recommend processor & memory for off-line solving

[mlco...@gmail.com]

Hello,

Astrometry.net works fine on my single board computer, *but* it is a bit slow, taking 60 sec on average to solve.

Astrometry.net is running off-line under Ekos and is solving a 6' x 4' field, about.

Among the many variables that one adjusts for, I was wondering what processor and memory other people might be using to get solve times down.

Thanks.

[Dustin Lang]

Hi,

Fundamentally, the Astrometry.net code spends a lot of its time searching for matches of quadrangles in your image to quadrangles in the index files.  These are searches in random places in the shape space, so it jumps around the index files.  We do this using the Linux mmap() call, so if you have enough physical memory, then each time a block is requested from the index file, it gets cached in memory.  Eventually, the whole file is mapped into memory, and then subsequent searches should be much faster.  One thing to check in your setup would be, once you have solved a field, how long does re-solving it take?

If you can reduce the amount of sky (in RA,Dec and in scale) to search, then that memory requirement will be less.  For tiny fields like 6 arcmin x 4 arcmin, if you know roughly the RA,Dec of your fields, consider the "5200-series (LITE)" files (https://portal.nersc.gov/project/cosmo/temp/dstn/index-5200/LITE/), where the sky is split into 48 pieces.  If you provide solve-field with --ra, --dec, and --radius, then it will only search the sky tiles that overlap.  Similarly, if you know the image scale (--scale-low, --scale-high, --scale-units), that will restrict the number of index files that have to be searched.

You might also want to check how much time is spent detecting stars, vs searching.  Turn on solve-field -v to get some details on what it's doing.  The "--downsample 2" option may help.  And if you don't need them, "--uniformize 0" and "--no-remove-lines" will skip some steps.

cheers,

--dustin

--
You received this message because you are subscribed to the Google Groups "astrometry" group.
To unsubscribe from this group and stop receiving emails from it, send an email to astrometry+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/astrometry/cf2e1222-8335-47ce-96a0-655198911e91n%40googlegroups.com.

[Bryan]

I have used several PCs for local solvers.  I always set the --scale-low and -scale-high to something close to the actual image, i.e. scale-low  minus 15%, scale-high plus 15%.  I also always set --downsample 2.  I have tried --downsample 4, but the longer image processing for downsampling offsets the shorter solution time.

Windows 10 Pro

3.4 GHz Intel i3 with 8 GB RAM

Windows 10 Himw

1.8 GHz Intel i5 with 16 GB RAM

Windows 10 Pro

1.9 GHz Intel i7 with 16 GB RAM

On a 18 MP TIF file, all the above take 16 s.   I have tinkered with other parameters, including  --noplots, --guess-scale, and --depth.  NONE, except for --no-plots, have the impact that --scale-low and --scale-high have in reducing solve times (as Dustin has stated before).  --no-plots is also significant, esp for large files. 

Bryan

[Dustin Lang]

For large files, I often find that --plot-scale 0.5 or --plot-scale 0.25 makes more useful plots, and does it faster.

--

You received this message because you are subscribed to the Google Groups "astrometry" group.
To unsubscribe from this group and stop receiving emails from it, send an email to astrometry+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/astrometry/0370190e-df5c-4991-906f-8cc44adcfd8fn%40googlegroups.com.

[Bryan]

Thanks for the tip, Dustin!

One other observation:  I use short exposures for images to be solved, e.g. <5s, as opposed to using a typical exposure for processing purposes, e.g. 10-25 minutes.  While the solve time is solve time is roughly the same, perhaps a bit shorter due to fewer stars, the time savings due to reduction in exposure is significant.  Astrometry has no problem solving such short exposure images.

Bryan

[mlco...@gmail.com]

Hello Dustin,

Thank you for taking such care in answering my question. I will be away from my telescope for a couple weeks so what I can say now is limited.

Using the Kstars Ekos alignment module, I was trying to center an object known only by coordinates in the camera FOV, but it failed to converge to the 30 arcsecond radius target, resulting in solving the same approximate (within 3' or so) field over and over, but the time to plate solution stayed between 60 and 90 seconds.

By default the Ekos command line includes -ra, -dec, -radius, scale, and downsample options which I tweaked without any noticeable effect. The only queer thing was that the "Pixel Scale" in the logs varied around .39" -- strange because unbinned pixels should be .238" so downsampling by 2 should be .48" or so...

How does the 5200 series compare with the 4200 series?

-------------

Thanks for your data, Brian. I have ordered components for a mini-itx system with comparable specs to those you list to replace my little telescope-side SBC running at 1.5 MHz and 1 GB RAM.

[Dustin Lang]

The 5200 series is built from Gaia-DR2, while 4200-series is built from 2MASS.  I expect 5200 to be better for almost all purposes (except maybe infrared imagers).

That's odd about the unexpected pixel scale.  Note that adding "--downsample 2" to the solve-field command does *not* change the pixel scale of the WCS outputs, because the downsampling is done before source detection and then the source positions are scaled back up; basically it only affects how we detect sources, not the solving results.

cheers,

--dustin

To view this discussion on the web visit https://groups.google.com/d/msgid/astrometry/4405ef32-d774-44dd-85ef-ffd81bcce627n%40googlegroups.com.

[mlco...@gmail.com]

Dustin, what is the difference between the 5200 series under index-5200 and under index-5200/LITE? Thanks.

[Dustin Lang]

The LITE ones contain just what you need for astrometry solving.  The non-LITE ones also contain all the Gaia proper motions and parallaxes -- useful if you want to do more stuff with the matched stars.

To view this discussion on the web visit https://groups.google.com/d/msgid/astrometry/8a7a16da-ed43-490b-8097-ded61583fd68n%40googlegroups.com.