Transcript michalis

COMPUTER
Science & Engineering
Scalable Fault-Tolerant Networking:
Re-evaluating The Network Layer
Michalis Faloutsos
Srikanth Krishnamurthy
C.V. Ravishankar
Jakob Eriksson
COMPUTER
Science & Engineering
Security and Reliability Challenges
 Security vs. fault tolerance vs. performance
A security enhancement creates new vulnerabilities
Example:
if we allow collaborative blocking of attacker, a new attack
is to collaborative block good guys
Scalability is critical
Self-configuration
Distributed
solutions
Localization of overhead
Network operation as a control-theory loop
Net. management and planning as a data mining problem
How
many nodes are within 200m of each other? (Ravishankar, FTN)
DARPA Workshop
2
COMPUTER
Science & Engineering
The Vision: What Lies Ahead
 Meganode networking



Pockets of wireless connectivity
Community based connectivity (rural areas)
Commercial interest: Starbucks, European cell phone companies
 Ubiquitous and plug-n-play connectivity
 Distributed peer-to-peer approach:

No central control, no single point of failure
DARPA Workshop
3
COMPUTER
Science & Engineering
PROBLEM: Are We Ready For This?
Now is the time to pop the question:
How
would we design a network from scratch?
The Principle: the IP address is also the ID of a node!
Initial
Internet design does not consider mobility!
Problems: mobile IP, multicast routing, dynamic address allocation
Routing information cannot scale: BGP routing entries 150000
DARPA Workshop
4
COMPUTER
Science & Engineering
PeerNet: a Novel Network Layer
application
application
transport
transport
Network
Network
Link
Link
physical
physical
The innovation: separate node identity from address
Analogy:
postal service ID = name, address = street address
Not an overlay: An alternative to the IP Network Layer
Partially funded under DARPA FTN
DARPA Workshop
5
COMPUTER
Science & Engineering
PeerNet: The Overview
The innovation: permanent nodeID =/= transient address
The address reflects network location
Consequences:
Routing
is simplified: given address, I know where you are
Nodes with similar addresses are “near” each other
Challenges:
Address
allocation: When I move, change the address
ID to Address mapping: Given an ID, find the address
DARPA Workshop
6
COMPUTER
Science & Engineering
The Address Tree in PeerNet
Addressed can be though of as leafs in a binary tree
Address = network location
Nodes
of a subtree are a connected subgraph
Address reallocation and tree balancing
DARPA Workshop
7
COMPUTER
Science & Engineering
How Routing Works
Scalability through information abstraction
Check destination address one bit a time
Route packet to the appropriate subtree
Routing
state: O(log N) for a well balanced tree
DARPA Workshop
8
COMPUTER
Science & Engineering
Important PeerNet Characteristics
Efficient scalable support for multicasting and anycasting
Exploit
the virtual address tree to “establish” a multicast tree
Loop-free routing
Efficient
loop-avoidance (log N bits per path)
DARPA Workshop
9
COMPUTER
Science & Engineering
Open Issues
Security and authentication in a purely distributed world
Dynamic control loop:

Topology -> Address allocation -> Movement -
Evaluate performance in realistic scenarios
DARPA Workshop
10
COMPUTER
Science & Engineering
Conclusions
Security as the result of well engineered system
Efficiency,
scalable, self-organizing
Time to re-evaluate our network architecture
Future
networks =/= static Internet
Our approach: a new network layer
Address
=/= identity
Address = network location
PeerNet: this could work!
An
implementation is on its way
DARPA Workshop
11
COMPUTER
Science & Engineering
Thank you
www.cs.ucr.edu/
www.cs.ucr.edu/~michalis
DARPA Workshop
12