What, exactly, has actually failed...

[Michael A. Patton]

[This is (currently) my personal "prime concern" about actually
        operating GENI.  -MAP]

GENI is not a network, but a collection of components, control of
which happens over "the legacy Internet" (or something like that).
But, that means the NOC is monitoring (at least some) components over
a network that it is NOT monitoring.  This is very different from
existing practice at most ISPs.  So, how will GENI handle this?

In a traditional ISP, the NOC is monitoring everything between itself
and any piece of equipment that's being managed.  Traditional NOC
tools draw diagrams of the devices being managed and the connections
between them (also being managed).  So, when a particular device
becomes unreachable, they have the information to know whether that's
because something else on the path has also failed or if the device
itself has failed.  With GENI there will be failures where making that
distinction is much harder because we don't know the actual path.

All GENI components will have a "Component Manager" (CM) interface
that is reachable with traditional IP and which will export a
management interface.  But where that CM interface is plugged into the
Internet is ultimately up to the component's owner, so practices will
vary.  The traditional NOC tools that draw network diagrams don't
really seem to map to this paradigm very well.  So, how do you build a
NOC that can deal with that heterogeneity?

There are organizations other than ISPs that manage distributed
servers.  In some ways that seems to be more like the GENI paradigm
than the ISP model (and less like it in others).  Unfortunately, my
experience is primarily with ISPs, so I don't know if these other
folks are using COTS tools or home grown stuff.  What can we learn
from them, and should we encourage development of some cross-over
tools?  Anyone on the list familiar with some of these want to
comment?

I have a few ideas on some of these, but no solid answers.  That's why
it's my prime concern at this time.

        -MAP

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