Welcome back to the video series on virtual memory based on my article on the same topic. It is also available in ebook format on Gumroad (pdf/epub) and Amazon Kindle, if you prefer those formats.
So far in this video series we have covered the foundational pieces behind virtual memory:
From here on, the next set of topics is going to be more advanced and important for debugging memory issues in real-world systems. The topic of this video is demand paging, which is central to how memory allocation works inside the kernel.
Demand paging simply means that when you ask the kernel for more memory, it reserves a virtual address range for you but does not perform physical memory allocation until you actually need it. This can lead to tricky situations and unpredictable system performance if you do not design your systems keeping this in mind.
The famous OOM killer that appears and kills your processes is also a result of demand paging, because the kernel may end up overcommitting more memory than what is physically available, and ultimately it has to kill a process to free up some memory to recover the system.
In this 45-minute video, we cover in depth what demand paging is, what its benefits are, and how it works inside the Linux kernel. And if you watch the full video, you would also get an idea of how the kernel handles page faults for various situations, such as when the page has been swapped, illegal address access, accessing file-backed pages, accessing an unmapped, copy-on-write fault, etc.
In the next follow-up video, I plan to do a hands-on demo of demand paging in action. We will use mmap and inspect the virtual memory of the process to see that merely calling mmap does not cause a physical page allocation, but accessing a page does. Wait for the next video!
