You, and others, have received a disproportionate share of the "short" WUs. Most of the hosts that have received them are not crunching them at the rate your rig does, I think you realize that, so for them, it's no problem.
It occured to me, as it has to a few others, that one possible remedy would be to "match" those shorties to slower hosts. I took the opportunity to pose this possibility to Dr. Bruce Allen, here, and received a reply that I think you'll like to read, including the few posts before and after. Essentially, there are few short WUs remaining to distribute, and then things will return to something resembling normal.
Increasing daily quotas as you suggested, would increase the severity of a lot of other problems that we have in abundance already, WU hogging, missed deadlines (especially to be avoided now that a WU is assigned to only a triplet of hosts), and other abuses. Believe me, there is more aggravation from waiting weeks or more for credit on a whole string of long-completed WUs than for having a rig sit idle for a few hours a day. The problem of an idle machine, after all, is easily solved by attaching to another project and giving it a tiny resource share, as has been suggested earlier. The idle machine thing is not new at all. Many Mac crunchers have been dealing with the thing since the newer Mac app came out several months ago, and handling it very well and graciously.
So, I hope the news I've given you helps you "deal with it", because deal with it you must - increasing daily quotas across the board to accomodate the few of you who momentarily find yourselves with idle rigs just ain't gonna happen.
Respects,
Michael
microcraft
"The arc of history is long, but it bends toward justice" - MLK
[So, I hope the news I've given you helps you "deal with it", because deal with it you must - increasing daily quotas across the board to accomodate the few of you who momentarily find yourselves with idle rigs just ain't gonna happen.
Michael; Thanks to you and all for the wise counsel. Last evening 1 of my rigs ripped through some of these WU's in under an hour each. When I uploaded, detached, and reattached, I finally got a sweet download of "regular" Alberts. For the time being I can return to autopilot for my rigs. Micro managing the farm due to the abbreviated crunch times was not fun. Remember, I depend on my fully loaded machines to heat my residence. An idle box generates few BTU's. I appreciate all your responses (I hope the CD Rom guy got some help).
IMHO the main failure here is one of communication....for whatever reason. It's unfortunate that this question could not have been addressed in this thread in a more timely fashion. Thanks Michael for supplying the link to the answer we were seeking....Cheers, Rog.
Since these short WU represent only about 1% of the WU that we're issuing in the new S4 search, I don't think they will be a large effect over time. Are many people still running into this problem? I have increased the daily result quotas, and am willing to increase them more if needed. I don't want to leave fast machines idle!
Since these short WU represent only about 1% of the WU that we're issuing in the new S4 search, I don't think they will be a large effect over time. Are many people still running into this problem?
Bruce
Dr. Allen,
My host is curently working on a set of short ones, about 21% the size of the einsteins, or 1hr 2min each, so I don't think they are the ultra-shorties. I've attached to Seti and given each a 50% share, so it completes 11 of one and 12 of the other per day. No problem.
Michael
edit - I know that the flip side of the "idle machine" problem is the "WU pending" problem, which is also looming large, so I'd personally not like an increased quota - it would exascerbate the delays in validation, credit return time, science return time and extra load on the servers, due to the ever-present hit-and-run crunchers.
microcraft
"The arc of history is long, but it bends toward justice" - MLK
Since these short WU represent only about 1% of the WU that we're issuing in the new S4 search, I don't think they will be a large effect over time. Are many people still running into this problem?
Quote:
Thanks for the information Bruce....always good to hear from you when you have time. The MDQ of 16 is OK for most of our machines. We must monitor the few faster machines, though, as they can process about 23 LF WUs/day when they have them. We can live with the present MDQ if/as it makes for a more stable project in the long run. One percent odds are pretty good I think.....Cheers, Rog.
Bruce; This evening one of my rigs uploaded 21 WU's. Congratulations! I got new work. You seem to be mixing enough regular Alberts with the "24 divided by 1 equals 16" WU's to keep me out of the dead box club. I hope that my detaching and reattaching to keep my rigs crunching had no negative impact on the project. Thanks for you response. I believe I am associated with the primo distributed computing project out there due to your support and server availability. Bravo.
Hmmm.. color me dense, but I've managed to read through this entire thread and yet I'm failing to understand the perceived problem(s).
Currently, I have 19 rigs running solely this project. They all plow through workunits constantly(24/7) whether the work be Alberts or normals. I have yet to find any of the machines sitting idle at any time.
"Chance is irrelevant. We will succeed."
- Seven of Nine
Hmmm.. color me dense, but I've managed to read through this entire thread and yet I'm failing to understand the perceived problem(s).
Currently, I have 19 rigs running solely this project. They all plow through workunits constantly(24/7) whether the work be Alberts or normals. I have yet to find any of the machines sitting idle at any time.
GRRO,
Nobody would like to suggest that you are dense, just that you have not yet experienced any of these sets of short albert WUs.
For instance, my rig used to crunch the old "regular-size" einstein WUs in ~ 5 hrs. In the set of "alberts" it's currently working, crunchtimes are ~ 1hr 2min, so they are only 21% the "size" of einsteins. See here. You can do the arithmetic. 3750sec x 16 = 60000 secs, or 16hrs 40 min.
Notice that I've made no complaint over this, I've simply followed my own suggestion and added another project, but also keep in mind that with the much-varying size of albert worksets, there are some with even smaller WUs that my rig is now crunching, and a few classes of machines (most notably the newer PowerMacs) that crunch faster than mine. I omit references to multi-core rigs and hyper-threaded P4s that are more productive than mine, because they are allotted 16/day times the # of cores or virtual cores.
Regards,
Michael
edited for more info
microcraft
"The arc of history is long, but it bends toward justice" - MLK
Tweakster, You, and
)
Tweakster,
You, and others, have received a disproportionate share of the "short" WUs. Most of the hosts that have received them are not crunching them at the rate your rig does, I think you realize that, so for them, it's no problem.
It occured to me, as it has to a few others, that one possible remedy would be to "match" those shorties to slower hosts. I took the opportunity to pose this possibility to Dr. Bruce Allen, here, and received a reply that I think you'll like to read, including the few posts before and after. Essentially, there are few short WUs remaining to distribute, and then things will return to something resembling normal.
Increasing daily quotas as you suggested, would increase the severity of a lot of other problems that we have in abundance already, WU hogging, missed deadlines (especially to be avoided now that a WU is assigned to only a triplet of hosts), and other abuses. Believe me, there is more aggravation from waiting weeks or more for credit on a whole string of long-completed WUs than for having a rig sit idle for a few hours a day. The problem of an idle machine, after all, is easily solved by attaching to another project and giving it a tiny resource share, as has been suggested earlier. The idle machine thing is not new at all. Many Mac crunchers have been dealing with the thing since the newer Mac app came out several months ago, and handling it very well and graciously.
So, I hope the news I've given you helps you "deal with it", because deal with it you must - increasing daily quotas across the board to accomodate the few of you who momentarily find yourselves with idle rigs just ain't gonna happen.
Respects,
Michael
microcraft
"The arc of history is long, but it bends toward justice" - MLK
[So, I hope the news I've
)
[So, I hope the news I've given you helps you "deal with it", because deal with it you must - increasing daily quotas across the board to accomodate the few of you who momentarily find yourselves with idle rigs just ain't gonna happen.
Michael; Thanks to you and all for the wise counsel. Last evening 1 of my rigs ripped through some of these WU's in under an hour each. When I uploaded, detached, and reattached, I finally got a sweet download of "regular" Alberts. For the time being I can return to autopilot for my rigs. Micro managing the farm due to the abbreviated crunch times was not fun. Remember, I depend on my fully loaded machines to heat my residence. An idle box generates few BTU's. I appreciate all your responses (I hope the CD Rom guy got some help).
Regards-tweakster
IMHO the main failure here is
)
IMHO the main failure here is one of communication....for whatever reason. It's unfortunate that this question could not have been addressed in this thread in a more timely fashion. Thanks Michael for supplying the link to the answer we were seeking....Cheers, Rog.
Since these short WU
)
Since these short WU represent only about 1% of the WU that we're issuing in the new S4 search, I don't think they will be a large effect over time. Are many people still running into this problem? I have increased the daily result quotas, and am willing to increase them more if needed. I don't want to leave fast machines idle!
Tweakster, what is your current situation?
Cheers,
Bruce
Director, Einstein@Home
RE: Since these short WU
)
Dr. Allen,
My host is curently working on a set of short ones, about 21% the size of the einsteins, or 1hr 2min each, so I don't think they are the ultra-shorties. I've attached to Seti and given each a 50% share, so it completes 11 of one and 12 of the other per day. No problem.
Michael
edit - I know that the flip side of the "idle machine" problem is the "WU pending" problem, which is also looming large, so I'd personally not like an increased quota - it would exascerbate the delays in validation, credit return time, science return time and extra load on the servers, due to the ever-present hit-and-run crunchers.
microcraft
"The arc of history is long, but it bends toward justice" - MLK
RE: Since these short WU
)
RE: tweakster, what is your
)
Bruce; This evening one of my rigs uploaded 21 WU's. Congratulations! I got new work. You seem to be mixing enough regular Alberts with the "24 divided by 1 equals 16" WU's to keep me out of the dead box club. I hope that my detaching and reattaching to keep my rigs crunching had no negative impact on the project. Thanks for you response. I believe I am associated with the primo distributed computing project out there due to your support and server availability. Bravo.
Regards-tweakster
Hmmm.. color me dense, but
)
Hmmm.. color me dense, but I've managed to read through this entire thread and yet I'm failing to understand the perceived problem(s).
Currently, I have 19 rigs running solely this project. They all plow through workunits constantly(24/7) whether the work be Alberts or normals. I have yet to find any of the machines sitting idle at any time.
"Chance is irrelevant. We will succeed."
- Seven of Nine
RE: Hmmm.. color me dense,
)
GRRO,
Nobody would like to suggest that you are dense, just that you have not yet experienced any of these sets of short albert WUs.
For instance, my rig used to crunch the old "regular-size" einstein WUs in ~ 5 hrs. In the set of "alberts" it's currently working, crunchtimes are ~ 1hr 2min, so they are only 21% the "size" of einsteins. See here. You can do the arithmetic. 3750sec x 16 = 60000 secs, or 16hrs 40 min.
Notice that I've made no complaint over this, I've simply followed my own suggestion and added another project, but also keep in mind that with the much-varying size of albert worksets, there are some with even smaller WUs that my rig is now crunching, and a few classes of machines (most notably the newer PowerMacs) that crunch faster than mine. I omit references to multi-core rigs and hyper-threaded P4s that are more productive than mine, because they are allotted 16/day times the # of cores or virtual cores.
Regards,
Michael
edited for more info
microcraft
"The arc of history is long, but it bends toward justice" - MLK