I understand that you refer to integration rather than registration, mainly because registering bias frames doesn't make too much sense.
From what you say I also assume that you are working with raw DSLR images. To integrate 100 of these images, the ImageIntegration process has to load all of them in memory at once (this is because DSLR raw formats don't support incremental file reading). Let's assume that your raw images are something like 4000x3000 pixels, and let's assume also that you are loading them as single-plane CFA data (no deBayering). ImageIntegration works with 32-bit floating point data internally. So 100 of these images require:
4000 x 3000 x 4 = 4.47... GiB
which is more than the size of your physical RAM. Since you are running a 64-bit operating system, you could complete the task using virtual memory on disk (although very slowly). However, you only have 1 GB of swap space. Consider that your Linux kernel needs at least 1 GB to run, plus some extra space to run vital services and processes, plus the space consumed by your X11 server and desktop manager, among other things. This means that your system is facing a severe out-of-memory condition.
When this happens, the
Linux Out-of-Memory (OOM) Killer starts killing processes, especially those processes that have caused an OOM condition. When a process is killed this way there's no room for elegance: it simply dies without any chance to fix anything.
So the problem here is twofold: your machine clearly lacks the necessary resources to accomplish this task, and your Linux swap partition is too small (not just for PixInsight; with just 4 GB of RAM, you should have at least 8 GB of swap space for a normal desktop Linux installation).
On a side note, if you pre-convert your DSLR raw frames to the FITS format, you should be able to integrate 100 bias frames. Take a look at the BatchFormatConversion script.
