Short version: The current number and rate of growth in the number
of multihomed end-user networks is not our
primary concern.
Our aim, as best I understand it, is to design
an architectural enhancement for the Internet
by which the much higher numbers of end-user
networks which want/need multihoming, TE and
address portability can have these in a manner
which is technically and economically sustainable
for ISPs and for the end-user networks
themselves.
A system which would work nicely with 10M
such end-user networks would be good. One
which scales well to 100M or 1B would be better.
Ideally the system should have an architecture
and business arrangements whereby each end-user
network does not impose unreasonable burdens on
others due to its activities, including how
many prefixes they advertise, how often they
change these advertisements and other matters
such as their share of traffic on OITRDs
(Open ITRs in the DFZ = LISP PTRs). When
such a system can technically scale well to
100s of millions of end-user networks - or
perhaps billions - and can be run as a series
of profitable businesses, then there will be
no scaling limits (or at least not for decades
to come) and little or no unreasonable burden
of costs.
I think Ivip has a much better chance of
achieving this than other proposals.
http://www.firstpr.com.au/ip/ivip/Ivip-summary.pdf
In "Re: BGP scaling limit?" Tony wrote:
> We already know that we're growing the table faster than
> technology can keep up. That much is clear.
I am not convinced that this is the case right now, since I can
imagine the continued advances in multi-CPU chips (with each CPU
running at 3GHz or more, with its own cache) with sufficiently large
amounts of static RAM, would be able to perform (separately) the RIB
and 1/10/40 Gbps FIB functions. However, that would be expensive and
inelegant - and would raise the real cost of DFZ routers in a manner
I think is unreasonable. Keeping the current system and souping up
the routers like this would perpetuate and exacerbate the current
problems, including:
1 - Difficulty and high costs for any end-user network which wants
or needs multihoming, TE and portability.
2 - Every such network advertising its one, two or more prefixes
placing an unreasonable burden on ISPs and multihomed end-user
networks all over the world by adding to the RIB and FIB
workload of their routers. (Assuming these DFZ routers do
carry the prefixes.)
3 - This burden is shared across all Internet users. A result of
this is higher barriers to Internet use than would otherwise be
the case - which shuts out some poorer people from using the
Net at all.
4 - A grey area of investment, unreasonable burdens and risk as
DFZ router operators are increasing tempted not to accept
some prefixes of multihomed end-user networks.
5 - Consequently, an increasing number of end-user networks
not getting the reliable service they need, at a time of
increasing demand, increasingly diverse usage, and increasing
moment-to-moment dependence on Internet communications.
I thought we agreed that our task is to modify the Net so the current
scaling difficulties with the BGP routing system are resolved, not
just for a continuation of current growth rates, but for 10s or 100s
of millions of multihomed end-user networks gaining multihoming, TE
and portability.
If not, I think now would be a good time to check consensus that we
are not just trying to cope with the existing, unacceptably
constrained, growth in the number of end-user networks with
multihoming, TE and portability, but to create a network which
technically and commercially cope well with some very large number of
such end-user networks.
A scheme which could clearly cope with 100M would be a good start.
Ideally - in my view - the system would be extendible to support a
form of mobility in which billions of mobile devices could keep their
IP address, or prefix, no matter where they physically connect to the
Net. A plan for this is in the paper Steve Russert and I wrote:
TTR Mobility Extensions for Core-Edge Separation Solutions to the
Internet's Routing Scaling Problem
http://www.firstpr.com.au/ip/ivip/#mobile
The key to this is that a mapping change is only needed when the
point of attachment moves such a distance (such as ~1000km) from the
currently used TTR that it is best to choose another TTR. This will
happen far less frequently than the changes in connection point, or
change of which access networks are used for the connection.
Each Ivip mapping change is paid for, so there are little or no
unreasonable cost burdens resulting from these changes.
- Robin
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg
