Transcript 03-wiojw - PURSUIT project

Lecture 3
- Why the Internet only just works
- What can we do about it?
D.Sc. Arto Karila
Helsinki Institute for Information Technology (HIIT)
[email protected]
17.09.2012
M.Sc. Mark Ain
Helsinki Institute for Information Technology (HIIT)
[email protected]
T-110.6120 – Special Course in Future Internet Technologies
1
*** NOTICE ***
 The following readings are now mandatory:
 DONA
 I3
 SEATTLE
 The following lectures are cancelled:
 Tue 25.09
 Mon 01.10
 Mon 08.10
17.09.2012
2
Computer networking
 Developed for mainframes
(on the left ENIAC and on the right IBM S/360)
 Sharing devices: computers, mass memory, printers
etc. with addresses
 Point-to-point traffic
between two devices
or network interfaces
 The old paradigm still
lives even though the
world around has
completely changed
Picture source: IDG News Service
17.09.2012
3
Contents
 What’s wrong with the Internet today
 What can we do about it?
17.09.2012
4
What’s wrong with the Internet
today?
1)
2)
3)
4)
5)
6)
7)
8)
9)
Sender empowerment
Endpoint-centrism
Infrastructure trustworthiness
Application deployment
Congestion control
Inter-domain routing
Multi-homing
Address space
Identifier-locator unification
 Mobility
10) QoS
11) Multicast and caching
17.09.2012
5
Sender empowerment
 In the 1960s…
 Computers were large and very resource-limited by
modern standards
 Data was stored, input, and output by physical means
e.g. punch cards; you took your data with you
 ARPANet was organized to address the need for
efficient resource-sharing amongst computers of the
time (NOT content sharing!)
The send-receive communication paradigm was
simple, arguably obvious, and well-suited for the
purposes of the time.
17.09.2012
6
Sender empowerment
 Today…
 Computers are small, resources are abundant, content is at the
forefront of (user) attention
 Send-receive may not be optimal
Result: SPAM, DoS, concealment (firewalls,
middleboxes etc.) etc.
17.09.2012
7
Quick look: SPAM (2009)
Estimates…
 Upwards of $130 billion USD in global losses
(2009 USD, average EUR-USD exchange rate,
unadjusted for inflation 2012)
 ~62 trillion messages per year
 Server-side filtering could hypothetically save
~135TWh of energy per year = ~17 million
metric tons of CO2 emissions
17.09.2012
8
Endpoint-centrism
 The future: content-centrism
 Get “x” simply by asking for “x”… the network
finds “x” and delivers it innately based on “x”
17.09.2012
9
Endpoint-centrism
 The reality: endpoint-centrism
 Get “x” by asking WHERE is “x”, receiving
response “y”, then fetching “x” from ”y”.
17.09.2012
10
Infrastructure trustworthiness
 Trust is irrational – however, there is a mathematical
foundation for it
 The Internet was developed for a community where
everybody was assumed trustworthy
 Now that the Internet is used by everybody, it is vital to
enable communication between parties that don’t trust
each other
 We need mechanisms by which people and companies
can build and evaluate trust
 Good reputation can be made an asset worth protecting
 Combining privacy and reputation is challenging
17.09.2012
11
Infrastructure trustworthiness
17.09.2012
12
Application deployment
“There is a vicious circle – application developers will
not use a new protocol (even if it is technically
superior) if it will not work end-to-end; OS vendors
will not implement a new protocol if application
developers do not express a need for it; NAT and
firewall vendors will not add support if the protocol is
not in common operating systems; the new protocol
will not work end-to-end because of lack of support
in NATs and firewalls.”
- M. Handley
17.09.2012
13
Application deployment
 E2E principle should in theory make it easy to
deploy applications over many hosts without
worrying about interactional problems across
network
 Unfortunately, this is not the case, as
evolutionary developments and patchwork
solutions (e.g. NAT) have broken E2E on
many levels
17.09.2012
14
Application deployment
Internet stagnancy feedback loop
17.09.2012
15
Internet stagnancy feedback
loop
 A “chicken and the egg” problem
 Discussion: what happens first?
17.09.2012
16
Congestion control
 It was implemented at the transport layer
(TCP, mid-late 1980’s) because it was too late
in the Internet’s development to change the
core protocol stack
 TCP congestion control is largely successful,
but incremental, and plagued by
insufficiencies (next slide)
17.09.2012
17
Congestion control
 TCP problems…
Only reacts to congestion, does not proactively
prevent it; insufficient convergence times
2. Changing application and per-flow requirements 
variety of security, performance, and compatibility
problems
3. Poor performance over links with high B*D product;
too slow to converge, too aggressive backoff
4. Not designed for wireless environments; TCP reacts
to packet loss as though it were congestion.
1.
17.09.2012
18
Inter-domain routing
 The current inter-operator routing protocol BGP4 does not fulfill modern requirements but there
is no successor to it in sight
 Tier-1 operators (AT&T, MCI, Sprint, C&W etc.)
are a group of about a dozen global operators
with mutual peering agreements
 In Practice they form a cartel, which wants to
cement the market and is not advocating
development
17.09.2012
19
Inter-domain routing
 BGP policy routing mechanisms were a reaction to
an abundance of users and the potential for
commercial competition
 BGP operation centered on…
 AS’s are separate and equal
 Route-path information is commercial sensitive
 BGP attempts to avoid unecessarily releasing
route-path information  subject to
misconfiguration, vulnerability, slow convergence
etc.
17.09.2012
20
Multi-homing
 Reliability, transparent-failover, and load-
sharing often necessitate multi-homed
connections
 Problem: the mere presence of multiple IP
prefix announcements on a wide-scale
removes the benefits of hierarchical IP
aggregation
17.09.2012
21
Address space
 IPv4 once though inexhaustible
 Despite evolutionary patchwork (NAT, DHCP,
improved allocation policies, reclamation
projects etc.), IPv4 is exhausted
 IPv6???
17.09.2012
22
IPv6

IPv6 was defined in 1995 and expected to spread fast

It is still hardly used in Western countries

The main improvement of IPv6 is moving from 32-bit to 128-bit addresses

IPv6 was defined at a time when nobody could foresee all of the uses and
needs of the Internet that we have now
?

Planned
Actual
The transition to IPv6 will be a long one and it won’t solve most of the
problems
17.09.2012
23
Identifier-locator unification
17.09.2012
24
Identifier-locator unification
 Mobility raises 5 fundamental problems…
1) Locating the mobile host and/or service
2) Preserving communication
3) Disconnecting gracefully
4) Hibernating efficiently
5) Reconnecting quickly
The root cause of problems 1 and 2 is IP
semantic overload i.e. identifier-locator
unification; the last 3 are largely unaddressed!
17.09.2012
25
QoS
 DiffServ and IntServ are NOT built-in to the
network or protocol independent
 DiffServ does not provide end-to-end
guarantees
 IntServ requires cooperation amongst
providers and network state
 How do we provide protocol-independent
QoS, built-in to the architecture, preserving
E2E, without necessarily requiring network
state etc?
17.09.2012
26
Multicast and Caching
17.09.2012
27
In summary…
 No major changes have been made to the core protocols
of the Internet since 1993 (CIDR)
 The core protocols of the Internet are ossified while the
needs have developed significantly
17.09.2012
28
Contents
 What’s wrong with the Internet today
 What can we do about it?
17.09.2012
29
Evolution vs. revolution
 The Internet has developed from the 1970’s in an
evolutionary way, with no big changes
 As concluded before, this has led into a situation where it
is very hard to make changes to the core protocols
 Among researchers and developers of the Internet, there
is a growing opinion that something fundamental has to
be done at some point
 It the Internet was to be designed from the scratch, it
would probably become very different from what it has
evolved to
17.09.2012
30
Evolution vs. revolution
 Various clean-slate solutions are current research topics
and some of them may lead into a new Internet
 It is possible that all the protocol layers, including the
Internet Protocol, will change
 However, it looks like any new solution would have to be
able to operate as overlay above the existing IP
infrastructure, in order to have a change to proliferate
 The publish/subscribe paradigm (pub/sub) is one of the
most promising new paradigms (for more information
see www.psirp.org and www.fp7-pursuit.eu
17.09.2012
31
Microeconomics

Over the past ten years, microeconomics have grown in
importance

We need economic mechanisms that encourage people to
do good for the community

The Internet was developed with public funds for research
and education without any commercial considerations

If we want to inject resources into the network, it must be
possible for the party paying for them to also receive (some
of) the revenues

We need to create ways for companies and people to
improve their own economies by doing things beneficial for
the community
17.09.2012
32
For tomorrow…
 READ
 Van Jacobson, Diana K. Smetters, James D. Thornton, Michael F.
Plass, Nicholas H. Briggs, and Rebecca L. Braynard. 2009.
Networking named content. In Proceedings of the 5th
international conference on Emerging networking experiments and
technologies (CoNEXT '09). ACM, New York, NY, USA, 1-12.
DOI=10.1145/1658939.1658941
http://doi.acm.org/10.1145/1658939.1658941
17.09.2012
33
Thank you for your attention!
Questions? Comments?
17.09.2012
34