Well it makes the kernel not able to make any decisions at all lol
No, it doesn't. All I/O is done via paging and the kernel can still evict unmodified pages from memory, which it does with or without a pagefile. All you've done is restrict it's ability to put modified pages into the pagefile, paging to normal files, binaries, shared libraries, etc will still happen.
From what I remember about XP's memory management, it's very proactive about paging, and would page out as much as it could including the important parts of the kernel just based on inactivity rather than necessity of RAM
No, the whole thing is based on LRU lists and while it may proactively put some modified pages into the pagefile that haven't been used in a long time it's not evicting them from memory until necessary and that's generally a good thing. It's not like XP is looking to keep memory free for no good reason.
I know in Linux you can very specifically set the behavior of the paging to determine how proactive it is.
A bit. You can change /proc/sys/vm/swappiness if you want, but that's directly related to the swap space usage and not paging in general.
The advantage of proactive paging is the RAM is ready to use right when you need it, the disadvantage of proactive paging is things you may still want in RAM get pushed out by the whim of the OS because it decided it was inactive too long. I prefer non-proactive paging because have more RAM than all my apps will need (so none of them should be paged out proactively because I have nothing else I plan on loading, even if I haven't alt-tabbed to it in an hour).
You'd be right if it worked like that.
But at 8GB an XP system shouldn't even have to make decisions about paging.. it can easily keep every part of the kernel (as you said those parts are small) and all running apps in ram without paging anything unless you're running some monster apps.
Exactly, so in the best case scenario none of your tweaks do anything.
I know Vista and Win 7 have a much smarter memory management system worked out so I wouldn't bother in those OS's.
The core memory management is virtually the same, but they also include userland portions in SuperFetch that watch usage and preload things they think you might want to use soon.
Also using the RAMdisk for page file may be a workaround for the default 2GB user mode app limit since Windows uses PAE to redirect memory to the virtual ram (page file), and should give the app access to more memory through that. But I doubt OP has a >2GB app..
Not at all, where you put the pagefile has absolutely zero affect on the VM layout of the kernel or any userland processes. The only workaround for the 2G per-process VM limit is to boot with /3GB and make sure any binaries that need more VM are marked LargeAddressAware.
And PAE has nothing to do with the pagefile or the VM layout. All it does is add another level to the pagetables so that the kernel can address >32-bits of physical memory. Virtual memory is unaffected.
You really should get a copy of Inside Windows or Understanding the Linux kernel and read the chapters on memory management. The details of the implementations differ, but the core ideas about virtual memory are the same.