Ziad,
let me give it a try.
(I will be as simplistic as I can make it, so please, guru's, do not trip immediately over the simplifications!)
The basic concept:
A node has a certain, limited, amount of physical memory.
A number of processes are running, each with a certain amount of virtual memory at their disposal.
The amount that a process might be able to address is its Address Space (divided in Process Space and System Space).
Process space is 'private', system space is also available to other processes, be they system service, or Job logical name table, anything that is shared or sharable with anything other than the process itself).
At any time a certain amount of that address space is allowed to be present in physical memory, directly addressable by the process.
This is called the Working Set.
Somehow, a connection has to be made between a process's memory, and the actual physical location. This is done by having the link between virtual and physical address in a special table (on Alpha: a cascade of tables).
(Stepping past all fine-tuning details now.)
If a process expands its used process space beyond its working set, it first has to 'give up' other pages from the working set. The 'least-recently-used' pages get released first, because the more recent a page was accessed, the more chance it will be needed again soon.
Physically, of course, these pages stay where they are, but they are no longer pointed to from the Working Set.
Of course the OS will have to keep track of those pages, because THOSE are the pages that are now available to be overwritten for other use.
But: just MAYBE, the process WILL need a just-released page again! Now we can get that page from disk again, but if we still know where it is in memory, that would make it thousands of times faster to retrieve it!
So, that is what is done.
Now to your question:
if a page has to be released, there are two possibilities:
the page containes information that has once been gotten from disk, and its only use has been reading the info. So, the info at this page can simply be retrieved from disk again if necessary.
Upon release, the pointer to this page is move from the working set to the end of the free list.
(Any needed pages are taken from the head of the list, so the page is kept in memory as long as possible. It more or less functions as a cache).
If, however, the page HAS been modified, then there is a good chance that it will be needed again, WITH the modifications. And THAT is available nowhere else! So, it has to be saved. That is what the Pagefile is for.
But the same argument of speed reading from memory vs from disk as for the free list applies double here: not only may we have to READ again, we must also WRITE it first.
And if we need to give up pages because we want to address other pages, we certainly do not want to wait for that write to the pagefile!
That is what the Modified Page List is used for: any page that leaves the working set of a process and has been modified, goes to the Modified Page List. If needed again, it can be retrieved from there.
The modified Page List has a certain size, with certain tresholds. If it grows too big, the SWAPPER system function (on Vax: the SWAPPER process) kicks in, and writes a big chunk of the Modified page List to the Pagefile, and moves the pages to the free list.
Now, if you need such page again, it might still be on the MPL. Great, just 'page in; again (by only changing a pointer in some tables).Maybe it has already be written to the pagefile? It might still be on the Free list! Then, once again, update some pointers, but now also deallocate that page in the Pagefile.
And if it is no longer there, then an actual read IO from the Pagefile is needed.
So, what is the Modified Page List?
Basically, just a cache for pages that some process has modified, might need again, but can not hold in its own working set because of size limitations.
-- if these cocepts are quite clear, and you want to 'dive in' some more detail, then things like WorkingSet Adjustment, Image- Activation and -Rundown, Shared pages, Installed Images, process creation and deletion, those all modify these activities somewhat, but basically they are all different reasons to perform these same activities.
And of course, this text is only trying to globally describe the mechanisms. Any exactness will have to be taken from the above-quoted manuals. This is just intended to give the background idea!
Hope this helps.
Proost.
Have one on me.
Jan
Don't rust yours pelled jacker to fine doll missed aches.
