Short version: I think this article has nothing to do with
Internet routing and so does not provide any
significant insights into or practical solutions
to the routing scalability problem.
This article:
Sustaining the Internet with Hyperbolic Mapping
http://www.nature.com/ncomms/journal/v1/n6/full/ncomms1063.html
http://www.nature.com/ncomms/journal/v1/n6/pdf/ncomms1063.pdf
doesn't seem to be relevant to AS networks and the BGP-based
interdomain routing system as I understand them. Nor does it appear
to address the problems we have identified - a more scalable method
than the current BGP-based DFZ for providing multihoming, inbound TE
(traffic engineering) and mobility.
The biggest challenge, I think, is mobility for billions of devices,
each responding to its own IP address, so it can interoperate with all
other hosts using standard stack and application protocols - with
generally optimal or close to optimal paths and session survival via a
persistent IP address, no matter how the "care of" address(es) change.
The only way I can see of doing it is TTR Mobility:
http://www.firstpr.com.au/ip/ivip/#mobile
To see my understanding of the goals of scalable routing, please see
section 17.2 of:
http://www.ietf.org/mail-archive/web/rrg/current/msg06219.html
My sense that the article's premises don't accord with reality begins
with the 3rd paragraph:
The scaling limitations of the existing Internet routing stem
from the requirement to have a current state of the Internet
topology distributed globally.
This doesn't accord with my understanding. BGP doesn't generate or
distribute a map or anything resembling "the current state of the
Internet". It is simply a method by which the routers can work
together so that each router can, decide which neighbor to forward a
packet to, based on the shortest BGP-advertised prefix the destination
address matches. This is done in a manner by which outages and
topology changes result in changed route advertisements by each
router, and so to changed choices and further changes of route
advertisements by other connected routers.
For some routers - such as those where "the rest of the Internet" is
reached by a single neighbor, this means that packets addressed to
pretty much every BGP advertised prefix are all sent to one particular
neighbor. For other routers in the "DFZ" (Default-Free Zone), there
will be two or more neighbours which the router decides are the best
ones to handle packets addressed to "the rest of the Internet" (that
is, the prefixes which are not ones this router handles itself).
Such global knowledge is unavoidable, as routing has no source
of information other than the network topology.
I think this too is incorrect.
The BGP routing daemon in each BGP router certainly does have another
source of information: the routes advertised by its neighbours. This
is what it uses. No BGP router has a clue about topology.
Routing in these conditions is equivalent to routing using a
hypothetical road atlas, which has no geographical information
but merely lists road network links, which are pairs of
connected road intersections, abstractly identified.
This describes a text-form of a map, similar to a printed circuit
board netlist - where distances between connected items are
irrelevant, but the total set of connections is specified exactly.
This has nothing to do with how a BGP router chooses the best path for
packets addressed to a given BGP-advertised prefix. There's nothing
resembling a road map or any other map. There's just a bunch of
advertised routes from neighbours, each with its own list of ASes
which must be traversed to reach the destination. Then the router has
its BGP algorithm and local policy for choosing one of these. It then
advertises this path to its other neighbours, perhaps with extra AS
numbers added.
So how does the rest of the article relate to the Internet in general,
or to the interdomain routing system in particular? As far as I can
see, it doesn't.
How would this approach be adopted?
What changes to routers would be required?
How could this work with BGP? Modifications to BGP, creating a new
network to replace the BGP-based DFZ?
Why would anyone adopt it?
When would they derive any benefits? After the rest of the world
adopts it too?
If it was adopted, how would ASes get their routers to implement the
policies they current set for them?
How would this provide multihoming to more end-user networks? I guess
this would have to be either without requiring each to have an ASN, or
allowing millions of such networks to get an ASN, but somehow have
this not matter.
How would mobility be achieved in a robust and scalable manner, for
hundreds of millions or billions of devices - with IPv4 or IPv6?
As far as I can see, this is a paper about some new approach to
constructing maps in coordinate systems which do not resemble ordinary
Cartesian coordinates. I guess this is of interest to some folks -
but I don't agree with the central premise that these techniques are
relevant to Internet routing or to a new approach to routing which
would better meet the needs of ASes.
Can anyone point out exactly why the techniques in this article could
help with scalable routing? Without using an analogy, such as "The
Internet's routing system operates like a road map. We have developed
an improved way of making a map for this purpose." - because that
analogy is false.
- Robin
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg
