Gamma-ray pulsar search #5 longer computing time
5 Apr 2024 21:19:52 UTC
Topic 230966
(moderation:
On my old Xeon machine ( https://einsteinathome.org/host/3426951 ) the Gamma-ray pulsar search #5 v1.11 () x86_64-apple-darwin tasks suddenly take 20.000 instead of 10.000 seconds for the same credits. Was there a change in the work units. Or is there something wrong with my machine?
B.I.G wrote: On my old Xeon
)
Best as I can tell, it is the Task I.D. "LATeah2011F" vs "LATeah2014F" with the former taking ~10,000 secs. Nothing that you can do about it. Sorry.
Proud member of the Old Farts Association
Do 2011 and 2014 mean the
)
Do 2011 and 2014 mean the year when the data was received?
Pavel_Kirpichenko wrote: Do
)
No, they were both built in the same year: Jul 26 2017. You can find this information in the 'Stderr output' file of the task in question on the 4th line. It's really easy.
Proud member of the Old Farts Association
No George, he was asking
)
No George, he was asking about the data sets, not the date the science application was built.
The naming of the data sets COULD be the year the data was acquired, OR it could just be sequential numbering of the data set the tasks are part of.
Likely to be the year the data was acquired but just a guess. A scientist would have to chime in with the truth here.
B.I.G schrieb: On my old
)
These tasks analyze different raw data files (Keith explained it already). The analysis configuration changes with different raw data. Even within the same science run (crunching same LATeah20xxF file) different tasks apply different analysis configuration, that is, e.g. the number of skypoints differ and thus the specific calculations (see task descriptions in client_state.xml or command line printout in stderr.txt. There are lots of parameters). From my observations: The recent FGRP5 tasks with only four skypoints require significantly less CPU time compared to current ones with 40...44 skypoints. FGRP5 task with two digit numbers (NN) in task name after prefix "LATeah2014F_", e.g.: "LATeah2014F_72.0_3252_-1.7e-11_0" have few skypoints (four, or six for previous raw data files).
In this regard the credit mechanism for FGRP5 tasks is not perfectly fair. They all get the same credits independent of required computation effort, which differs significantly.
But hey, We're crunchin for science not credits, aren't we? ;-)
Scrooge McDuck wrote: But
)
Thank you for the detailed explanation, and yes, it's all about the science, I was more worried there could be something wrong with that old machine than about my credits. Credits are nice and all but science rocks :)