Well I am doing what Jeroen told me to do: Torture the server with requests for GPU WUs every few minutes :D
That seems to work. Thanks for the advice.
Well I am doing what Jeroen told me to do: Torture the server with requests for GPU WUs every few minutes :D
That seems to work. Thanks for the advice.
I set my update interval from 15 - 60 minutes depending on how much work is available. That is generally sufficient to keep the GPUs busy.
Just to give an idea how fast we are crunching the data with GPUs:
What is the BRP4 actually doing with a workunit?
Each workunit ist really a bundle of 8 preprocessed datafiles from Arecibo.
In essence, each data file contains roughly 4.5 minutes of observation. The pre-processing that has been done on the server is the so called de-dispersion: compensating for distortions of a signal from a given range of distance (roughly speaking). So it's like that datafile is made to look for pulsars at a certain dispersion measure (or roughly distance, or within a certain "onion layer" around earth, if you want).
After removing some known interferences from the signal, the main loop of the application will analyse the signal under the hypothesis of a certain orbit of a pulsar source. In fact, lots of orbits: 6662 orbit templates AFAIK for the Arecebibo BRP4 data. Each template analysis requires an FFT of the 4.5 minute signal.
Because there are 8 datafiles per BRP4 workunit, the total number of templates (= number of FFTs) per workunit is roughly 53,000
With a pretty inexpensive CUDA card ( each template (checking 4.5 minutes of data for a binary pulsar at a given distance and given orbital parameters) takes 60 milliseconds. [/b]
I think this is amazing.
So it is perhaps understandable that the servers are having problems keeping up with the clients, especially in an early phase of BRP4 where clients may need to refill their work caches.
RE: But everything works
)
Hi
As far as I can see, your PC with app_info file just got several BRP4 tasks, so I guess the problem, if any, is solved now, right?
HBE
Well I am doing what Jeroen
)
Well I am doing what Jeroen told me to do: Torture the server with requests for GPU WUs every few minutes :D
That seems to work. Thanks for the advice.
RE: Well I am doing what
)
I set my update interval from 15 - 60 minutes depending on how much work is available. That is generally sufficient to keep the GPUs busy.
RE: RE: But everything
)
Thanks. Don't need one here, but boinc's 4 hour timeout for any server glitch goes very poorly with MW's only giving 15-30minutes of work for my 5870.
Hi all, Just to give an
)
Hi all,
Just to give an idea how fast we are crunching the data with GPUs:
What is the BRP4 actually doing with a workunit?
Each workunit ist really a bundle of 8 preprocessed datafiles from Arecibo.
In essence, each data file contains roughly 4.5 minutes of observation. The pre-processing that has been done on the server is the so called de-dispersion: compensating for distortions of a signal from a given range of distance (roughly speaking). So it's like that datafile is made to look for pulsars at a certain dispersion measure (or roughly distance, or within a certain "onion layer" around earth, if you want).
After removing some known interferences from the signal, the main loop of the application will analyse the signal under the hypothesis of a certain orbit of a pulsar source. In fact, lots of orbits: 6662 orbit templates AFAIK for the Arecebibo BRP4 data. Each template analysis requires an FFT of the 4.5 minute signal.
Because there are 8 datafiles per BRP4 workunit, the total number of templates (= number of FFTs) per workunit is roughly 53,000
With a pretty inexpensive CUDA card ( each template (checking 4.5 minutes of data for a binary pulsar at a given distance and given orbital parameters) takes 60 milliseconds. [/b]
I think this is amazing.
So it is perhaps understandable that the servers are having problems keeping up with the clients, especially in an early phase of BRP4 where clients may need to refill their work caches.
HBE
RE: RE: RE: But
)
Except I might end up pulling the plug. Each time it fires I get a dos window strobing on my screen for one or two frames. Really annoying.
RE: Except I might end up
)
You could setup a script file in windows to run boinccmd. The 0 in the Run command hides the window.
bupdate.vbs
Set WinScriptHost= CreateObject("Wscript.Shell")
WinScriptHost.Run Chr(34) & "C:\Program Files\BOINC\boinccmd.exe" & Chr(34) & " --project http://einstein.phys.uwm.edu/ update" & Chr(34), 0
Set WinScriptHost= Nothing
RE: RE: Except I might
)
Thanks. This seems to be working perfectly.