Short version: The current DFZ growth rates are not our target.
In the long term - including ideally the next
6 to 10 years - we need to alter the architecture
so it can cope well with hundreds of millions of
multihomed end-user networks.
I eyeballed the graph:
http://bgp.potaroo.net
and estimated 16 to 17% per annum growth in the last 3 or 4 years.
I expect this rate of growth will increase in the next few years as
IPv4 space is subdivided more intensively once the last of the fresh
pastures at the edge of town are engulfed.
I think there is little point in discussing whether DRAM speeds, CPU
speeds etc. can cope with the current rate of growth in the DFZ in
the long term.
In the long term, we need to enhance the architecture of the Net to
cope happily with a much larger growth in the numbers of end-user
networks which have multihoming, TE and portability.
I suggest we should be happy with a system which is technically and
commercially agreeable with numbers of such end-user networks like:
2016 1M
2020 16M
2024 256M
This is 100% PA growth. Each end-user network will probably have
multiple prefixes too, for load sharing over multiple links and for
their multiple sites - so the total number of portable, multihomable,
prefixes needed will be 2 to 10 times this.
Ideally, the scalable routing solution should be technically and
commercially capable of scaling within a few years of introduction to
a hundred million or so separate prefixes (or non-binary length
spans) of end-user network address space. However, I guess it would
take a few years for such levels of demand to eventuate (not counting
the immense demand for stable IP addresses for mobile devices).
There's no way that CPUs and RAM, for either the RIB or the FIB, can
cope with this sort of growth, as long as the prefixes are provided
by the current BGP-based techniques.
Even if we could somehow build and afford the routers, this would be
an inelegant, unstable and probably unmanageable number of prefixes
to expect the interdomain routing system to cope with.
The current system is unacceptable since it doesn't fit the needs of
many end-user networks. The disparity between the unadorned BGP
system's capabilities and the needs of end-user networks will get
worse with every passing year.
I don't think there is any brick wall where the interdomain routing
system as we run it today will stop working. I am sure DFZ routers
can be built to handle 20% PA growth in DFZ prefixes. (Despite what
I wrote about DRAM latency, routers could be built with faster
memories, including static RAM and/or with multiple CPU-RAM systems
sharing the RIB and FIB loads.)
Our success will be measured in how soon we can devise a better way
of providing hundreds of millions of end-user networks with
multihoming, TE and portability.
IPv4 address space will become harder and harder to get, since every
customer needs one such address. A good core-edge separation scheme
could improve on this considerably, by enabling improved address
utilization. It could do this by enabling many end-user networks to
multihome etc. with as small a number of addresses as they really
need - not in multiples of 256, 512 etc.
Another measure of our success will be how well the scalable routing
solution enables more efficient IPv4 address utilization. That will
provide more time for figuring out an alternative to IPv4.
Ideally, a well designed solution to the routing scaling problem
could help with the transition to whatever lies beyond IPv4.
Ideally, it will help with mobility too.
- Robin
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg
