Re vkernel and all

Matthew Dillon dillon at apollo.backplane.com
Fri Jan 12 11:04:18 PST 2007


:
:Regarding the vkernel work and other exciting things happening: how
:(if at all) will this affect the scope of the application base
:supported by dragonfly?
:---

    My original reasons for wanting to have a working virtual kernel
    have not changed.  For me, having a working vkernel reduces the
    engineering cycle time for testing new kernel code from 5 minutes
    to 5 seconds.   I decided it was a must-have development tool to
    be able to make any progress on my upcoming clustering work.

    What other uses are there?  Well, any kernel development done by anyone
    on any subsystem that does not directly involve hardware will benefit
    *hugely* from the virtual kernel feature.  Filesystems, Networking,
    Clustering, etc.  The linux folks apparently have used the linux UML
    primarily for kernel development.

    For real world apps, clearly anyone farming out server style services 
    would be well served by having this sort of feature.  I believe this is
    the other main use of UML in Linux, but it must also be considered in
    the type of clustered environment that I intend to build for DragonFly.
    I don't expect performance in such an environment to be great, at least
    not initially, but that isn't the point of being able to do it, at
    least not at the beginning.

    Anyone for which security is far more important then performance
    would also be well served.  The load on the machines in my server room
    is fairly low... performance isn't an issue.  I have all those machines
    primarily for security separation and resource management purposes.

    For other real world apps... well, it depends on what kind of app it is
    and what kind of performance is needed.  As has been mentioned, the
    performance of this sort of abstraction suffers when it comes to anything
    that requires a lot of page table manipulation or makes a lot of system
    calls.  Strangely enough, cpu-bound programs such as compilers tend NOT
    to suffer anywhere near as badly.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>





More information about the Kernel mailing list