Cache coherency, clustering, and Kernel virtualization

[Bill Hacker]

Thomas E. Spanjaard wrote:

*snip*

This is hardly the logical step forward beyond SMP, and does nothing to 
make proper use of consumer-level NUMA equipment (AMDs Athlon64/Opteron 
family of processors for example), I don't see multiple virtual kernels 
work in a NUMA system. Or do you intend to not neglect this class of 
machines? :)

Let us not forget that what Matt has proposed to do is not the be-all-and 
end-all of DFLY. Wasn't really even the core part of the 'Prime Directive'.

The reverse, actually. A side-trip, using a byproduct of the accomplishments 
to-date of the core work focused on clustering and 'portability' of resources 
and running code.

NUMA, and the AMD instructions set, as well as Intel's sudden rush of brains to 
the anatomy with Vanderpool, are not yet where IBM's mainframe-class CPU have 
been for a very long time.  I'm not talking about speed or cost, or even 
elegance of the instructions set or pipeline - or any of that.

Simply that these chunks of IBM transmetals have *hardware* as well as 
instruction-set and 'mode' mechanisms to make it easier for them to export or 
virtualize themselves well.

That remains a side-issue for now, and maybe forever, w/r DFLY.

I am sure that in time, DFLY will run on whatever architecture is widely 
available and well-supported. 'Supported' meaning by F/OSS in general and right 
here in specific, not just in IBM, Intel, or AMD Research parks.

I've believed for some time that a *BSD could make a far better hypervisor than 
Linux, and DFLY perhaps the best of all, so I am *delighted* to see Matt choose 
to pursue that - even if only as a better demo / devel tool.

JM2CW, but Virtualization and hypervisor capability are of far more utility to 
me, here and now, than clustering is likely to be ... ever.

YMMV, YOMD

Best,

Bill