New brainfart for threaded VFS and data passing between threads.

Matthew Dillon dillon at apollo.backplane.com
Wed Mar 31 12:17:45 PST 2004


:
:Matthew Dillon wrote:
:>     The recent PIPE work adapted from Alan Cox's work in FreeBSD-5 has really
:>     lit a fire under my seat.  It's amazing how such a simple concept can
:>     change the world as we know it :-)
:
:Out of curiousity, have you considered some of the ideas from exokernel
:designs?

    Not really.  As far as I can tell, all exokernel does to get its 
    stated performance improvement is to vertically integrate everything,
    e.g. such as the web server and TCP protocol stack, into what is
    essentially a single execution context.

    Well, I stopped doing that the day I set aside my Amiga for a PC running
    Linux, and you won't see me going back.  I don't see how such a system
    could ever be used in a complex production environment.... it would be
    nearly impossible to track down and fix bugs.

    I will be frank... I could integrate a TCP stack into a web server and
    fast-path the most common connection/data/disconnect situations, but the
    resulting mess would be unsupportable.  It might look good in a one-off
    performance benchmark but the cost of trying to maintain such a beast
    over time would be horrendous.  The last embedded hardware project I wrote
    was based on a 10MHz 68000, and even the little UDP stack I wrote for
    that was message-based and protected (via a simple protection map tacked
    on to the memory subsystem) rather then vertically integrated.  So,
    no, exokernel does not appeal to me at all.

					-Matt
					Matthew Dillon 
					<dillon at xxxxxxxxxxxxx>





More information about the Kernel mailing list