|
Subject |
Re: globally unique identifiers for lightpaths? |
|
From |
Leon Gommans <lgommans@xxxxxxxxxxxxxx> |
|
Date |
Fri, 07 Dec 2007 09:40:09 +0100 |
Ronald,
Within our work with Internet2 we discussed the same issue. Essentially
your need to refer to a "network session id"
that needs to be created in a globally unique way. Each domain may have
to take the initiative to create such
an identifier. In our discussion with Internet2 we started to called it
a Global Resource Identifier (GRI).
This identifier essentially allows local domains to assign its resources
to this GRI. The information related to local resources can was called a
Local Resource Identifier (LRI). Each domain can use its own conventions
here. The GRI was essentially build using a domain identifier and some
number which each domain is free to choose as long as it is unique and
adheres to some format constraints posed by the inter-domain signalling
protocol. The random number you mentioned could be one way. One may
start to think about domain-id's that some authority will issue and this
might need discussion in the GLIF.
Regards .. Leon.
Ronald van der Pol schreef:
On Thu, Dec 06, 2007 at 19:26:29 +0100, Erik-Jan Bos wrote:
Hi Ronald and Dave:
I too support the global naming of lightpath. As one of the "design
criteria" I think it should adhere to the KISS principle as much as
possible.
About KISS: you know I fully agree :-)
Hence, I think that the system we come up with should:
* Ensure uniqueness.
Two possibilities, I think:
- central repository
- random number
I prefer the latter.
* Be maintenance free as much as possible (i.e. a name should not change
when one of the characteristics of the lightpath changes).
Do you think the name should remain the same when the lightpath
is re-routed? That rules out an identifier which lists all
the GOLEs or all the local names in the path.
I am not sure. It is convenient to be able to deduct the topology
of a lightpath by looking at the name. If we do not do this, we
really need a good inter-domain monitoring system setup fast.
That system must provide domain and topology info for each lightpath.
* Avoid any notion of a central authority or repository.
* Be able to generate a name on the spot by an organization (i.e.
without going back and forth with one or more peer organizations).
I agree. How about these options:
- generate a random number of X digits
this will be a unique ID when X is large enough
- take the current time in UTC
YYYYMMDDHHMMSS, e.g. 20071206200042
It is highly unlikely that two lightpaths get created at the
same second
I prefer the latter. I does not need a random number generator program.
Just look at your watch to create this number.
At the moment, my preferred ID would look like this:
YYYYMMDDHHMMSS-GOLE1-GOLE2
GOLE1 and GOLE2 are UN Location codes (http://www.unece.org/cefact/locode/)
Examples:
20071205200042-AMS-CHI
20071206070809-NYC-TYO
- it is easy to parse because it has a fixed format (22 bytes)
- the number YYYYMMDDHHMMSS ensures uniqueness, especially in combination
with the two locodes
- it gives the egress and ingress GOLE and thus some kind of topology info
rvdp
