Einstein Tasks' Time "To Completion" Incorrectly Going Up Instead of Down
30 Apr 2010 22:35:46 UTC
Topic 194900
(moderation:
My Einstein tasks are behaving strangely. My SETI tasks are not.
My Einstein tasks' time "To completion" incorrectly goes up instead of down. Their "Elapsed" time correctly goes up. Eventually, the tasks do complete but it's weird to see the time to completion incrementing up instead of down.
I am running BOINC v6.10.43 on my Apple iMac computer under Mac OS X 10.6.3 (Snow Leopard).
Einstein Tasks' Time "To Completion" Incorrectly Going Up Instea
)
Did you really have to post that 3 times? This thread, in this thread and in this one?
My Einstein tasks are
)
My Einstein tasks are behaving strangely. My SETI tasks are not.
My Einstein tasks' time "To completion" incorrectly goes up instead of down. Their "Elapsed" time correctly goes up. Eventually, the tasks do complete but it's weird to see the time to completion incrementing up instead of down.
This time to completion bug seems to occur only in Einstein jobs using an application of "Global Correlations ...". For example, right now on my iMac (Apple desktop computer), I have two (2) Einstein jobs running. The first job is a "Global Correlations S5 search ..." & the second is a "Arecibo Binary Pulsar Search ...". Only the Global Correlations job shows an incorrectly increasing time to completion. The Arecibo job displays it correctly (decreasing)
The time to completion bug (incrementing instead of decrementing) happens on both my iMac desktop and my Mac Book Pro laptop. Currently, I am running BOINC v6.10.56 on these Apple computers under Mac OS X 10.6.3 (Snow Leopard).
RE: My Einstein tasks are
)
I am using Linux and BOINC 6.10.56. My Time to completion goes up too and then goes down abruptly in a seesaw curve. This happens also in other projects.
Tullio
Indeed, BOINC needs to
)
Indeed, BOINC needs to collect a few statistics on the runtime of the apps before it can make accurate predictions of the runtime, it should eventually "learn" to estimate more precisely. This happens all the time, as Tullio mentioned.
Having said that, there are a few real error scenarios where this symptom can happen. What you will want to check is whether the Global Correlations task in question is progressing at all or if it's somehow frozen (e.g. by a false alarm of a virus scanner).
On a moderately modern PC, the GC1 tasks should complete within at most (say) 12 hours of total processing time. If it takes longer, there might be something wrong. If the task seems to be hung, try stopping boinc completely and restart it (or reboot the whole PC if it's not too inconvenient to make sure), and check the output from your virus checker whether it has put the app into quarantine.
Hope this helps
HB
RE: On a moderately modern
)
That's correct for my 2 core Opteron 1210 running Linux. One WU took longer but the wingman with a 8 core Xeon, also on Linux, errored out and a second 8 core on Linux is still crunching, so maybe there is something wrong with this GC1 WU.
Tullio
Latest GC1 result:12:57:46,
)
Latest GC1 result:12:57:46, still pending, like two others.Many errors on my wingmen, all Linux.
Tullio
RE: If the task seems to be
)
Don't restart it too quickly. I have a BOINC Manager (own make) that shows both CPU and Wall time and there's a consistent 30 minute discrepancy between the two. The Wall time shows the amount of time it took from start of task to finish of task, while the CPU time shows only the amount of time the CPU actually did any work.
On AQUA@home, multithreading,
)
On AQUA@home, multithreading, CPU time is much greater than elapsed time.
Tullio
I have noticed the same
)
I have noticed the same "increasing time remaining" on my PIII 1.4 Netservers (W2Ksp4) OS. So what is the solution? Just leave all alone and monitor if it completes by the deadline? Abort and hope for a WU that counts down the correct direction? I thought maybe I had caused the problem by setting my server clocks after I had started E@H.
RE: I have noticed the same
)
Which type of task are you looking at?
The Gamma-Ray Pulsar Search tasks (task name starts 'LATeah...') only register progress in 2% increments - and each 2% could take 40 minutes or more on your PIIIs. During that time, BOINC's estimate will slowly increase, then jump back down as the next 2% progress is registered.
Other task types are more linear (though h1_... Gravitational Wave searches take a while to get started), but in general I'd suggest you let them run for a while and try to get a feel for how they behave over time.