Re: Memory management for Gsoc

[Samuel Thibault]

Maksym Planeta, le Wed 30 Mar 2011 16:51:13 +0300, a écrit :
> I'd like to participate in Gsoc this summer and I'm interested in memory
> management. I was looking through the code, involved with memory
> management, and concluded that kalloc subsystem can be improved.

Ideally we should try to focus on parts which do impact performance
most. I don't believe kalloc is very critical for performance (though it
might for e.g. port management), ext2fs paging is probably much more of
concern.

> I've tried to understand virtual memory in Linux and I'm investigating
> virtual memory management in Hurd. I have theoretical knowledge of
> virtual memory of x86 and x86_64 architectures.

You'll have to learn a bit about how pagers work in GNU/Hurd, since
that's quite more involved than on Linux.

> In gsoc project ideas for vm tuning was written:
> >The goal of this project is to bring the virtual memory management in
> >Hurd/Mach closer to that of modern mainstream kernels (Linux, FreeBSD),
> >by comparing the implementation to other systems, implementing any
> >worthwhile improvements, and general optimization/tuning. It requires
> >very good understanding of the Mach VM, and virtual memory in general.
> 
> and in http://savannah.gnu.org/task/?5489:
> >We need a new design to the Virtual Memory system, and Mach based ones
> >are not a good idea for microkernels.
> 
> So, vm system should be redesigned.

Ideally, yes.  I however don't believe it'll be feasible to do it during
the timeslice of a GSoC project.

> Can you advise me what way it can be redesigned, if you have such
> ideas. And can it be part of Gsoc project?

We probably need Sergio's output on this.

Samuel