Processing time

Are you using our latest beta version already? Beta 4 can be downloaded via the links on top of the forum. Speeds have been improved since 1.082. But, APP does perform way more complicated calibration calculations than DSS for instance. This is where the differences in speed also come from, we chose to focus on quality and algorithms that enable users to process a lot of different and complex datasets, we do continue to improve on speed, but it is simply more complex.

No not using Beta I always thought that meant experimental rather than reliable. I gave up on that first attempt after 5 hours. I am trying again with about 25% of that dataset and will then switch to Beta thanks.

Yes it is a test, but beta 4 is very stable actually so that shouldn't be an issue. 5 hours seems a bit too long though, so when you checked it with beta 4, we can see what the cause might be. You are processing on an internal drive right?

@vincent-mod I would have installed to my 1Tb normal HD and windows is on an SSD as that is not big enough for all my software. Am I going to need to reinstall on the SSD?

The website isn't showing beta just the one option to doqwnload APP for windows.

@woldsweather If it's an external HD, that will either cause issues or be much slower yes. APP is using your drive quite a bit during processing and if it's limited that will slow it down as well. An external SSD will work better, when connected to a fast connection.

Let me give you a direct link: APP for Windows, beta4

@vincent-mod Ah thanks I googled and found ikt. Have uninstalled the one I have and installed this on the C:  SSD drive and its calibrating but its my bedtime so thanks for that and I'll report back in the morning!

Have moved on to integrate this morning and its taking about 10 minutes and a lot of fan noise to move up 1%. Is that OK?

Its on INtegrate and after 3 hours is at 24%. Surely it shouldn't take this long?

2000 and still on 81% of Integration, something can't be right.

Hi Trevor @woldsweather,

What exactly are you doing? How many frames are you processing from which camera?

Are you using defaults in all the steps of 2) to 6) ?

Mabula

500 .cr2 raw frames from my Canon 6D MK2 around 20 darks, flats and bias. Yes all the defaults.

So you installed APP itself on your SSD, but where is your working directory located? Besides that, 500 is quite a lot, but I don't think it should take all day.

The folder with the Darks fl;asts etc is on my HD. Do I need to move them to the SSD?

If your HD is external, then yes, move them to the SSD as well. Basically all files APP works on are preferred to be on a SSD for processing speed. The working directory you select when starting APP also should be on a local disk.

I have to agree.  there are data sets that cause the software to basically go into an idle mode.  I just stopped a process where i had taken all my data for Orion's nebula (some of which i did have previous FITS files, but chose to try them from the raw files), and loaded the different data in to different sessions, combining data from different evenings where i used exactly the same optics, including filters, some shots with a 560mm telescope, some with a 600mm telephoto lens, all taken on the same 6D.  i put 1400 images into the process and got through normalization and selected the top 35% for stacking and had just under 500 pictures.  I selected the smallest area possible around the target (Orion and Running man) for the crop, selected 2nd degree LNC, 3 iterations with integration for each session and all.  After 30 hours on a Dell Xenon processor with 32 gigs of ram and a 4TB hard drive, I stopped the process.  I then selected 25% of the images to stack and after 20 hours stopped the process. I have just changed it to just integrate with no per session integration, only 20% (total of 280 files, which i have been able to integrate successfully before in a single session) and will let it run over night to see if it will do it.  The largest i've ever done before was about 500 images and that takes quite a while.  I"m on version 1.082. 

If it doesn't make it through, i'll process each session individually and then integrate each set of FITS files instead, as you suggested above.  I didn't think to look up processing time issues before.  I'll have to download latest version when i renew my subscription.  

Edit:  additional information.  In addition to xenon processor, 32 gig ram, the laptop has internal 1TB SSD hard drive, and connected by USB3.0 an external 4TB SSD hard drive.

the 2 previous stalled at 10%, sometime after initiation LNC.  The size of the processing was for the first run at 35% stack 667 gigs, 2nd run at 25% stack 450 gigs (roughly), and current process at 20% stack but single integration is 333.5gigs. 

As a follow up.  The process has been running since 6pm yesterday (4/14).  It is now at 28% and seems to have stalled in the same place, the LNC phase during integration (set to only perform LNC on the initial normalization and registration, not on final integration)  It started the LNC process at 10PM yesterday.   It has now been running the integration for 10.5 hours.  I will give it until 6PM today (a full 24 hours) before i stop the process again.  If it completes, i'll update this post.

@gsxrraycer In the current version of APP, LNC runs as a single thread and therefore is very slow when a large number of files are involved. Mabula is working on a fix to make it multi-threaded so in the next release that should be much faster. Sorry for the trouble!

@wvreeven is a final stack of 290 images considered large?  or is the problem that i started with 1400 images and after the various pre-processing stages only 290 images are being stacked.  I've stacked more pictures but in one session and without trying to do multiple sets of integrations, and that also took at least 12 hours total.  

at least i know now that this is normal.  I probably shouldn't have stopped the 30 hour or 20 hour processes since they might have completed.  

I'll definitely move to doing each session individually then combining at the FITS file stage.  But, if this is the optimal way to do large data set processing some discussion of that might be useful to newer users so we don't load to much to create these long jobs, or use multiple sessions per process if that's the issue.

Hi all @gsxrraycer, @woldsweather,

I think most of the processing delays indeed are caused by the current LNC algorithm i.c.w. more than 100 frames.

LNC in it's current form is a single thread calculation and requires an increasing amount of RAM and processing time with

• higher LNC degree (more parameters need to be calculated)
• larger number of frames

The situation is such that the calculation for 200 frames with LNC will take much more than double the time of the same LNC degree for 100 frames, so it does not scale linearly.

Luckily, I have good news on this front on multiple levels 🙂

1. The next APP stable version will run on a new development platform with much more efficient data throughput, so this in itself will help big time with processing large datasets. You can expect speed increase from 20-60% depending on hardware as well. I know from testing that especially strong computers will benefit a lot! So computers with a large amount of RAM and CPU Threads will see big performance increases.
2. We should be able to implement a new LNC algorithm, let's us say LNC 2.0 which will be able to run multi-threaded. So then the long lasting LNC task should be reduced big time ! Furthermore, I should also be able to implement the new LNC algorithm in such a way that it uses less RAM, so it will take less overhead and than it can speed up also from that.

We are very close to release a first beta of this new APP version where you will already benefit from 1).

Step 2) is probably part of a next beta afterwards.

All in all, an integration of 1000 frames in the new APP with LNC should be much faster ! I will do some benchmarks with graphs to visualize the difference with the old APP version 😉

Mabula

Thanks for the update and information on the development path of APP.

I look forward to trying the newer versions when they are released.

i shut the process down, downloaded the newest version and then tried stacking using the FITS files for the different sessions.  I was able to get one integration, but then when i went back and selected a different reference, renormalized etc, and then during the integration it gave me an error message saying something went wrong in registration.  I tried to go back using the original reference, and then it gave me the same error message.  

I'm completely baffled, it's like sometimes things work the right way and then redoing it the same way gives completely different results or errors.

i'm gonna take a break, watch a few of the tutorials and then regroup and try again.

@gsxrraycer Out of curiosity, why do you want to change the reference frame? APP selects the best one based on internal algorithms and in general this isn't needed.

Also, what error do you get? If you tell us then possibly we are able to help you get that solved. 

@mabula-admin that’s great news Mabula! I recently began imaging with a QHY268C CMOS camera, which is 26 MPixeks per sub. I also went to lots of shorter subs. I found processing 150-300 lights along with darks, flats and biases, was taking 8-10 hours even with a laptop workstation, which although 5 years old, was equipped with Quad core Xeon processor with 64GB memory and SSD. Gave up using LNC altogether. 

Since I really want to continue with APP and other compute-intensive applications, I decided to go ahead and upgrade and got a desktop based on Threadripper 3 24-core processor with 128 GB RAM. Uses high speed SSD boot drive for programs and working directory and a 10 Terabyte HDD for image storage. 

Very pleased so far. I ran a project with 268 lights, 50 biases, 25 darks and 26 flats - 34 GB of raw data - from start to integrated image in 2 hours and 50 minutes. That was with leaving all the raw files on the HDD  But still don’t use LNC 😂. 

Thanks again for your hard work and really looking forward to the updates!

Wayne