1.6.1 ImageRegistration: hangs, can not be interrupted

[Nocturnal]

Hi,

I'm trying to register a downloaded HST image onto an image I took. The HST image covers only the central area of mine. It's sitting here burning up a few cores (fans howling) on iteration 16 but that doesn't seem to finish. Clicking cancel doesn't do anything, presumably because that only gets checked when it comes out of an iteration. So now I have to kill PI which means I have to reproduce it before I can grab a process icon and all that. I'll let you know when I get to that. I do notice that during the RANSAC operation it seems to cycle between using only 1 core or maybe 3. Definitely not all 8.

[Nocturnal]

Oh, it used up 4 GB of RAM. Sweet! Ok, a little excessive First time I saw my memory utilization pegged at 97%.

[Niall Saunders]

Time for 8GB?

With your machine's 8 cores (compared to my 4), but with only 4Gb RAM (compared to my 8), I wonder which machine actually ends up 'more capable' in the world of astro-image processing? I suspect that RAM wins - which is good, because RAM upgrades are usually cheap and therefore cost effective.

What are your thoughts?

Cheers,

[Juan Conejero]

Hey Sander,

Oh, it used up 4 GB of RAM

What's the size of your images (WxHxN and sample type)? SA doesn't hang; it's just very slow because it needs to do a lot of work on disk swap files. If you'd have the patience to wait (it seems eternal at the beginning, but once Windows organizes its swap file the process speeds up), it would complete the task.

That happens because you're using a 64-bit OS. If you were using a 32-bit OS you'd have had a nice "out of memory" error With 4 GB, anything exceeding 2 GB approx. requires disk swapping activity. When you reach the limit of your physical memory, everything becomes extremely slow.

As Niall says, time for an 8 GB upgrade!

[Juan Conejero]

I do notice that during the RANSAC operation it seems to cycle between using only 1 core or maybe 3. Definitely not all 8.

The RANSAC routine actually uses all the cores (with several parallel RANSAC threads, the probability to find the optimal registration model gets multiplied by the number of threads). It is the star matching routine that implements algorithms which cannot be parallelized so efficiently. There are parts that use all the cores but they are so brief that performance monitors interpret them as narrow peaks of partial CPU usage (that's why you're seeing only 3 cores used). The performance monitor on Windows (is it part of the Task Manager if I remember well?) isn't reporting your CPU usage in real time. The same happens with the 'top' command on Linux for example.

[Nocturnal]

I have 6 GB, 4 of which was used by PI. The reference image is a 3000x2000 color image, the image to be registered is 3500x3500 mono. To 'help' I flipped and mirrored the image to make it match closer.

Anyway, I tried again. With a preview roughly in the center of the reference image a solution was found in 23 seconds. Without the preview it gave up after 17 iterations but didn't print how long it had taken so I don't know. When I saw the problem before I had unchecked the brightness relations and scale differences checkboxes. That led to far more potential star matches I assume.