// DBOINCP-300: added node comment count condition in order to get Preview working ?>
John_Waters
Joined: 1 Jul 11
Posts: 20
Credit: 76699157
RAC: 0
14 Feb 2018 20:07:18 UTC
Topic 213408
(moderation:
)
My gamma-ray/gpu units reach 89.997% with 1:40 to 1:45 remaining, then seemingly hang up for~ 3 Minutes, after which they go to 100% and vanish. Can anyone tell me what"s going on with this?
That is the normal behavior. The last phase of the analysis is computationally different (if you monitor measures of GPU activity you will seen it suddenly drop in that phase, along with temperature and power consumption) and the application does not update guesses of completion percentage in the midst of that phase. If you look around here, you'll find plenty of descriptions of the behavior, and some more details of what is going on.
Also, the detailed GPU behavior in that phase varies with the frequency of the data in a Work Unit.
My gamma-ray/gpu units reach 89.997% with 1:40 to 1:45 remaining, then seemingly hang up for~ 3 Minutes, after which they go to 100% and vanish. Can anyone tell me what"s going on with this?
Crunching of FGRP tasks (both CPU and GPU type) is done in two stages. The first stage (~90% of the total time) is for identifying possible candidate signals in the data. The follow-up stage (~10%) is for re-processing the ten most likely candidates using double precision.
There is a continuously changing estimate of progress, and of remaining time, during the main stage. At 89.997% when the main stage is finished and the follow-up stage commences, there are no further increments in %done until the very end. Also the remaining time estimate changes very little, simply because it is tied to %done which has stopped increasing. None of this is any sort of problem - it's just normal behaviour as Archae86 mentions.
Your completed tasks don't 'vanish'. On completion, when the %done suddenly jumps to 100%, the results get uploaded and the status changes to 'Ready to report'. They may not get reported immediately - it all depends on whether or not your BOINC client sees the need to re-fill the work cache. When BOINC decides to make a work request, any completed tasks will get reported automatically.
That is the normal behavior.
)
That is the normal behavior. The last phase of the analysis is computationally different (if you monitor measures of GPU activity you will seen it suddenly drop in that phase, along with temperature and power consumption) and the application does not update guesses of completion percentage in the midst of that phase. If you look around here, you'll find plenty of descriptions of the behavior, and some more details of what is going on.
Also, the detailed GPU behavior in that phase varies with the frequency of the data in a Work Unit.
John_Waters wrote:My
)
Crunching of FGRP tasks (both CPU and GPU type) is done in two stages. The first stage (~90% of the total time) is for identifying possible candidate signals in the data. The follow-up stage (~10%) is for re-processing the ten most likely candidates using double precision.
There is a continuously changing estimate of progress, and of remaining time, during the main stage. At 89.997% when the main stage is finished and the follow-up stage commences, there are no further increments in %done until the very end. Also the remaining time estimate changes very little, simply because it is tied to %done which has stopped increasing. None of this is any sort of problem - it's just normal behaviour as Archae86 mentions.
Your completed tasks don't 'vanish'. On completion, when the %done suddenly jumps to 100%, the results get uploaded and the status changes to 'Ready to report'. They may not get reported immediately - it all depends on whether or not your BOINC client sees the need to re-fill the work cache. When BOINC decides to make a work request, any completed tasks will get reported automatically.
Cheers,
Gary.