Re: Re: Some thoughts on policy and governance issues with respect to open lightpath exchange points
Harvey Newman <newman@xxxxxxxxxxxxxxx>
Tue, 05 Apr 2011 13:56:39 -0700
I would like to add another 2cents, based on the origin
of the OEP and hybrid network concepts, and why this
was taken as a common, compelling direction:
(1) Research teams were foreseen to generate
flows at some point that are either impractical,
or too expensive to support using a general
purpose network fabric. That time is now upon
us, during this LHC run of 2011-12.
(2) "Too expensive" means for the research teams
involved, especially when they operate "at scale".
The scale is progressing from a few hundred to
a Terabit/sec in aggregate over the next "couple"
(3) Hence the issue of costs raised 5-10 years ago
still applies. One can/should ask "what does a
typical configuration/map cost" and is it within
the power of the players to actually provide it
to the research community concerned ? If not,
go for a cheaper solution, that could be affordable
when operating at N X 100 Gbps at each OEP.
(4) Openness = freedom of (sub-)communities to
interconnect transparently. Free ? Not necessarily.
Affordably even at full scale ? Yes.
(5) Lightpath: overloaded from the beginning. It was
meant to be an end-to-end path at the optical layer,
as in a Layer 1 infrastructure. Over time, the use for
dedicated Layer 2 paths came to be more and more
common. Call that a "vulgarization" of the original
concept if you like.
(6) Essential characteristics:
o Point-to-point as perceived by the user, or the
o Dedicated, with hard bandwidth guarantees:
for deadline scheduling, fair sharing of a resource with
a known capacity, mutual isolation of GP traffic with
many small flows and research traffic with a limited
number of large flows
[launching thousands of small flows just blurs the picture
and makes it all less manageable.]
o Static or dynamic: dynamic is cheaper unless
one is in a resource-rich environment (which we are not)
or where true 24 X 7 utilization is required (which is rare,
and even more rarely affordable at the full 2011-12 scale).
o Highly reliable/available: Provided through a multiply connected
infrastructure though arrangements for fallback (not just
among the partners.
o Affordable at scale (as above): this means that not all
implementations are equally effective. It also means that
the partners have an interest in providing affordable solutions,
even if it affects their implementation choices.
On 4/5/2011 1:20 PM, Artur Barczyk wrote:
I haven't seen Jerry's original mail (was it sent to glif-all?), but
the ones followed up by Victor make already a very good set of
Here's my 2c for the discussion:
Indeed, "distributed exchange point" is an oxymoron :-)
"Distributed exchange" would be a better name, but apart from
that (c.f. below), since we had this discussion in the LHCONE
context, what is meant is an exchange with multiple nodes, where
the interconnecting links/trunks form an integral part of the
"exchange". Hence to the end-user, whether you connect in New York
or LA or Seattle, you can exchange traffic with the parties
connected to any node making up the "exchange". The tricky parts
of right-sizing etc. are of course to be dealt with properly by
the operator, but the theoretical concept is useful.
Whether it is practically, depends on the implementation.
But, I had the following question on my mind: what is the difference
between "distributed exchange" and a transit network between
exchange points? Are we just inventing a new term? Rebranding?
Victor raises an important point: "Open". I think important part of
the discussion will be centered on that.
My definition is that "open exchange points" are basically neutral
facilitators acting on behalf of connecting parties (users=networks).
If two (or more) networks present at the exchange point agree
between themselves to exchange data, the "open exchange point"
does not impose any other restrictions on them. They also
accept any user/network willing to connect - the final
agreement for "peering" (at whichever layer is supported)
is between the users of the exchange.
And I agree, open does not have to mean free.
"Federated" - my belief is that people using this term often
really mean "confederated". To the latter, I can stand.
On 04/05/2011 08:29 PM, John Silvester wrote:
In the light of Jerry's search for definitions, here are my comments,
some perhaps clarifying some no doubt muddying.
We should probably distinguish "exchange" from "exchange point". The
former can have its own internal topology whereas the latter tends to
imply a single point (place or locus) of operation. With this
distinction, the term "distributed exchange" makes sense but
"distributed exchange point" doesn't (it is an oxymoron since a point
can't be distributed, can it?)
The goal not GOLE :-) of an exchange point is to provide a capability
for other entities (usually networks) to exchange traffic where the
interaction between the participants may occur at different protocol
Lightpath is a heavy overused term - I am not sure what it is.
Global is probably an incorrect term here. Perhaps we should use
International (meaning providing services to more than one country).
I have an aversion to the term Federated.
Open means that the services are available to all (not necessarily for
free) in a non-discriminatory way (maybe restricted to some set of
similar entities - e.g. non-profit or education community or
non-commercial or ...)
Should be an interesting discussion.
On 4/5/2011 8:17 AM, Victor Reijs (work) wrote:
Just my few cents on your questions:
Jerry Sobieski wrote:
2. What really is an "Exchange Point?" How can you tell an "Exchange
Point" from "Not an Exchange Point"? How do networks fit into this
picture? What for pete's sake is a "distributed exchange point"?...
I see a GOLE/Exchange Point; just as a domain (with its policies, AUP,
access rules, etc.).
I understand the policies of the GOLEs are more relaxed than for
instance of an NREN (which is only allowed to be used by the NREN's AUP)
Physically it does not have to be one device (and it can be at any
layer), IMHO. It is a transit network/node (at any layer).
I even see HEAnet's IP/BLUEnet p2p network; GEANT IP/GEANT Plus/Lambda
DANTE; Internet 2/ION; etc. as Exchange 'nodes' (how the node is being
made is not a real concern, so things as resilience, disaster recovery
can all be catered for). These are not 'Open' (as there is an explicit
INEX, AMSIX are an open exchange point (at layer 2), correct (although
some AUP must be there, even it it is empty).
3. What really is meant by "federated" multi-channel lightpaths
[interconnecting lightpath exchanges]? The definition of "federation"
involves ceding certain rights/roles to a central governing authority...
What rights are we referring to? and to what governing authority? Is
this already being done? Or is this a proposal?
Perhaps with 'federated (management)' is meant that the management is
federated (so each party who owns the domain, manages it with the aim
to make a more general service). 'Federated (access)' could also be
that with in token (SSO, eduGAIN, etc.) one can access all the
Don't know if this helps of confuses things more.
All the best,