Lecture Notes for CSE 561, 10/2

Question 1:  How long does it take to adopt a new technology?
	Tech			Invented	Adopted
	Compact Disc		early 80s	mid 90s
	cars			1890s		1920s
	OOP			early 60s	mid 90s
	RISC			early 80s	early 90s
	public key crypto	70s		late 90s - not yet
	transfer protocol	60s - 70s	mid 90s
	RAID			mid 80s		early 90s
	video codecs		???		90s

	conclusion 1:  nothing useful done in the 90s
	conclusion 2:  it takes a long time to adopt, and there's
	a high degree of variability in the time to adopt (TTA).

Question 2:  Why the variation in TTA?
Possible Answers:
	- somethings are just intrinsically better
	- market forces (supply/demand and cost/benefit)
	- interoperability, legacy vs. plug and play

Question 3:  What is the upshot for the Internet?
The most important advantage of today's Internet design (as compared to 
OSI/ISO, for example) is its interoperability.  Not coincidentally, this
is also the fundamental design goal of the DARPA net, according to Clarke.


Question 4:  Were design decisions made in the early 80s the right ones 
for today?  In other words, how would we redesign the Internet if given
the opportunity today?

But first we need to answer
Question 5:  What constraints, design goals, and parameters must 
Internet designers work with?
A) Constraints and parameters
	1. Have existing networks, links, and hosts
	2. Number of hosts -- unbounded? 64 bits (~10^19) enough?
	3. Number of companies -- in the millions
	4. Number of device manufacturers -- thousands
	5. Kinds of applications to support -- ideally, all
	6. Kinds of media to run on -- ideally, any
	7. Kinds of networks (as opposed to media) to run on
		a. heterogeneous in speeds (1 kb/s - 10 Gb/s)
		b. wired/wireless
		c. simplex/duplex
		d. bus/switched
		e. circuit-switched vs. packet-switched
		f. packets can be corrupted/reordered/lost/split

B) Design Goals (some already implied by above)
	1. According to Clarke (in order of importance)
		a. efficient operability (for existing networks)
		b. availability/robustness
		c. flexibility in application support
		d. flexibility in support of new types of nets
		------------- (unmet or partially unmet goals below)
		e. distributed management
		f. cost-efficiency
		g. low administrative costs
		h. ability to control resource usage
	2. According to class (in no particular order)
		a. security
		b. extensibility
		c. performance/efficiency (as opposed to cost-effic.)
		d. scalability
		e. measurability
		f. conceptual simplicity/understandability
		g. predictability
		h. modularity
		i. anonymity (at odds with accountability)
		j. ability to accomodate conflicting interest groups
		k. mobility
	
	Side note to explain what is meant by predictability:
	If we add buffers to the network, will they improve or hurt
	performance?
		- hard to say
		- or, one could say, it depends on def. of "performance"
		  RTT might be longer, but packet loss rate is lower
		- so maybe this is an example of poor measurability (?)

	Another side note:  in a vote by industry engineers, the most
	important design goals were 1. interoperability, 2. availability, 
	and 3. extensibility (because with it comes everything else)

So now we can begin to answer Question 4 ...
The class voted for a different prioritization of design goals from the
one described by Clarke.  Chief among the issues raised were security,
scalability, and extensibility.

Question 6:  How do we build for the desired design goals?
The key ideas of the current Internet implementation are
1) networks of networks (hierarchical management and routing)
2) interoperable gateways
3) store & forward
4) few assumptions about underlying networks
5) end host recovery -- no state in network (or soft state)
6) independent of type of data
7) best effort
8) layered implementation

New ideas for a new implementation:
1) information about paths/capabilities
2) Quality of Service (QoS) requirements
3) allow capability of underlying networks to be exploited by hosts

Virtual Private Network (VPN) example:  there is a business demand for
secure circuits.  So security is currently added to the network in an
ad-hoc way, one business at a time.

It was pointed out that there is no incremental advantage for a service
provider to provide any innovation (like security, QoS) locally.  Only
when a reasonably large portion of the network adopts the innovation will
it have an impact.  This creates a significant barrier to adoption of new
technologies.

Corruption of data example:  end-to-end principle says that services should
be provided at the lowest level where it can be fully implemented.  For the
detection of corrupted data, almost every link does some kind of error
detection.  Still, about 1 in every million packets gets corrupted.