virtual memory 4518

Andy Glew

You're certainly in a far better position to answer that than I am, but my read is that PAE really has minimal impact on the ISA and implementation, plus the three level page table extends naturally into a bigger virtual address space (as it obviously has with x86-64). So unless you wanted to do something radically different than the hardware walked x86 page tables, you'd have ended up with something similar to the PAE scheme (perhaps without above-the-line PSE support) anyway.

The impact on the OS vendors is a different story. One the one hand adding PAE support to the OS is clearly less work and less disruptive than doing a 64 bit port, on the other hand, it's all work that gets thrown away a few years down the road, although it might happen anyway (see comment below).

Heck, right now I'm working on an application that will get AWE (the Windows API to PAE) support for storing data in phase 2. When this goes 64 bit somewhere down the road, the option to allocate a memory pool in AWE memory will simply get disabled. Because we've been able to plan ahead, all of this is isolated in a couple of clbuttes.

a major blunder in the design of two instructions (unfortunately the two subroutine call instructions - which stored additional data in the high byte of the register getting the return address), and the decision to define address space wrap-around at 2**24 (there were a couple of other areas, but they would have been minor changes). The OS writers, who were under considerable memory space pressure made this worse by storing three byte addresses all over the place, including application visible structures.

I think you'd be hard pressed to call either the PDP-11 to VAX or VAX to Alpha transitions "easy." Although in the Alpha case DEC managed to limit themselves to doing a pure Alpha port of VMS and then hanging a VMS-VAX emulator on the side.

zOS transition). Basically the first few versions of the OS had limited support for 64 bit mode, very limited at first, but over time more and more functionality became available in 64 bit mode. In the case of AIX and zOS, the first versions of 64 bit support allowed little more than paging, and a very primitive ability for applications to run above the line - in most cases OS API calls had to happen from below the line. So more than a bit like a PAE aware OS. Over time the full interface became available to 64 bit applications.

Hard to say, but the biggest systems at the moment are pushing towards 50 bit real addresses (Altix, for example, can do 128TB). That may imply a need for virtual address spaces bigger than 64 bits in a couple of decades, less for shared memory clusters. Although none of us are likely to have to worry about the 128-to-whatever transition (but the great-grand-kids will undoubtedly curse our lack of foresight).