Re: What, exactly, has actually failed...

[Kobus van der Merwe]

I was motivated by the WG discussion on "GENI is not a network"
to read through the OMIS archives and happened upon this
related post from Mike Patton. So herewith some thoughts.

So, while I have to agree that GENI is not a network in the
same sense as a current day ISP, the fact that it has distributed,
interconnected and shared resources, does mean that GENI
would operationally have many of the same concerns that current
networks have. (I.e., a slice is only useful to the extent that
the underlying substrate components are operational and
connected.)

As a simple example, suppose an optical (or other transport)
path between two GENI nodes, further suppose no currently
running GENI experiments using this path and finally assume
a fiber cut somewhere between these two nodes. So the question
is whether operationally GENI would/should be aware of this
fiber cut (even in the absense of any experiments using the
path in question)? I think it should. And I think that illustrates
the need for an operational "in-band substrate view", i.e.,
being operationally aware of the state of the substrate independent
of any experiments. (And to some extent this is a network view,
albeit not a conventional IP network view.)

I think of the  operational view via the component manager (CM), as
Mike described it, as an "out-of-band substrate view". I.e., monitoring
the path to the CM only tells you about the availablity of
an out-of-band view, but does not tell you anything about the
health of the substrate itself. Although, this would be the
way to obtain information about the in-band substrate view.

The third operational view needed (i think), is a "slice view",
which would tell us something about the health of a slice.
Here I would agree with the sentiments expressed at the WG
that this view could be a noop (in the case of experiments
that either don't care about that, or want to role-their-own),
or could be something of an outsourced operational service/function
provided by GENI for long lived experiments that do not want to
do it themselves.

Kobus

_______________________________________________
omis-wg mailing list
omis-wg@...
http://lists.geni.net/mailman/listinfo/omis-wg