Local Astrometry server slow


#1

Hi, this is my first post on this form, I’m just getting started with SGP, so please bear with me.

I’m trying to blind solve a saved image file, if I use the online server it does it in about 14 seconds. If I use a local server, it takes forever, approx 250 seconds. I assume I’m doing something wrong.

Thanks for any tips, and btw, so far I love what SGP has to offer.

Huw


#2

Huw,

Maybe you could post the problematic .fit image somewhere so we can look at it?

Thanks,
Andy


#3

Thanks, but it’s not the file that’s the problem, I’ve tried a few, it’s much more likely the steam powered laptop I’m using. Didn’t think of the amount of calculating it’s having to do.

Huw


#4

Huw,

Your PC processing power will definitely be a factor, but there may be some settings that can speed up the solves so it may still be worthwhile posting a .fit file for others to play around with.

You should confirm that your image scale setting in SGP is accurate (Camera tab in the Control Panel.) If that is off, it will really make the the local solver take longer. Another tip to speed things up is to remove the index files that are below your image scale. You can look in the ansvr log (Start Menu => Astrometry.net local solver => Watch ansvr Log) to see which index files are giving the solution, then remove index files for smaller scales. The smaller scale files are much larger and there is no point having the solver search those indexes for a solution.

Andy


#5

Thanks again Andy. The number of indexes I’m running is likely to be a large part of the problem. Trouble is,I image at quite a few different scales, but for the solves I’ve done so far I’ve filled the image scale box, this should limit the search, does it?

Huw


#6

Hi Huw,

The steam powered laptop is the most likely cause, I set up a few local servers for people in the days before the installer and one of them found the internet version solved quicker as the laptop could not cope well.

An expensive but worth is consideration is to get a copy of Pinpoint and use ANSVR as a fallback which is what I do now. You can get a trial of Pinpoint to see if you think the gain is worth it.

Trevor


#7

Yes, Trevor raises a very good point. I would add that Elbrus is also a very good choice. You can setup Elbrus or Pinpoint as the primary solver, then use local Astrometry.net as the fallback. Personally I use Elbrus with ANSVR as the fallback and the combination has always worked great for me. Jared and Ken have an Elbrus installer on the MSS downloads page.
Andy


#8

How old is your PC? My main imaging PC in my obs is a fairly old dual core laptop. Certainly not fast by today’s standards. When I was imaging with my Qhy9 I had ANSVR solves down to about 8 seconds through tuning the settings. This was also with a known sampling and a 5% error. If I were to do a blind solve this number would increase.

Now that I have an STL-11000 my solves take about 2 minutes on the same machine. And for that matter they take about 90 seconds on my development machine which is an i7 w/32gigs of ram and an SSD. I haven’t spent a ton of time looking into why this is happening as I now use ANSVR strictly as a fail over.

Jared


#9

Thanks Jared. Yes, the camera is the other problem, I’m also using an 11002 based camera, and my laptop is a little netbook with an intel atom 1.6G processor (windows experience rating 3.3 !!). Also, as I image at various scales , I have loaded quite a few indexes.
Taken all together, it’s probably not a wonder that solves take a while.

Huw


#10

Andy,
I have recently changed cameras and I’ve noticed that my blind solves take considerably longer than they used to. I moved from a KAF-8300 (scale 1.21) to a Sony-694 (scale 1.02). My solves used to take 10-15sec but now take 60-70sec. I have the scale error estimate set at 5 and downsample 2. This is on a fast i7 computer. I have removed the indexes I don’t need. Is this to be expected because of the smaller image scale?

I also use Elbrus but it seems as though Elbrus has a harder time solving with the new camera. 50% of the time it reverts to ansvr, so I’d like to try and get my solve time down.


#11

Joel, it sounds like you have tried the most likely things – scale estimate, downsampling, and removing unnecessary indexes. Maybe you could post a sub or two for us to try solving?
One of my setups is 0.55"/px and it does seem to take a bit longer on average than with my 1.7"/px gear for example, but not as dramatic a difference as you are seeing.
Andy


#12

Here’s the image I was solving:


Solved image scale is 1.013.


#13

Hi Joel,

Just for your info:
I have just run it on my computer…
First tried Elbrus…solved in no more than 3 seconds (57 sides). Then tried blind solve with local astrometry net…it took 40 seconds. I did it a second time with a blind solve and it took 9.5 seconds…I forgot to check the numbers before & after to see did the 1st blind solve change the info in the header (which I think it does ???).
The final (9.5 sec solve) shows the following:
RA Hint 05:22:31
Dec Hint 33° 21’ 37"
Angle 350
Scale 1.013


#14

Thanks Kinch. Your solve info is exactly the same as I get. When Elbrus solves is does so in 2sec. But I’m at a loss for why ansvr now takes so long. On my OBS computer (which is slower) it takes 2min to solve. What are your settings in ansvr (scale estimate, downsample)?


#15

Scale estimate: 5%
Downsample: Factor : 2
Epoch : J2000


#16

Well that’s what I use too. I’m baffled.


#17

Now let me see if this makes sense…I blind solved a very widefield image taken with a 55mm lens with NO hint information at all (no dec, ra, scale or angle) and it solves in 15sec. This is at an image scale of about 19.6. Then I go and try to solve the above linked image and it takes 70sec. hmmm…


#18

It took me just 9.5 sec to solve blind Joel’s file on the first try.
I used a Surface Pro 3 with i5 processor.
I’m wondering the disk speed is important?
This laptop has a SSD.

Scale estimate: 4%
Downsample factor : 2
Epoch: JNow


#19

I also have an SSD in my computer.
I have discovered something though. Up to this point I was doing all my solves within the target settings dialogue, under “reference data.” With almost all of the images in that same data set solves would consistently take 65-70sec.

However, when I actually open the image in SGP, right click on it and plate solve that way, the solves only took 6-8sec!

This is fairly consistent. So what’s different between solving under target settings/reference data and opening the image and plate solving?

The issue that prompted this question is that when running a sequence I noticed that “center now” and meridian flips etc that use plate solving were taking very much longer than they used to, and a couple of times it even timed out after 480sec. So again I’m wondering if there is a different path or way that SGP handles solves from different starting points?


#20

Joel,

Could you post an ansvr log file with both a slow and a fast solve?

You can find the log in c:\Users\USERNAME\AppData\Local\cygwin_ansvr\var\tmp\platesolve\ansvr_DATE-TIME.log

Andy


www.mainsequencesoftware.com