Theli-Forum

Hi all,
what can you say about multithreading in THELI GUI?

Hi Oleg,

I'm not sure what you mean? When reducing data from a single-chip camera, then only some processes are multi-threaded, such as astrometry, sky modeling and coaddition. If you reduce data from a multi-chip camera, then the data from the chips will be equally distributed amongst the available CPUs.

mischa

I say about the multi-chip cameras data processing.

Look at my screenshots from OmegaCam@VST data processing:

step #1


step #2


step #3


step #4

I have the question about step #3 (or Coaddition: coadding images...) in menu "Coaddition".

Why does this step use so small powerful of CPU/threads???

CPU1 7.9% CPU2 7.9% CPU3 3.9% CPU4 3.0%
CPU5 6.0% CPU6 10.7% CPU7 2.0% CPU8 0.0%
CPU9 6.8% CPU10 2.0% CPU11 2.0% CPU12 2.0%

Hi Oleg,

not all processes can be fully parallelised. Step 2 has a full CPU load because each CPU can resample data from the chip assigned to it independently from the other CPUs.
At step 3 all the resampled images are read and combined into the output image. I guess this is a more sequential progress. I remember Emmanuel Bertin mentioning that this part of Swarp is more difficult to multi-thread.

You can cook 10 chicken at the same time, but you can only eat them one after the other.

mischa

Look at my another screenshot from "Astro+photometry":

This step uses only one thread (P.S. pardon: three threads) for Calculation astrometric solution.



Is it normally?
OR

Is it the problem with SCAMP compilation???

I see three CPUs at 100%. Not all parts of Scamp are fully parallelised. The initial matching with the WCS is fully multi-threaded, the calculation of the higher order distortions is only partially multi-threaded.

As I said before, not all algorithms can be done in parallel.

mischa hat geschrieben:I see three CPUs at 100%. Not all parts of Scamp are fully parallelised. The initial matching with the WCS is fully multi-threaded, the calculation of the higher order distortions is only partially multi-threaded.

As I said before, not all algorithms can be done in parallel.

Ok. Many thanks!