1 |
On Tue, Oct 08 2013, Alan McKinnon wrote: |
2 |
|
3 |
> That is correct, with 3G physica RAM, you will not benefit from using |
4 |
> PAE at all. I don't think it interferes with anything if you do have it, |
5 |
> I recall a time when RedHat shipped 32 bit kernels that were PAE-enabled. |
6 |
> |
7 |
> Briefly, the way it works is that the kernel assigns blocks of memory to |
8 |
> different processes. So a single process can still only access 4G of |
9 |
> memory, but two different process don't anymore have to address the same |
10 |
> 4G of memory like you must do without PAE. But you still don't get to |
11 |
> give your 32 bit database more than 4g of RAM |
12 |
|
13 |
Agreed. Virtual addresses refer to those in the program (really |
14 |
process). Physical addresses address refer to those in the hardware |
15 |
(i.e. addresses in the RAM itself). To have a single process able to |
16 |
access extra memory would be to increase the *virtual* address range. |
17 |
PAE (*physical* address extension) enables more RAM to be accessed (by |
18 |
the hardware not by a single process), but does not increase the virtual |
19 |
address range. |
20 |
|
21 |
When pdp-11s added I and D space, that increased the virtual address |
22 |
range by a factor of two. The I/D bit (instruction/data) was |
23 |
essentially an extra bit of virtual address. |
24 |
|
25 |
allan |