Re: VM cache policy change

[Richard Braun]

On Sat, Feb 21, 2015 at 04:02:00PM +0100, Justus Winter wrote:
> He mentions three problems preventing the patch to be merged as-is:

And I'll mention another one.

Assuming the system doesn't freeze because of a swap deadlock (and I
only name it that way because I've read it happened and it looks like
a deadlock), and assuming file system servers are scalable enough,
another dreadful problem is the size of the kernel heap.

More cache means more memory objects and more IPC ports, but the
kernel only uses a small fraction of its virtual space for kernel
objects. This means that the patch will cause kernel panics if it
caches lots of pages.

As GSoC 2015 is starting soon, I'll submit a project idea about
reworking memory managment to overcome this problem on 32-bits
machines (64-bits ones would be a lot less prone to it, although
they would benefit from improved memory managment as well).

-- 
Richard Braun