http://www.ispcon.com/SNEditorial/sn_detail.asp?ID=147
Wednesday, October 30, 2002
The stupid network: Why I'm not sold...yet
There may yet be a place for special packet treatment, and the silicon
to do it
By Scott Mace, SERVICENETWORKS.com
You know things are bad in the Internet industry when the best-sounding
idea out there is David Isenberg's call for the return to or acceptance
of his vision of the Internet as an all-IP-based "stupid network" � a
network designed only to move packets from one place to another with no
special treatment of special traffic.
It's a seductive theory at this point in time. Let's look at the
evidence:
* Manufacturers and operators of "smart" networks � networks
designed to treat some telecomm traffic, such as voice or video, better
than other traffic � have taken an absolute drubbing financially. The
Quality of Service Forum (for which I edited the last few white papers)
is a distant memory. Multicast technology remains an academic
curiosity, and traffic to the IP Multicast Mailing list has been fairly
light of late. (This Web site hosts the list, and it is still the
preeminent mailing list regarding multicast in the world.)
* Service providers have yet to roll out special-treatment packet
delivery services to any large extent (a notable exception of late is
anti-spam services). Instead, they've been turning to other techniques,
such as route optimization and content delivery networks (caching plus
monitoring), or simply overprovisioning their networks, providing
surplus bandwidth at all but peak times. (I don't consider services as
VPNs to be special treatment. Although the packets are encrypted, they
are otherwise treated just like other packets).
* Bandwidth costs continue to plunge. The fiber glut continues
unabated. Predictions of the glut's demise were predicated on Internet
traffic growth estimates which themselves turned out to be false.
(Listen to my interview with Andrew Odlysko for details.)
* Managing prioritized traffic in the end-to-end Internet is still
too expensive and complicated, despite a surge of research and
development aimed at putting the necessary software in silicon and
baking it into switches and routers.
* The technology used to develop smart networking equipment is
evolving too quickly and the state-of-the-art renders existing
solutions obsolete overnight. This is not a solid basis upon which to
invest in such equipment for a physical network which cannot be
upgraded very often.
* The human engineering and management expertise needed to manage
and optimize smart networks is in short supply and will remain so.
These arguments are powerful enough to sweep away most thoughts of
implementing smart networking any time within the next few years. So
why aren't I buying the theory hook, line and sinker?
I won't argue that anything listed above should compel any service
provider or carrier to rush out and install smart networking gear. But
conversely, I also know that carriers who have spent enormous fortunes
installing such smart gear as voice switching equipment won't rush away
from that investment eagerly to embrace dumb networks.
Clearly, the pendulum will swing far in the direction of stupid
networks. I expect the traditional telecom companies to continue to put
up a front of championing smart networks for voice, while at the same
time, inside these companies, they hollow out their infrastructure from
the core, placing dumb networks with overprovisioned resources there
and moving the intelligence to the edge � all very quietly.
At the edge (where, as Isenberg says, IP packets are broken apart and
applications read and utilize the contents of the packets), I certainly
would like to be able to prioritize my Internet traffic. If I'm
involved in a critical data upload, I still want to surf the Web, but I
want to be able to take away bandwidth from the Web surfing process and
give to the critical process. The gear and software to do this exists
today and there are many success stories in the enterprise arena,
especially where a high-latency, low-bandwidth WAN is in place.
I also want route optimization and the ability to grab additional
bandwidth on demand as needed, and I believe the technology for this
exists too. There is an opportunity here for service providers to
automate these processes and make them available in a seamless fashion
to customers.
The most interesting action that Isenberg has taken in the course of
his stupid network crusade was his October 21 call (with a few
distinguished friends) for the FCC to allow the legacy
telecommunications companies to fail. If they aren't allowed to fail,
the "hollowing out" I refer to above will take place at a much slower
pace � for instance, the telcos will continue to slowly depreciate
obsolete "smart network" equipment which should be written off as
quickly as possible (bankruptcy would do it really quick) � and as a
result the build-out of IP-based stupid networks will take years longer
� years the United States may not have to stay competitive with a world
where some stupid networks are being built out by governments in a
fashion Al Gore could appreciate, the result being that the U.S. is now
trailing Estonia in its rate of high-speed, always-on ("broadband")
Internet adoption, among other countries.
And yet, there are some questions that swinging to a stupid-network
model raise, that give me just enough pause to avoid becoming another
one of Isenberg's troops. These questions include:
* What do we do about IP-based distributed denial-of-service
attacks? In my interview with him this week, Isenberg admits that
stupid networks are more vulnerable to DDoS attacks than smart networks
are. In a world where all critical applications including telephony are
collapsed onto a single pipe, some fancy redundancy will be necessary
to thwart DDoS, raising the overhead costs of a stupid network. For
this reason alone, it would be worth attending next week's SERVICE
NETWORKS / ISPCON conference, to hear three different views about how
DDoS attacks can be pre-empted.
* How will carriers grab market share in any fashion other than
speed or price? Providing "five nines" or more of service may become
less compelling if all a customer has to do to insure a reliable pipe
is to "buy as many nines as you want" as Isenberg puts it, by having
redundant connections open to the Internet. As bandwidth becomes even
more of a commodity, thanks to Moore's law it will be less and less
profitable.
* Isenberg's solution is to treat it like other money-losing
infrastructure: the highways, the airlines, the airports, Amtrak. In
other words, let governments build and operate the Internet as a key
national infrastructure. Isenberg touts a public/private partnership in
Sweden, but the heavy hand of government is there as well. Can the
United States stomach that sort of government role in this day and age?
More importantly, must it? It's one thing to build sewers and roads,
but these facilities aren't effectively made obsolete overnight by
technology. Can the same be said about Internet carrier equipment?
Maybe the government ends up owning only the conduit, but as wireless
comes on strong, might even conduit in the ground be worth less than it
is today?
* Wireless spectrum may eventually be reallocated into smaller and
smaller pieces that can be shared by smart radios at the edge, but
today's regulatory reality is that spectrum remains scarce, a fact that
802.11 radio users are likely to run up to in a hurry with the
exponential growth of that technology. The Open Spectrum movement might
change that, but it's got its work cut out for it.
* Firewalls exist to filter traffic far away from edge devices.
Yet, in Isenberg's perfect stupid network, a firewall exists only on
the edge device itself. This to me seems as nonsensical as expecting
every edge device to be able to combat a DDoS attack by itself.
(Imagine if G.W. Bush's IPv6 cell phone address ever got out.) Whether
the unwanted content is spam, adult content, or malicious code, it's
evident that firewalls somewhere inside the edge of the network, which
peek not just at packet headers but somewhat deeper inside the packet,
serve a valuable purpose. Any stupid network which throws them out or
expects to shrink them into each device is imagining a technological
future beyond what I can foresee.
* I've often written about how smart networks and QoS are needed to
carry the low-latency "killer apps" of the future � everything from
videoconferencing to immersive applications ranging from telepresence
to gaming to simulations � and along the way, they will break the
monopoly held by legacy content distribution networks such as the TV
networks. But Isenberg says the killer apps of the Internet have
already arrived � email and the Web among them. Are we content to let
the others struggle without special network consideration until
bandwidth is virtually free? That's a pretty poor return-on-investment
for Internet2.
The answers to these questions elude many. For the past four years I've
examined the potential of smart networks for the Internet (not just
voice), to optimize a resource I considered scarce at bottlenecks. All
the time, others such as George Gilder have argued, to limited effect,
that bandwidth was becoming free and we will simply add more bandwidth
everywhere any time we need it. The truth remains somewhere in between.
By examining stupid networks more closely, I too am joining the ranks
of those who believe that bandwidth may be too cheap to meter. And yet,
some vast surge of silicon innovation may yield a solution which sweeps
away all the concerns about the high overhead of metering bandwidth, or
the lack of human skills to manage such systems. As always, the
pendulums in technology swing back and forth.
Last week, my colleague David Kopf gave some advice to service
providers which could be a saving grace in the short term: get your own
data center facilities in order. As improbable as it may seem, Web
hosting could emerge from the current downturn as the one area of
business where service providers and carriers can make a buck, if they
work hard, keep it simple and figure out how to maximize their use of
an Internet which remains, for the time being, pretty stupid. So, the
other SERVICE NETWORKS / ISPCON panel that I recommend to you would be
Next-Generation Web and Application Hosting: Distinguishing Hype from
Reality.
One thing is for certain: the world isn't waiting for answers to my
questions or Isenberg's. Every day, in the proving grounds of Wall
Street and world economies, technologies, businesses and even national
infrastructures are being tried, tested, and many are found wanting.
The important thing is to keep asking the tough questions, and not to
let any one factor, least of all mere campaign-contribution-stoked
politics, dictate the answers.
-----------
FoIB mailing list -- Bits, Analysis, Digital Group Therapy
http://www.ianbell.com/
