freebsd exokernel

Fabio Kaminski fabiokaminski at gmail.com
Fri Jul 30 17:41:40 UTC 2010


Hi folks,

i know its a kind of off topic, but i think this is the perfect list for
this..

anyone here think a little bit like me, and like the exokernel idea?

the primary idea is to leverage only things like schedulling, and drivers to
kernel ring .. and downgrade things like VFS and MM to userland rings
as library.. so an aplication could optionally use those as libs abstracting
things generically.. (like a bicicle with wheels)...
and when you really need or want.. you can go into the bare metal and create
your own application abstraction..

imagine what it could represent in performance since the layers get
optimized and are not on top of other layers... and without the context
switch between user an kernel ring??

for a applicartion virtual machine like java(with its own schedulling, mm
and fs layers), or a database (fs and memory layers) or a virtualization
software..

if we write a database for instance and want to outperform disks, the actual
scenario is: or you invade the kernel of the OS and implement your
abstraction(you has to know all the sou rce of it)
and part you code in the userland :s , or you dont mess with the kernel at
all (its too impratical) and keep in the userland.. and everybody has to be
ruled by only one homogenic way to "see" things.. your application may have
luck.. this kernel abstraction its good for you.. but you may has not.. and
even if you can see the gold.. you cant advance any further..

the mit guys create one based on 98 (i think) openbsd, and they created a
web server that (now optional) tcp protocol where persisted on disk, so its
protocol agnostic, and can change
its communication wall in runtime..

sometimes im looking to where evething is going in technology, and we are
kind of stepping back.. putting more layers on top of others layers... and
slowing everything.. instead of getting it faster as it can..

i would like to share experience and what you think about this..

would it be a feasible project to borrow things from freebsd, and start a
project like this? anyone like this idea ??

anyway, just some thoughts for now..


More information about the freebsd-hackers mailing list