Re: [dc] draft-khasnabish-vmmi-problems-00.txt

[Thomas Narten <[e-mail]>]

Ashish,

> Let's assume we are building protocols for hypervisors. Can you suggest
> what type of protocols will allow a hypervisor based solution interact
> with a non-hypervisor based server? I.e. encap-decap done in the
> hypervisor is not understood by the non-virtualized server.

In which case, in something like NVO3/VXLAN/NVGRE, the switch that the
physical server connects to can do the encap/decap.

But the kind of question you are asking above about protocols
completely misses the point.

We are not building "protocols for hypervisors". Hopefully, at some
point we talk about building "solutions" to "real concrete
problems". One of the requirements for such a solution might be that
hypervisors can particpate in the protocol. And that non-virtualized
server are not excluded (and indeed that is what the NVO3 problem
statement talks about). If so, those sorts of requirement get factored
in. But the key thing is not "hypervisors" but how well a proposed
solution addresses the problem and all of its assocated requirements.

<broken-record-mode>

So, first, we articulate a problem or pain point that we need a fix
for. Then we start talking about requirements a workable solution must
address, then mechanisms that address the problem.

Talking generically about "hypervisor based solution" needing to talk
to "non-hypervisor based servers" seems like a pointless exercise. At
least not without being able to answer "solution to what? part of the
question"

</broken-record-mode>

Thomas