DragonFly kernel List (threaded) for 2007-01
Re: VKernel progress update
I will do a full write-up before the release. Right now, insofar as I
can tell, it is very similar to UML. But I am not familiar enough with
UML to be able to compare them. Ours eventually will be able to take
advantage of Intel's and AMD's new virtualization stuff, but my
intent (for now) is not to create a high performance virtualization.
It will take considerable development to make VKERNEL perform well
and I hope that other developers will start doing work on it once I
get the thing working. Current short-term goals for VKERNEL are:
* Extremely low overhead in the real kernel to support a VKERNEL process.
Virtual kernels are highly compartmentalized environments.
* Ability to run virtual kernels as a normal user (not just root).
* Highly secure (meaning minimal interactions with the real kernel).
* Ability to test anything that does not require direct hardware
interactions. Networking will be via a pseudo device.
My own personal use of vkernels will be to ease development of the
clustering and cache coherency code. Being able to recompile a vkernel
and instantly run it will greatly reduce the engineering development
The core design of VKERNEL will allow for performance, SMP, and other
enhancements to be added in the future. SMP in particular will be no
big deal since VKERNEL links against libc. The main performance issues
revolve around caching execution contexts, dealing with page tables,
and copying data between VM contexts.