Technology Changes
- Bandwidth is no longer expensive.
- Reliability as bit-error rate is increasing.
- Costs are dropping.
Complexity vs. Simplicity
- As things move faster, there's less time to operate on each thing.
- Working with fast technology may also be difficult.
- Complexity doesn't scale.
Stupid Networks
- Stupid networks are simple networks; they do little more than transport
data.
- Lack of complexity leads to easier scaling.
Doctrine vs. Strategy
- Learning to deal with dogs vs. "Dogs don't do that".
- Doctrine doesn't work in changing or complex environments.
Changing Technologies
- As the environment changes, environment-based assumptions may no longer
hold.
- Networks have to react to environmental changes.
- Stupid networks can more easily and nimbly react to changes than can
intelligent networks.
Environmental Changes
- And the POTS environment has changed, due to the Internet.
- The rise of data traffic, with its differing statistics.
Intelligent Networks
- Move intelligence (computers and programs) into the network to make it
easier to add new services.
- Service creation involved mucking around in the network.
Expectations
- Avoiding supplier lock-in via publicly available components.
- Add new services and revenue streams.
The Stupid Advantage
- The stupid network is based on three capabilities:
- Cheap and plentiful infrastructure.
- Underspecification
- Universal transport.
- The result is a platform for rapid innovation.
Infrastructure
- The stupid approach to problems is to over-provision.
- Too much congestion? Add more bandwidth.
- Too many disconnects? Add more bandwidth elsewhere.
- In a technology advanced environment, over-provisioning is simple and
effective.
- Until computers, this was the classic engineering approach to
uncertainty.
Technological Abundance
- Transmission capacity per square inch.
- Switching operations per hour.
- Costs move in the opposite direction.
Reacting to Change
- Intelligent networks require exacting analysis and preparation to react
to change.
- Stupid networks grow more easily with simpler planning.
Underspecification
- Configuring for specific environments may make adaptation for new
environments difficult.
- It depends on how deeply the assumptions are embedded.
- Minimalist assumptions make reacting to changes easier.
- On the other hand, it may be harder to know about what you're getting.
Minimalist Assumptions
- Minimalist assumptions are harder to contradict.
- Make no assumptions about packet size (apart from the maximum).
- And they're easier to recover when contradictions apply.
Internetworking
- IP makes minimalist assumptions about the underlying networks.
- As long as the underlying networks meet the minimal assumptions, IP
works.
- The less restrictive the assumptions, the more networks that can be
used.
Commodity Components
- IP-style assumptions reduce components to commodities, even if the
component's owners don't view them as such.
- It's tough to make money with commodity components.
- Everything comes down to bandwidth and connectivity.
Networking Irrelevancies
- By making the underlying networks mostly irrelevant, IP simplifies the
end-points interaction with the network.
- IP is the only network that you need to deal with.
The End-Point
- Minimalist interfaces simplify by pushing complexity to the end-points.
- You can't do much through a minimalist interface, and the interface
controls the network.
- That leaves the other side of the interface: the end-point.
- Three-way calling? The network doesn't do that, but you can using the
network.
- It may not be as efficient as network-based multicast; but it may
be.
Innovation in Stupidity
- Developing and deploying new services is easier in stupid networks.
- The network is easier to understand.
- All the work is done outside the network, isolating problems at the
end-point.
- But there is a difference between getting it going and getting it going
well.
Innovation in Complexity
- Fitting new services into a complex network is complex.
Protocol Changes
- IP version 6 (IPv6) adds new capabilities which can be directly
exploited by new services.
- Larger address space, real-time features, mobility, authentication
and encryption, multicasting.
- However, IPv6 has not been adopted as quickly as originally thought.
- Once the addressing problem was rendered moot, the remaining features
(e.g. security) could be solved using end-to-end approaches.
IP vs. ATM
- Simple but limited IP transport can replace complicated but featureful
ATM transport.
- The ATM features are moved out to end-points.
- This makes the network cheaper and, for the same underlying
bandwidth, faster (due to less overhead).
-
Traffic Differentiation
- Traffic differentiation, the Holy Grail.
- Traffic differentiation leads to price differentiation and profits.
- Telephony requires low delay, TV requires high bandwidth, financial
transactions require low error rates and security.
- The question becomes: are these end-to-end functions or are they
inherent in the network?
Quality of Service
- Quality of Service (QoS) is one way of implementing traffic
differentiation.
- Streams through the network are guaranteed to get specified levels of
service.
- For a suitable definition of "guaranteed", and assuming the stream
end-points behave.
- Is QoS an end-to-end property?
QoS vs. Technology
- Rely on technology to outgrow the need for QoS.
- Also consider the smarts at the end-points.
The Skeptical View
- Is technology moving that fast?
- Yes; the real question is "Is it moving fast enough?"
- Has
Parkinson's Law
been repealed?
- No, but that's a universal counter-argument (electricity too cheap to
meter).
- You can drag economics into the argument, but that's independent (but
not separate from) technologies.
Bibliography
This page last modified on 13 November 2004.