1. Enumerate and describe three generations of networking systems ...

gazecummingΔίκτυα και Επικοινωνίες

26 Οκτ 2013 (πριν από 3 χρόνια και 7 μήνες)

103 εμφανίσεις

1.

Enumerate and describe three generations of
networking systems proposed by Van Jacobsen


Generation 1: The phone system



circuit switching

Based on wires
-

very expensive to put them everywhere (make them ubiquitous
'wszechobecne'). Revenue ('dochód')

comes from calls, wires are sunk costs. Calls are treated

as a
side effect of connecting two points by wires

(system based on the paths)
. Phone number may be
treated as a program put in the system that builds a new path.
It started as a
very primitive
tec
hnology.


Throughout the history, when the operators where connecting the callers the phone number
stated coordinates of a point where the operator had to put one end of the wire. First automatic
switch
-

Strowger
switch (inwented around 1890) allowed to
select two
-
digit numbers


each pulse
caused the digit to increase (by switching horizontally), then when it got a long pulse the second
number could be chosen (this time by switching vertically after each pulse) [2D matrix] . Two
switches of this type lin
ed serially allowed to choose 4
-
digit numbers (10k possibilities).
Calling was
impossible when somebody was already using the section of a wire that was necessary for creating a
path for you or any element in the path fails (
reliability goes down exponenti
ally as the system
scales up



the only way of increasing the reliability was making individual elements of the system
extremely reliable


component reliable

[opposite of the internet philosophy, which focus on
creating numerous cheap component and doesn’t worry about
failure of one of them, making the
system
structurally reliable
]
).


A huge drawback of using telephony to send data is the
relatively long
time
of sett
ing up a
path (typically between 100ms and 1s
-

a small share of a phone call, but a really big part of sending
data). The problem was what to do to make quicker setups.

A large changed occurred in 60’s


electronics came up, enabling to dream abo
ut
decentralization.


Generation 2: The
I
nternet



packet switching

It had a good start


all wires were already in their places. The main idea was to
focus more
on the endpoints rather than the paths

and
split the data into independent chunks

(packets).
Another important thing was
transitivity

(‘przechodniość’ ?)


a way of forcing node to sent the
packet that got to it
further if it’s not addressed for it. Those ideas were brought up to life in
ARPAnet
, using already estab
lished phone
networ
k [however, it could use also Ethernet, radio,
satellite etc.


it was designed agnostic about the delivery technology


possible due the change of
view: endpoints, not paths). The only thing seen externally were the addresses


all the knowledge
about top
ology became
irrelevant

to users



distributed routing spreading load and avoiding
hierarchy problems
. Joining a lot of independent networks bring out a need of regulate the
addresses
-
it was the beginning of TCP/IP.

Here,
reliability increases exponential
ly with system size

(more and more alternative paths available).
Transitivity allowed to remove the need of call setup
.

However, the TCP/IP have it own problems. First of all is treating a ‘connected’ state as a
binary attribute (you can talk to everythin
g or be isolated).
Next, the addresses need to be stable.
Finally, a huge
wastefulness

while broadcasting


if a lot of machines demand the same information,
the server needs to send it to each of them individually, creating numerous copies of it.



Genera
tion 3:
D
issemination

networking

Both first and second generation focuses

on enabling conversations between two
machines
. Nowadays, a huge part of internet traffic are just request for some named packets of data
(webpages
, mails, files), which can be obtained not by conversation, but by
dissemination

(
point to
multipoint or multipoint to multipoint

like in “Does anybody got the time ?”).

Dissemination through a conversation is possible, but
ineffective


there’s a huge p
roblem
with a security (network is blind for data which is wrapped between TCP/IP headers)

and users have
to specifically state the goals and
their realization each time

they demand data.

Basic
assumptions

of dissemination networking

are:

*
enabling data r
equests by name



the means of obtaining it doesn’t bother the user; the protocol
uses them all (a bit of a analogue to using a lot of mediums to transmit internet packet data)

*
anything that hears the request and has a valid copy of the data can respond
to it

*
the responded data is signed and secured, and it’s validation could be checked

(security model
needed !)



trust are derived from the data, not the channel it arrives on

2. Describe advantages of packet switched networks over circuit switched
netwo
rks.




No need of call setup, which takes a significant part in overall time of c.s. networks



R
eliability increases

(in c.s.


decreases)

exponentially with system size



Knowledge of topology is irrelevant



Easily spread load and avoidance of hierarchy
problems


thanks to focusing on
endpoints rather the path between them


3. Describe the properties of dissemination oriented networks.

First of all,
let’s copy
the last point from the answer to the first question: data has a
name, not a location
-

trust a
re derived from the data, not the channel it arrives on
*
. Next,
anything that moves bits in time or space can
and will be used to communicate


there is not
distinction between data in a wire, in a memory or on a disc.


Another
property of d.o.n. is the as
sumption, that

the network is doing more on user’s
behalf


the user is not giving a definite instruction but in a fact a he does simply state what he wants
and the network delivers it to him.

Also, the network do not depend on conversations, but on diss
emination


the popular
content does not generate huge unnecessary traffic (like it does in TCP/IP).

Organizing local wireless or sensor nets are easier


the nods doesn’t need to have a name
.

The data can be remembered, so intermittent (‘przerywane’)
operations are possible, also
with use of opportunistic transport.

*
Trust and data integrity are foundation of dissemination network design, not an add
-
on like in
TCP/IP.