This system should be fairer for everyone.
Cheers,
Bruce
Sounds like the PowerPC platform is at the losing end.... but what the heck, it's just science....
Sounds like the PowerPC platform is at the losing end.... but what the heck, it's just science....
Wow, Is it typical for PowerPC's to claim so much more credit than other platforms (about 3 times as much) for the same work? Must be a bug in BOINC?
I don't know what you mean with "at the losing end" though - the quorum system usually sorts out that bug anyway so you get similar credit as everybody else.
In a few weeks when we start doing analysis of LIGO/GEO S5 data, we plan to start widely distributing optimized applications containing Akos's modifications. We will also begin to award credit in a more uniform way: the credit awarded for 'the same length' workunits will be predetermined by the project and will be the same for everyone.
This system should be fairer for everyone.
Cheers,
Bruce
Excellent news. :)
Since Einstein@home doesn't seem to have all the wu that for various reasons ends early like some other projects is having, using a pre-determined credit for each wu-size, similar to Folding@home and CPDN, is a good idea.
The biggest "problem" will be the cross-project-calibration, so Einstein@home gives roughly equal granted credit as the other BOINC-projects...
... And, handling all the complaints from users running optimized S4-applications that suddenly doesn't get 4x or so more credit/hour than other users...
BTW, now you don't need 3 results to make granted credit "fair" any longer, will this also mean can decrease min_quorum to 2 and get a further 50% speed-up, or is the results so variable that still needs 3 to make sure validated scientific results?
Also, what will the wu-size be under S5, similar range as now, or longer now you've got the large speed-up from optimizations?
"I make so many mistakes. But then just think of all the mistakes I don't make, although I might."
Sounds like the PowerPC platform is at the losing end.... but what the heck, it's just science....
Wow, Is it typical for PowerPC's to claim so much more credit than other platforms (about 3 times as much) for the same work? Must be a bug in BOINC?
I don't know what you mean with "at the losing end" though - the quorum system usually sorts out that bug anyway so you get similar credit as everybody else.
Don't make a fool out of yourself. And mind you, I'm utilising a superbenched version of BOINC from Team MacNN for SETI@home....
And when I say "at the losing end", I don't mean the deeply-troubled quorum system based on similar credits, but to the upcoming centralised credit/WU system (whatever they call it) on a dissimilar CPU crunch time....
I observed that a PowerPC was claiming a lot more credit for each work unit than most other hosts. How does that make me a fool? If there's something important I missed, perhaps you could explain that, rather than resorting to making nasty comments.
As I read it everybody will (with the new system Bruce referred to) get the same credit for the same job, regardless of hardware or software used. Fast computers will do more work per day than slow ones and thus get more credit per day. So I don't understand the distinction about the PowerPC being at the "losing end" any more than any other computer under the new system.
But don't bother explaining it - I've lost interest.
I observed that a PowerPC was claiming a lot more credit for each work unit than most other hosts. How does that make me a fool? If there's something important I missed, perhaps you could explain that, rather than resorting to making nasty comments.
As I read it everybody will (with the new system Bruce referred to) get the same credit for the same job, regardless of hardware or software used. Fast computers will do more work per day than slow ones and thus get more credit per day. So I don't understand the distinction about the PowerPC being at the "losing end" any more than any other computer under the new system.
But don't bother explaining it - I've lost interest.
What the point of me explaining anyway since you're not a PowerPC user anyway....? It's not that you'll understand the PPC architecture and Akos' optimisation implication to the PPC platform.... It's good now that you've lost interest...
I might as well lose interest in this project entirely thanks to people like you....
As you mention in your post, the enhanced features are as significant for me as any calibration/optimization capability. And since the machines I have access to are all Intel, most with HT turned on for other applications that run on them, the CPU affinity and binding one project to each real/virtual CPU are very important features. I suspect that as dual core CPU's become more prevalent, they'll probably be more significant to others, also.
Which means that BoincStudio is probably not for me right now, but you can be sure I'll be keeping an eye on it to see if those show up!
In a few weeks when we start
)
In a few weeks when we start doing analysis of LIGO/GEO S5 data, we plan to start widely distributing optimized applications containing Akos's modifications. We will also begin to award credit in a more uniform way: the credit awarded for 'the same length' workunits will be predetermined by the project and will be the same for everyone.
This system should be fairer for everyone.
Cheers,
Bruce
Director, Einstein@Home
Thanks for the good news,
)
Thanks for the good news, Bruce.
I'm sure it will some solve the credit issue if crunch time among different platforms will be similar. We went that way on CPDN - just couple of types of climate models (phew, not a simple thing) with fixed credit per model.
No need of bechamarks, cool :-)
I see that project will benefit widely from cooperation with akosf...very good.
RE: We will also begin to
)
Good. The premise that point-of-use information should be part of the credit calculation makes some sense in the SETI case where some work units running on properly running machines essentially "error out" in seconds. As apparently the project can't figure this out in advance, only point of use information can adjust.
But for Einstein the consistency of behavior of the WU's seems to be very high, so having the project pre-calculate the relative effort for each WU and simply award pre-determined credit for success seems excellent to me.
One fine thing is that it preserves the incentive for any and all forms of optimization. Do more useful work, get more credit. No overclaiming, no underclaiming. No harm to quorum partners.
I like this much better than application flop counting!
RE: In a few weeks when we
)
Sounds like the PowerPC platform is at the losing end.... but what the heck, it's just science....
RE: Sounds like the PowerPC
)
Wow, Is it typical for PowerPC's to claim so much more credit than other platforms (about 3 times as much) for the same work? Must be a bug in BOINC?
I don't know what you mean with "at the losing end" though - the quorum system usually sorts out that bug anyway so you get similar credit as everybody else.
Join the #1 Aussie Alliance on Einstein
RE: In a few weeks when we
)
Excellent news. :)
Since Einstein@home doesn't seem to have all the wu that for various reasons ends early like some other projects is having, using a pre-determined credit for each wu-size, similar to Folding@home and CPDN, is a good idea.
The biggest "problem" will be the cross-project-calibration, so Einstein@home gives roughly equal granted credit as the other BOINC-projects...
... And, handling all the complaints from users running optimized S4-applications that suddenly doesn't get 4x or so more credit/hour than other users...
BTW, now you don't need 3 results to make granted credit "fair" any longer, will this also mean can decrease min_quorum to 2 and get a further 50% speed-up, or is the results so variable that still needs 3 to make sure validated scientific results?
Also, what will the wu-size be under S5, similar range as now, or longer now you've got the large speed-up from optimizations?
"I make so many mistakes. But then just think of all the mistakes I don't make, although I might."
RE: RE: Sounds like the
)
Don't make a fool out of yourself. And mind you, I'm utilising a superbenched version of BOINC from Team MacNN for SETI@home....
And when I say "at the losing end", I don't mean the deeply-troubled quorum system based on similar credits, but to the upcoming centralised credit/WU system (whatever they call it) on a dissimilar CPU crunch time....
RE: Don't make a fool out
)
I observed that a PowerPC was claiming a lot more credit for each work unit than most other hosts. How does that make me a fool? If there's something important I missed, perhaps you could explain that, rather than resorting to making nasty comments.
As I read it everybody will (with the new system Bruce referred to) get the same credit for the same job, regardless of hardware or software used. Fast computers will do more work per day than slow ones and thus get more credit per day. So I don't understand the distinction about the PowerPC being at the "losing end" any more than any other computer under the new system.
But don't bother explaining it - I've lost interest.
Join the #1 Aussie Alliance on Einstein
RE: RE: Don't make a fool
)
What the point of me explaining anyway since you're not a PowerPC user anyway....? It's not that you'll understand the PPC architecture and Akos' optimisation implication to the PPC platform.... It's good now that you've lost interest...
I might as well lose interest in this project entirely thanks to people like you....
Thanks, Honza! As you
)
Thanks, Honza!
As you mention in your post, the enhanced features are as significant for me as any calibration/optimization capability. And since the machines I have access to are all Intel, most with HT turned on for other applications that run on them, the CPU affinity and binding one project to each real/virtual CPU are very important features. I suspect that as dual core CPU's become more prevalent, they'll probably be more significant to others, also.
Which means that BoincStudio is probably not for me right now, but you can be sure I'll be keeping an eye on it to see if those show up!
.
