I've been watching the database rather closely (eg, every four to six hours) for the past month, since I started distributing Albert jobs on December 24th. I have watched this adjust to a number of changes: the decreased length of typical jobs, the fact that we had a target of 3 rather than 4 results, changes in internal scheduler parameters that govern how old an unsent result can be before it gets sent off to the first possible host, and a number of other small changes.
The major effect of changing the maximum daily quota per CPU is that (1) machines that are misconfigured and error out more work will wipe out a few more results before being shut down by the punishment algorithm and (2) very fast machines won't be starved for work. But currently only 4% of Albert results fail on the host machines. So the effect of (1) will be small.
I am not overly worried about the 'time to credit' issue. The average age in the E@H database of workunits WITHOUT a canonical result is 404302 seconds (4.7 days). The standard deviation is 4.3 days. This means that the average workunit is FINISHED (credit granted) in 4.7 days. All but about 25% are finished in 9 days. I think this is a reasonable length of time. If I see this change significantly due to the modification in the maximum daily quotas, then I'll revert it back.
Note: the BOINCSTATS project graph showing credit per day for E@H during the past 60 days indicates that we are starting to do more work than we have historically accomplished. So from the scientific point of view the project is working better than it has in the past, not worse.
I've been watching the database rather closely (eg, every four to six hours) for the past month, since I started distributing Albert jobs on December 24th. I have watched this adjust to a number of changes: the decreased length of typical jobs, the fact that we had a target of 3 rather than 4 results, changes in internal scheduler parameters that govern how old an unsent result can be before it gets sent off to the first possible host, and a number of other small changes.
The major effect of changing the maximum daily quota per CPU is that (1) machines that are misconfigured and error out more work will wipe out a few more results before being shut down by the punishment algorithm and (2) very fast machines won't be starved for work. But currently only 4% of Albert results fail on the host machines. So the effect of (1) will be small.
I am not overly worried about the 'time to credit' issue. The average age in the E@H database of workunits WITHOUT a canonical result is 404302 seconds (4.7 days). The standard deviation is 4.3 days. This means that the average workunit is FINISHED (credit granted) in 4.7 days. All but about 25% are finished in 9 days. I think this is a reasonable length of time. If I see this change significantly due to the modification in the maximum daily quotas, then I'll revert it back.
Note: the BOINC SYNERGY project graph for E@H during the past 60 days indicates that we are starting to do more work than we have historically accomplished. So from the scientific point of view the project is working better than it has in the past, not worse.
Cheers,
Bruce
Dr. Allen,
I've noted the BoincSynergy data, too, and I'm pleased but not surprised that E@H has fared well. I think that we did as good a job as could be done at assimilating SetiClassic folks, and as one of those who contributed time on the helpdesk during those hectic, frantic weeks, I can say that our job then was made much easier by the remarkable stability and reliability of E@H. We continue to enlist Boinc noobies as they start to explore the expanding horizons that BOINC allows, and their praise for E@H is the more emphatic after they've experienced work shortages and other glitches on some of the other BOINC-based offerings - E@H .. just .. works .. well .. in .. comparison. I'm glad that you will be closely monitoring the effects of the MDQ change, because I'd surely like for the trend that we've been seeing to continue smoothly.
Respectfully,
Michael
microcraft
"The arc of history is long, but it bends toward justice" - MLK
......[snip]........
The major effect of changing the maximum daily quota per CPU is that (1) machines that are misconfigured and error out more work will wipe out a few more results before being shut down by the punishment algorithm and
......[another snip]........
Cheers,
Bruce
Bruce,
In (1) above does the max WU's allowed per day drop below one? If not, can it?
......[snip]........
The major effect of changing the maximum daily quota per CPU is that (1) machines that are misconfigured and error out more work will wipe out a few more results before being shut down by the punishment algorithm and
......[another snip]........
Cheers,
Bruce
Bruce,
In (1) above does the max WU's allowed per day drop below one? If not, can it?
I don't think this would be a good idea. A floor of one will always allow someone with problems to try again once a day to see if they've fixed it.
....[snip]....
Bruce,
In (1) above does the max WU's allowed per day drop below one? If not, can it?
I don't think this would be a good idea. A floor of one will always allow someone with problems to try again once a day to see if they've fixed it.
I keep running into crunchers that don't appear to check their computers at all. They have heaps of no work returned errors, lots of work still to be done and I'll be matched up with one of the top three in their "yet to be done" list.
Couldn't they wait a week to get a new WU in that condition?
If there were a way for the software to automatically send a POP-UP message to a user who has returned a set number of BAD results, and keep popping up as each bad one goes in, and if it does get down to minimial, after another set number it should totally stop, and keep popping messages up, telling them of the situation.
If there were a way for the software to automatically send a POP-UP message to a user who has returned a set number of BAD results, and keep popping up as each bad one goes in, and if it does get down to minimial, after another set number it should totally stop, and keep popping messages up, telling them of the situation.
Alas, in a perfect world.
What a brilliant idea!!! Seriously, why don't you submit it to the dev-board list?
Michael
edit - One thought...they'd have to get the POP-UP to appear on top of the darn screensaver, because I think the saver is the only part of the project that those users are interested in.
(Term "users" entirely intentional in this case.)
microcraft
"The arc of history is long, but it bends toward justice" - MLK
As a charter member of the "relative handfull", (since December 29th), I want to thank the Einstein staff for their work with the MDQ. The practice I got in the running dry, detaching and reattaching to avoid the dead box club will not be missed. Yes, tweakster is warm and happy. My house is warm and happy. Our mild winter (so far) here in the midwest has meant that my air conditioning has come on briefly at times to maintain my 78 degree environment. My Einstein farm is literally heating my home. For those of you who want to tinker with the screen saver to deliver messages of doom, just remember, for the pro's, it's the first thing we disable.
[edit]pending WU's [/edit]
)
[edit]pending WU's [/edit]
"The FUTURE is only a PARTICLE away from the PRESENT and the PAST."
Hi Michael, I've been
)
Hi Michael,
I've been watching the database rather closely (eg, every four to six hours) for the past month, since I started distributing Albert jobs on December 24th. I have watched this adjust to a number of changes: the decreased length of typical jobs, the fact that we had a target of 3 rather than 4 results, changes in internal scheduler parameters that govern how old an unsent result can be before it gets sent off to the first possible host, and a number of other small changes.
The major effect of changing the maximum daily quota per CPU is that (1) machines that are misconfigured and error out more work will wipe out a few more results before being shut down by the punishment algorithm and (2) very fast machines won't be starved for work. But currently only 4% of Albert results fail on the host machines. So the effect of (1) will be small.
I am not overly worried about the 'time to credit' issue. The average age in the E@H database of workunits WITHOUT a canonical result is 404302 seconds (4.7 days). The standard deviation is 4.3 days. This means that the average workunit is FINISHED (credit granted) in 4.7 days. All but about 25% are finished in 9 days. I think this is a reasonable length of time. If I see this change significantly due to the modification in the maximum daily quotas, then I'll revert it back.
Note: the BOINCSTATS project graph showing credit per day for E@H during the past 60 days indicates that we are starting to do more work than we have historically accomplished. So from the scientific point of view the project is working better than it has in the past, not worse.
Cheers,
Bruce
Director, Einstein@Home
RE: Hi Michael, I've been
)
Dr. Allen,
I've noted the BoincSynergy data, too, and I'm pleased but not surprised that E@H has fared well. I think that we did as good a job as could be done at assimilating SetiClassic folks, and as one of those who contributed time on the helpdesk during those hectic, frantic weeks, I can say that our job then was made much easier by the remarkable stability and reliability of E@H. We continue to enlist Boinc noobies as they start to explore the expanding horizons that BOINC allows, and their praise for E@H is the more emphatic after they've experienced work shortages and other glitches on some of the other BOINC-based offerings - E@H .. just .. works .. well .. in .. comparison. I'm glad that you will be closely monitoring the effects of the MDQ change, because I'd surely like for the trend that we've been seeing to continue smoothly.
Respectfully,
Michael
microcraft
"The arc of history is long, but it bends toward justice" - MLK
RE: Hi
)
Bruce,
In (1) above does the max WU's allowed per day drop below one? If not, can it?
Join the #1 Aussie Alliance on Einstein
RE: RE: Hi
)
I don't think this would be a good idea. A floor of one will always allow someone with problems to try again once a day to see if they've fixed it.
RE: RE: ....[snip].... B
)
I keep running into crunchers that don't appear to check their computers at all. They have heaps of no work returned errors, lots of work still to be done and I'll be matched up with one of the top three in their "yet to be done" list.
Couldn't they wait a week to get a new WU in that condition?
Join the #1 Aussie Alliance on Einstein
If there were a way for the
)
If there were a way for the software to automatically send a POP-UP message to a user who has returned a set number of BAD results, and keep popping up as each bad one goes in, and if it does get down to minimial, after another set number it should totally stop, and keep popping messages up, telling them of the situation.
Alas, in a perfect world.
RE: If there were a way for
)
What a brilliant idea!!! Seriously, why don't you submit it to the dev-board list?
Michael
edit - One thought...they'd have to get the POP-UP to appear on top of the darn screensaver, because I think the saver is the only part of the project that those users are interested in.
(Term "users" entirely intentional in this case.)
microcraft
"The arc of history is long, but it bends toward justice" - MLK
RE: What a brilliant
)
Add it to the screensaver, along with a POP-UP. So the screensaver tells them the same thing that the POP-UP does.
As a charter member of the
)
As a charter member of the "relative handfull", (since December 29th), I want to thank the Einstein staff for their work with the MDQ. The practice I got in the running dry, detaching and reattaching to avoid the dead box club will not be missed. Yes, tweakster is warm and happy. My house is warm and happy. Our mild winter (so far) here in the midwest has meant that my air conditioning has come on briefly at times to maintain my 78 degree environment. My Einstein farm is literally heating my home. For those of you who want to tinker with the screen saver to deliver messages of doom, just remember, for the pro's, it's the first thing we disable.
Regards-tweakster