The following thoughts have been on my mind for quite some time,
I've read
Geoff Chappell's blog and atm am about halfway through
Windows Internals 6th by Russunovich, and it's brilliant, I really recommend it.
Mark used to piss Microsoft right off so I have great respect for him.
Just to set the record straight, a pagefile is "not required" in Windows and never has been. Regarding applications which "need" a pagefile, applications don't know even the pagefile exists.
^This is true. You will find the above in Mark R's book so feel free to look it up.
Below is mostly opinion, but with some truths thrown in.
Due to events which have happened over the last few years, and some personal observations I've concluded MS are not being totally honest with us neither are Intel. And AMD has been getting the short end.
Many MSDN blogs and TN blogs are tailored to suit Microsoft interests, and tbh - Intel.
The reality:
https://www.raymond.cc/blog/make-windows-7-and-vista-32-bit-x86-support-more-than-4gb-memory/
https://blogs.msdn.microsoft.com/oldnewthing/20090706-00/?p=17623/
https://blogs.msdn.microsoft.com/oldnewthing/20040822-00/?p=38093/
http://www.geoffchappell.com/notes/windows/license/memory.htm
******
PAE and the PF is designed for x86 32 systems with 2GB sliced of for the kernel and a total of 2GB or less left for everything else. Or any platform with barely any ram. Like all Intel until around ~2008.
AMD64 and IA64 platforms can both support up to 2TB of RAM and likely more expert MS hasn't tested past 2TB.
TBh I don't understand the point of a PF because if you can map section objects from a hard drive what is the difference between that and a pagefile? Why use one, you're mapping the same pages twice.
Maybe someone else who knows WMM can throw in 2c?
Also, has anyone ever been able to make sense of "Windows Memory Limitations" on MSDN?
I've been in the industry for nearly 20 years and even I don't understand them. Can you?
An example:
https://msdn.microsoft.com/en-nz/library/windows/desktop/aa366778(v=vs.85).aspx
If those are MS Technical References the authors need to find a new line of work.
Wasn't till I read Geoff's Chappel's blog cited above I realised hey, I'm really not stupid after all!
Geoff said it years ago; the MS articles on memory limits and PAE are deliberately ambiguous.
Why can't MSDN seem to make it clear?
During 2004-2008 Intel did not have a single chipset which could handle over 4GB, except Xeon which was specifically designed for servers and astronomically expensive. In 2004 AMD released Athlon64 which was both cheaper than Xeon and could support about as much RAM as you could throw at it. At the same time XP users found they could install 8GB or RAM on their shiny new nForce4 board and the OS would use it.
This is because up until then XP had always been shipped with the PAE kernel enabled, the Client and Server versions of XP are exactly the same, except for the PAE kernel, it ships with both.
Obviously this did not sit well with Microsoft, because hell if end users could now access all that RAM on XP why would why anyone bother upgrading to Vista? Or XP Enterprise/Server versions.
It also did not sit well with Intel who at the time were busy paying of OEM vendors like Dell and HP in attempt to destroy AMD for good.
Consequently along came SP2, low and behold after updating end users running XP Client on Xeon or AMD64 could no longer use over 4GB even though 10 mins before they had 6-8GB installed.
MS eventually made some vague comments that disabling PAE was to "prevent crashes from buggy drivers which expect 64 bit addresses" and in the clients best interest.
The curious part was although MS were concerned by this, at the same time they were busy promoting 64bit Vista as the second coming of Christ. No, no, nothing contradictory aboiut that.
Another point I'd like to make is regarding the 2008 "Pushing windows memory limits" Technet blog, which I'm convinced (I hope) Russunovich didn't write since it has Alex Ionescu's name on the meminfo screenshot.
*****
There is a paragraph about halfway down the page with a screenshot of meminfo and a claim along the lines of "look see in 64bit Vista all my 4GB of RAM shows up".
Directly below it for comparison is another meminfo screenshot with "I bought a gaming system with 2GB of VRAM but because it came with 32bit Vista I have only 2-5 or 3.5GB of RAM available." Whatever it was.
I'm sorry, but that claim is bullshit.
The "Boutique Gaming system" with "32bit Vista" is in fact an Intel 945 laptop with onboard 8400 GPU and only supports 4GB RAM maximum.
In the top "Vista 64" screenshot the command prompt line
CLEARLY STATES AMD64.
It's an AMD64 chipset with remapping, obviously it will support over 4GB. The extra 500MB RAM has nothing whatsoever to do with the OS. Is this misleading the public? You decide.
I posted a comment on Technet last week questioning the screenshot, and it was promptly deleted.
Some time later all the "Anonymous" spam posts appeared, which you will see below the main article.
P.S: Russunovich never claimed "the PF shouldn't be disabled no matter how much RAM is installed" he said "it can be disabled but it's not recommended".
That was written nearly 10years ago in 2008.
What he didn't - or probably couldn't - say was "it's not recommended on x86 IA32 systems which have sweet FA ram available.
Windows Internals 6th Edition states Windows can run without a pagefile and the key to performance is physical memory. You have to read between the lines sometimes,
I think with 2TB available we can assume it's safe. but ofc check support with your vendor.
Thanks, I'm interested in any comments.
Please no flaming.
