Saturday, April 5, 2008

Further Dissection Of Paging And Swapping On Linux And Unix

Hey again,

Believe it or not, I actually got a few emails about our previous post on paging and swapping in Linux or Unix because it wasn't specific enough ;) While I can certainly understand the frustration, I was hoping to explain the main differences as completely and concisely as possible. I seem to have faltered a bit on each front: I glossed over two specifics which, I agree, deserve some attention and I wrote yet another novel ;)

With that in mind (and with a prayer that my fingers won't type any more than they have to ;) I'd like to address, and/or clarify, the level of depth I didn't descend to in my last post on the difference between paging and swapping and write about the difference between paging and swapping (The redundancy was intentional and any resulting confusion is expected, given the topic at hand and my writing style ;)

As I mentioned previously, the terms paging and swapping are used almost interchangeably these days. Some industry manuals will actually talk about "swapping out pages" which seems to be contradictory and, theoretically, impossible if swapping and paging are two separate concepts with distinct and unique definitions. This is where language and implied meaning become a barrier to actual definition. And, all the more reason to clarify this one last bit of the puzzle.

And here they come. The extra clarifications...

1. Difference in resident virtual memory management with paging and swapping.
When a system swaps a program, or process, it guarantees that it is resident (on disk or in memory) before it schedules it for execution, and will often hold onto the mapped resources reserved for that process from the time the process requests them until it notifies the scheduler that it is complete. When a system pages during the execution of a program, or process, there isn't any such direct correlation. You don't necessarily know (without specifically checking) how much of a process's virtual memory is resident or whether the process is entirely able to be scheduled for execution at the time paging begins. Pages can be selectively grabbed from a process (out of mapped physical memory) and never returned, unless re-requested by the process when it looks for the memory, can't find it and generates a page fault.

Phew... That's one down. Hopefully this isn't just becoming more confusing :)

2. More specific definition of paging and swapping with regard to page ins/outs and swap ins/outs.
In this instance, swapping specifically refers only to the transfer of memory pages from physical memory to dedicated swap devices or swap disk (on most systems this is now referred to as swapfs - or a unique swap filesystem) and vice versa. Paging, on the other hand, refers to the transfer of memory pages from physical memory to disk (regular disk or swap disk) and vice versa. So, really, the major difference is that swapping is limited to only transferring memory pages back and forth between the physical memory and a dedicated swap device or filesystem, while paging can transfer between physical memory and any sort of disk device.

Hopefully, we've reached a sufficient amount of explanation at this point, and this thing won't turn into the monster I'd hoped it wouldn't become ;)

Thank you, everyone who wrote in, for your helpful input. As many folks have also noted, we don't have comments set up on this blog (because of issues with "comment spam" and not wanting to get shut down). If you ever want to leave a comment, or an objection, we welcome you to email us directly at our most often-check email address or, if you have a lot to say (or want to attach video, etc), sign up (for free) on our sister Linux and Unix Menagerie Forum, and we'll get your remarks there, as well. We do our best to reply, personally, to everyone who takes the time to write us. So far, I think we're still batting a thousand in that regard ;)

Best wishes,

, Mike