[SOLVED] How stable PAE really is (kernel 3.18) ?

zmau · 03-18-2016, 06:38 AM

Hi all,

I have a real time application on an x86 system.
The HW has 4 GB of RAM, but a large portion of it as not available due to BIOS behavior (a lot of addresses are "reserved").

So, there is an idea of compiling the kernel with

HIGHMEM=64GB
PAE=YES

The above configuration seems to be able to use all 4 GB of RAM (cat /proc/meminfo).

The questions are :

How stable is this configuration ?
I Know that in the past this configuration use to be treated as "not stable".
What is the effect on the performance of the system ?

Thanks
zmau

dijetlo · 03-19-2016, 10:23 PM

Most of the issues with PAE I can recall involved drivers that misbehaved when they were loaded in the higher addressing space, currently it's the standard memory mapping scheme used by most Linux distros so you can assume it's pretty stable.
The extended addressing does impact performance, the reads would tend to be slower simply because of the extended addressing, but you have to ask yourself, does it degrade performance more or less than being able to jam 64 gig of high speed ram into the machine improves performance.

syg00 · 03-19-2016, 10:45 PM

A PAE kernel will only work if the hardware is PAE capable.

There would have to be good reasons to be throwing money at a 32-bit hardware these days I would have thought. PAE used to work (mostly), but was a kludge until 64-bit was generally available.

zmau · 03-20-2016, 11:15 AM

Thanks all

syg00: There are good reasons for working in 32 bit mode.

I have another question, but I will open another thread for it.

zmau