Transcript itb-2014-janx

Institut Teknologi Bandung
January 23, 2014
The World Trend of
Next Generation
Internet Technologies
Shigeki Goto
Waseda University
1
How to understand
new and old technologies?
• There is no royal road for learning.
• There is no single principle for understanding
the Internet at present and in the future.
• Computer Science has started with a negative
result by Alan Turing.
There are certain functions which
cannot be computed mechanically.
1912--1954
2
dualism
1. Binary opposition
e.g. good/bad, mental/physical, sun/moon
2. State of two parts
3. Complementally, conflict
logicalphysical
3
List of dualism approach
1. Wireless vs. Wired
2. Circuit switch vs. Packet
3. Centralized vs. Distributed
4. Do-it-yourself vs. Cloud
5. End-to-end vs. Hop-by-hop
6. De jure vs. De facto
7. Open vs. Trust
8. Array vs. Tree
9. Serial vs. parallel
10. Synchronous vs. Asynchronous
11. Fixed length vs. Variable length, and more
4
1. Wireless vs. Wired
• Wireless technology has been used for long
distance communications.
Marconi (1874—1937)
• Wire was first used to feed music program to
healthy people at home.
Butapest, Hungaly
• Now, we use fiber for backbone, wireless for
local access.
5
Hybrid wireless network
-- Cellular and WiFi -Queuing model for multiple WiFi stations
Cellular
Cellular
WLAN1
WLAN1
WLAN2
…
…
WLAN N
WLAN2
WLAN N
© Masaki Fukushima
6
2. Circuit vs. Packet
• ARPAnet was born in ＵＣＬＡ, California.
• Most professors were not willing to participate
in an ARPA project, because they know the
technology.
• Computers were slow. They could not process
the packets in a timely manner.
• One theorist, Prof Leonard Kleinrock, has
accepted the invitation.
• ARPAnet is the initial stage of the Internet.
7
Vint Cerf, Robert Kahn, Leonard Kleinrock and Lawrence G. Roberts
8
Packet is the right selection?
• IP packet (or datagram)
• Series of packets (or train)
A flow is a series of packets which shares the
same source and destination IP addresses,
port numbers, and the protocol indicator.
• Flow is used for measurement of routers.
• Flow is also used as an object of flow control.
OpenFlow
Future Internet
9
Reduce Power Consumption
•
•
•
•
10
(Dai Mochinaga, 2013)
Packet switching needs 17.52 TWh/yr in 2030. (estimated)
Hybrid approach by packet and optical circuit switch
67.7% down (5.67 TWh/year in 2030)
With CDN 22.2% down (13.6 TWh/year in 2030)
Optical circuit for backbone network 48.4% down
(9.04 TWh/year in 2030)
3. Centralized vs. Distributed
• Background of ARPAnet
Vietnam War
There is no single point of failure.
The Internet is a typical example of distributed
computing. Is it turue?
• Apparently it looks physically distributed.
There is a certain logical center.
E.g. DNS, domain names (tree structure)
OpenFlow switch is welcomed by network
operator because they can control switches
remotely.
11
DNS is a distributed database with a
single ultimate root
root
Logical root is
unique in the
world.
“ ”
…
jp
ac
org
arpa
ietf
in-addr
waseda www
info
goto
piano
piano.goto.info.waseda.ac.jp
int
ip6
…
133
90
18
10
12
OpenFlow
• Video 3 min.+ [click]
13
4. Do-it-yourself vs. Cloud
• Plain old public telephone service (POTS)
smart network
Switching units can do everything.
Phone sets are terminals.
• The Internet
stupid network
PCs and smart phones perform most of jobs.
Network is a pipe.
• Cloud computing (cloud)
Depend on other people or systems
We do not know who they are.
Your application software is not yours.
14
New idea of application set fingerprint
• Smart phones can be identified by
installed application set
© Fukumoto, Waseda Univ.
15
How many fingerprints?
• Found several tens of thousands of fingerprints
• Variation is enough to identify smart phone units
Big Data
1,000
The result shows that the distribution has an
extremely long tail and 92.2% of the smart phone
users have a unique application set fingerprint.
Frequency
100
44,248
fingerprints
10
1
1
© Hu Yao
10
100
1,000
Distinct Fingerprints
10,000
16
5．End-to-end vs. Hop-by-hop
• TCP is based on end-to-end principle.
IP addresses indicate source and destination.
Ethernet (MAC) controls link by link.
• ICN/CCN/NDN takes hop-by-hop approach.
PURSUIT in Europe has end-to-end functions.
• Logical or virtual level attracts our interests.
It is also meaningful to understand the physical
layer which is the basis of our communications.
17
© Masaki Fukushima
Architecture of CCN routers
IP-based network
Source
node
Packet
(address)
Router
Destination
node
Routing table
address 
outgoing link
CCN (Content-Centric Networking）
CCN Router
Like URL
Interest
(data name)
Data
Consumer
name
cached data
Content
Store
name
incoming link
name
outgoing link
Future
Internet
FIB
Pending
Interest
Table
Data
Source
Data
18
Content-Centric Network (CCN)
Content-Centric Network (CCN):
trustworthy information dissemination and retrieval
with cache memories/storages
CCN is one prospective approach for designing future network.
© Hu Yao
19
Content-Centric Network (CCN)
• Two basic data structures
Face 2
– Interest
– Data
CCN node
FIB
• Content Store (CS)
Interest
– to cache Data packets
Interest
Face 0
• Pending Interest Table (PIT)
Consumer
PIT
Face 1
Data
Data
Provider
– to store pending Interest requests
• Forwarding Information Base (FIB)
– to store forwarding rules
CS
An excellent design of FIB is essential both for Interest
and Data packets transfer in CCN.
© Hu Yao
20
Ethernet connector, RJ-45
Pin number | Signal name
Wave form by a Oscilloscope
21
6. De jure vs. De facto
• IETF is a Standardizing body for the Internet
A typical example of de facto standard
It is also called forum standard.
• Slogan of IETF
rough consensus and running code
It is not necessary to get a majority of votes.
• Standard Document
RFC (Request for Comments)
It is sometimes neglected（initial value of TTL）
There published a poem of a meeting.
22
Japanese students worked hard.
Many stories behind the scene of
Internationalized Domain Names (IDN)
Published in 2006.
23
7. Open vs. Trust
• Commercial Internet is allowed for CIX, Commercial
Internet Exchange in 1990. (January, 1991）
• It was a failure that the Internet was
commercialized as was.
• There was no anonymous users in old ARPAnet.
ARPAnet directory has been published until 1983.
(D)ARPA was not happy with free terminals at a
landing of a staircase at Stanford University.
• Encryption
Authentication
24
Authentication
in Internet2
Shibboleth
IdP
ユーザ
SP
Gaku-nin in Japan
No private information
Only the attributes are needed
1. User attempts to access Shibboleth-protected resource on SP site application server.
2, 3, 4. User is redirected to a Where Are You From (WAYF) server, where the user indicates their
home site (IdP).
5. User is redirected to the Handle Service at their IdP.
6, 7 User authenticates at their IdP, using local credentials.
8. Handle service generates unique ID (Handle) and redirects user to Service Provider site's
Assertion Consumer Service (ACS). ACS validates the supplied assertion, creates a session,
and transfers to Attribute Requestor (AR).
9, 10. AR uses the Handle to request attributes from the IdP site's Attribute Authority. The
attribute authority responds with an attribute assertion subject to attribute release policies;
SP site uses attributes for access control and other application-level decisions.
25
8. Array vs. List (Tree)
• Most programming languages have data types:
Array, Matrix, Vector, Table
• Preferable language in Artificial Intelligence
List （Symbolic Programming Language: LISP）
• LISP can handle with variable structures:
UNIX files and directories, Domain Names,
Decision Tree
Tree
26
ＬＩＳＰ
A Japanese textbook
SYMBOLIC PROGRAMMING
Published in1988.
John McCarthy
Sep 4, 1927 – Oct 24, 2011
Father of Artificial Intelligence
27