Dragonfly BSD kernel model

Matthew Dillon dillon at apollo.backplane.com
Sun May 22 09:38:07 PDT 2005


:
:Robert Garrett wrote:
:> so really none of the existing models really applies.. 
:> this is my understanding, I am quite sure i'll be corrected if wrong.
:
:That's where the term 'hybrid' comes in. Basically, DFly is monolithic 
:due to it's ancestry, with microkernel features added (the message 
:passing structure et al). The theory is, the current kernel could be 
:split up into independent userspace servers pretty easily because the 
:way they communicate is transparent: the message passing interface.
:
:Cheers,
:		-- Thomas E. Spanjaard
:		   tgen at xxxxxxxxxxxxx

    Right on both accounts.  In theory we will be able to export portions
    of the kernel interface to userland.  In reality this is going to be
    quite difficult due to its ancestory.  But eventually it will be 
    possible.  The only thing preventing e.g. VFS from being exported now
    is the I/O infrastrucuture.

    Probably a better way to think about it is not so much running 
    individual subsystems in userland as running them in their own VM
    context.

    It is not actually desireable to do this within the context of a 
    single machine, it's simply too inefficient.  But in the context of
    a CLUSTERED topology this capability becomes very desireable.

					-Matt
					Matthew Dillon 
					<dillon at xxxxxxxxxxxxx>





More information about the Users mailing list