Index  | Recent Threads  | Unanswered Threads  | Who's Active  | Guidelines  | Search
 

Quick Go ยป
No member browsing this thread
Thread Status: Active
Total posts in this thread: 1
[ Jump to Last Post ]
Post new Thread
Author
Previous Thread This topic has been viewed 2677 times and has 0 replies Next Thread
JmBoullier
Former Community Advisor
Normandy - France
Joined: Jan 26, 2007
Post Count: 3715
Status: Offline
Project Badges:
Reply to this Post  Reply with Quote 
How long will this new type-A Clean Energy Project (CEP) work unit (WU) last?

Warning (as of April 15, 2009): I just noticed that type A WUs are no longer exactly following the scenario described below. Percentage done jumps by 0.020 % now and checkpoints occur at multiples of 1.600 %. Therefore this post needs to be reviewed in more detail... if it appears that the process is as linear as before. I need first to confirm it.
End of warning.

Because of the big differences of durations of the type-A CEP WUs Boinc is often completely lost when estimating the duration of CEP WUs waiting to start.
Hopefully WCG techs will be able to standardize those durations in the future but meanwhile it can be useful to make a better estimate and it is possible once a CEP WU of type A has started.


Preliminary note: a type-A CEP WU is a WU whose name has the following pattern:
E000033_423A _00043y013
Later on there will also be type-B and type-C WUs, but at this stage of the Clean Energy Project only type-A WUs are distributed.

One good thing of type-A CEP WUs is that their progress is very linear:
  • first there is an initialization phase during which the percentage stays at 0 %. This phase may last more than one hour for very long WUs
  • next the percentage is incremented by 0.031 % (or more precisely 4 times per 0.125 %) at a regular pace
  • checkpoints are taken every 2.5 % from 2.5 % until a last one at 100 %.

    Consequences:
  • there are 40 checkpoints per WU, with a period almost constant. In fact it is slowly increasing all over the computation but the last interval is only about 6 % longer than the first ones.
  • the percentage is incremented 3,200 times per WU, and the interval between two changes of the percentage is almost constant, with the same slight increase as for the checkpoints.

    Last, it appears that the initialization phase lasts about half the time between the first two checkpoints.

    With all these characteristics it is possible to predict the total runtime of a WU with a reasonable precision (about 5 % and less than 10 % in the worst cases), once the percentage of a type-A CEP WU has started to move.
    Until then it is only possible to predict that it will need at least 81 times the CPU time already spent during this phase.
    For example if your WU is still in its initialization phase and the CPU time shows 10 minutes, then its total runtime will be at least 810 minutes, i.e. 13:30:00 hours.

    Method 1.
    Once the percentage has started to change watch how much CPU time it takes between two changes. The total runtime will be slightly more than 3,200 times this interval, actually 3,240 times with the initialization phase.
    For example if it takes 20 seconds of CPU time between two changes of the percentage your WU should need about 3240 * 20 = 64,800 seconds, i.e. 18 hours of runtime.

    Method 2.
    If you are logging the checkpoints in your message log you can use it to find the interval between two consecutive checkpoints, i.e. between the time shown for the checkpoint at 5 % and the time shown for the checkpoint at 2.5 %. Then multiply it by 40.5 and you will have the minimum runtime needed by your WU.
    For example the interval between those two checkpoints is 36 minutes, or 0.6 hour. Your WU will need about 40.5 * 0.6 = 24.3 hours.

    Method 3.
    Once the percentage is at least 20-25 % you can quickly extrapolate the total runtime with a reasonable precision from the CPU time already spent and the percentage reached as follows: 100 * CPU time / percentage.
    For example your WU just passed 12:15:00 hours (i.e. 12.25 hours) and your percentage shows 41.125 %. The total runtime will be about 100 * 12.25 / 41.125 = 29.79 hours.

    These formulas must be used inside their limits, i.e. method 1 until 5 or 10 %, then method 2 until 20 or 25 %, then only, method 3 will give acceptable results.
    They will only be approaching the final runtime in all circumstances, but they can be useful to find that some WU that Boinc was estimating at 12 hours will actually last about 55 hours, or vice-versa obviously.

    Note: For method 2 you need to see the checkpoints in your message log. If this is not yet the case see Project Checkpoint Saving - How to Minimize Progress Loss on Close/Restart to see how to have them logged.
  • ----------------------------------------
    Team--> Decrypthon -->Statistics/Join -->Thread
    ----------------------------------------
    [Edit 1 times, last edit by JmBoullier at Apr 15, 2009 7:49:24 PM]
    [Dec 16, 2008 4:48:56 PM]   Link   Report threatening or abusive post: please login first  Go to top 
    [ Jump to Last Post ]
    Post new Thread