Transcript Low-Power DoS Attacks in Data Wireless LANs and Countermeasures

Scalable Robust and Secure
Heterogeneous Wireless Networks
Guevara Noubir
College of Computer Science
Northeastern University, Boston, MA
[email protected]
1
The Heterogeneous Future of Wireless
Networks

Ambient intelligence aware of people’s presence, needs, and context

Ubiquitous computing: maintain seamless access to data and services

Nature and man-made disaster: require adequate operational modes




Safety services: better quality of life for elderly and disabled people
The need for the enabling technology

Limitations of current wireless technology:




No integration, QoS, seamless adaptivity, single-hop, limited data rates, battery life
Major issues: scalability, robustness, security
We need novel approaches!
As these applications become more ubiquitous new threats will appear:


Fast recovery through reconfiguration and prioritization of services
Resiliency to denial of service attack
Amplified by: untracability, limited resources (energy and computation power)
Talk focus on networking aspects
2
Outline

Characteristics of heterogeneous wireless networks

Some security aspects heterogeneous wireless networks



Some novel approaches to scalability and robustness




Physical, layer/link, and multi-layer attacks
Multicasting
Cross-layer design
Accumulative Relaying
Universal Network Structures
Conclusion
3
Characteristics








Limited radio spectrum
Shared Medium (collisions)
Limited energy available at the nodes
Limited computation power
Limited storage memory
Unreliable network connectivity
Dynamic topology
Need to enforce fairness
4
Flexibility








Use of various coding/modulation schemes
Use of various transmission power level
Use of multiple RF interfaces
Use of multi-hop relaying
Clustering and backbone formation
Planning of the fixed nodes location
Packets scheduling schemes
Application adaptivity
5
Multihop Heterogeneous Paths
Resource Efficient Paths:
Multirate, Power-Controlled, Contention and Mobility Aware
Cooperating paths:
Distributed MIMO, Accumulative Relaying
Internet
Access Points
Mobile Nodes
Sensor Nodes
Universal Network Design:
Universal Sensors Steiner Tree
Robust Distributed Compression:
Generalized Slepian-Wolf
Cross-layer power controlled MAC
6
Multilayer DoS in Wireless Networks

Physical layer


MAC layer


Jamming of control traffic and mechanisms
Network layer


Smart multilayer aware jammers
Malicious injection/disruption of routing information
Transport layer

Exploiting weaknesses in congestion control
mechanisms
7
Physical Layer Jamming

Leads to:



Network partition
Forcing packets to be routed over chosen paths
Low-Power: cyber-mines
8
Low-Power Physical Layer Jamming

Jamming effort:


IP packet:


Jamming duration/packet duration
1500 bytes = 12000 bits
Uncoded packet:

Jamming effort in the order of 10-4
9
Jamming IEEE802.11 and 802.11b
Modulation/coding
Rate
Packet length
IP packet
Number of bits
needed to jam
Jamming
Efficiency
BPSK
1500*8
1
12000
QPSK
1500*8
2
6000
CCK (5.5Mbps)
1500*8
4
3000
CCK (11Mbps)
1500*8
8
1500
10
Jamming Encoded Data Packets
Link Architecture
Jamming Unreliable
Communication
Jamming ECC Protected
Communication
UDP
UDP
EDP
…
Jamming Interleaved ECC
Protected Communication
UDP
EDP
IDP
JP
JP
JP
>dmin-1/2
UDP: Uncoded Data Packet
JP: Jamming Packet
EDP: Encoded Data Packet
in l codewords
RP: Received Packet
IDP: Interleaved Data Packet
DDP: De-Interleaved Packet
RP
DDP
P
dmin: code minimum
Hamming distace
>dmin-1/2 errors within
a single codeword
11
Traditional Anti-Jamming Techniques
Focus on bit-level
2
P
G
G
R
J
j
jr rj tr Lr B r

S Pt Gtr Grt R 2jr L j B j
Pj:
Gjr:
Grj:
Rtr:
Lr:
Br:

Pt:
Gtr:
Grt:
Rjr:
Lj:
Bj:
transmitter power
antenna gain from transmitter to receiver
antenna gain from receiver to transmitter
distance from jammer to receiver
jammer signal loss
jamming transmitter bandwidth
Spread-Spectrum in military provides:


jammer power
antenna gain from jammer to receiver
antenna gain from receiver to jammer
distance from transmitter to receiver
communication signal loss
communications receiver bandwidth
20-30dB processing gain
Low-power jamming requires:

40dB!
12
Mitigating Physical Layer DoS

Physical Layer:



Link Layer:


Spread-Spectrum
Directional Antennas
Cryptographic Interleaver + Efficient Coding
Routing:


Jamming-free paths
Use of Mobility
13
Proposed Solution for Link Layer
Cryptographic Interleaving
+
Efficient Adaptive Error Correction

For Binary Modulation:


Cryptographic interleaving transforms the
channel into a Binary Symmetric Channel
Capacity of BSC (Shannon):
C  1  H ( p)
C  1  p log 2 ( p)  (1  p) log 2 (1  p)
14
Practical Codes

Low Density Parity Codes:


Very Close to Shannon’s Bound
Best for long packets:

E.g., 16000 bits
Jamming Effort

Code Rate
Shannon Limit
8%
0.5
0.598
Code
Throughput
0.5
17.4%
0.25
0.333
0.25
Non-binary modulation e.g., IEEE802.11b (CCK): transmits 8 bits




Use a Reed-Solomon code with symbols of 8 bits
Maximum length: 256 bytes
Data: k  256bytes
Tolerates: (256-k)/2 errors
15
Conclusion on Physical Layer DoS

Existing Wireless Data Networks are easy targets of physical layer jamming

High transmission power, and spread-spectrum are not enough

Jammer effort in the order of 10-4 for an IP packet

Traditional anti-jamming focuses on bit protection

Cryptographic interleaving and Error Control Codes provide much better
resiliency to Jamming

Additional technique that derive from the J/S ratio: directional antennas

Need adaptivity and careful integration within the network stack
16
Link/MAC Layer DoS

Attack Control Traffic



RACH/Grant CH/BCCH channels in cellular
Authentication (e.g., sending deauth message)
MAC Mechanisms of IEEE802.11:

Reservation:




Backoff:


RTS/CTS are short packets: require less energy to be jammed
NAV: malicious nodes can force nodes to wait for long durations
EIFS: a single pulse every EIFS at high power
Backoff allows an attacker to spend less energy when Jamming
Selecting attacks on MAC/IP addresses
17
DoS on Routing

Malicious nodes can attack control traffic:




Attack goals: disruption or resource consumption
Techniques:









Jamming
Inject wrong information
Black hole: force all packets to go through an adversary node
Rooting loop: force packets to loop and consume bandwidth and
energy
Gray hole: drop some packets (e.g., data but not control)
Detours: force sub-optimal paths
Wormhole: use a tunnel between two attacking nodes
Rushing attack: drop subsequent legitimate RREQ
Inject extra traffic: consume energy and bandwidth
Blackmailing: ruining the routing reputation of a node
Proposed secure routing protocols are still not practical
18
DoS on Transport Layer

Transport layer should be able to differentiate
between:

Congestion



Wireless link packets loss



Due to traffic pattern change: new sessions
Requires source rate reduction
Due to mobility and interference
Requires modulation/coding/power/path change
Malicious nodes


Selective jamming and disruptions
Requires isolation of malicious nodes and dead areas
19

Protection against DoS in wireless
networks requires a careful cross-layer
design
20
Secure Multicasting
[with Kaya, Lin, Qian – Funded by Draper]

Goal:



Secure multicast applications:





Communication over a multihop wireless ad hoc network
Limited computation power, and energy
Services:


Secure remote tracking of mobiles
Sharing sensed data
Military: Data/Video streaming from UAV, multicasting of command decisions
Specificity:


Securely and efficiently acquire and disseminate time varying information
Example: location information
Authentication, integrity, confidentiality, revocation, group key management
Approach:

Overlay network of mobile nodes build secure multicast tree
21
Prototype Application
iPAQ PDA
Pharos Compact Flash GPS
IEEE 802.11 PCMCIA card
22
Ad Hoc vs. Wired Multicast

Wireless:



Mobility:



Higher packet loss
Necessity of frequent discovery of paths
Multihop:



Unreliable links
Loss of a packet results in node exclusion and necessity for new
join request
Cost of multicast depends on number of hops
Major factor because of radio resources scarcity
Ad hoc:


Limited computation: nodes cannot manage large groups
Active nodes
23
Group Management
1
2
5
3
4
9
6
7
10
8
11
12
x
13
Source
y Group member
24
Issues and Results

Efficient tree construction and maintenance

Under mobility greedy algorithms can be very good



Public key encryption is costly:


Close to optimal trees O(log n) in theory but in practice 1.5
approximation
Minimize broadcast cost and tree maintenance
Memory can be traded with computation
Revocation in an infrastructure-less environment
25
Novel Approaches to Scalability and
Robustness

Scalability to large networks with limited
resources requires novel techniques


Make use of specificity of the environment
Use techniques from a combination of fields:



Graph theory, linear programming, network flow
Information theory, coding theory
Accurate simulation and modeling tools

Accumulative relaying

Universal network design
26
Accumulative Power Relaying
[with Chen, Jia, Liu, Sundaram]
B
G
A

C
Reliable reception
Partial reception
Problem:

Determine a feasible schedule [(N1, P1), …, (Nk, Pk)] that
minimizes total energy consumption
27
Accumulative Power Relaying
[with Chen, Jia, Liu, Sundaram]
B
G
A

C
Reliable reception
Partial reception
Problem:

Determine a feasible schedule [(N1, P1), …, (Nk, Pk)] that
minimizes total energy consumption
28
Accumulative Relaying


Very similar to the relay problem in information
theory and still open in it’s general form
Simpler than the general relay problem:




Every energy optimal sequence can be transformed
into a canonical form called wavepath
In a wavepath each node in the sequence activates
its next hop neighbor and only its next hop neighbor
Finding a minimum energy wavepath is still NP-hard
for arbitrary networks
Heuristic for building a wavepath can achieve more
than 40% energy saving on a Euclidian plane
29
Universal Multicast Tree
[with Jia, Lin, Rajaraman, Sundaram]

Problem:




Given a graph G (V, E), n nodes, and a root/sink
Build a tree T such that for all subgroups T leads to a low weight
tree for all subgroups (through pruning)
CostT ( S )
}
i.e., build T that minimizes the stretch Max{
S V
OPT ( S )
Applications:




Environment: sensor network where routing is difficult
Dissemination: efficient multicasting to dynamic groups
Aggregation from changing groups
Distributed queries
30
Universal Tree for the Euclidian Space

Results:


Polynomial time algorithm to build a universal tree
with stretch O(log k) [where k is the size of the
selected subgroup]
Hardness result: no algorithm can build a tree with
stretch lower O(log n/loglog n)
31
Universal Structures

Other results:



Algorithm for a universal tree for non-Euclidian
metrics with poly-logarithmic stretch
Poly-logarithmic stretch for the universal Traveler
Salesman Problem
Extensions:



Universal tree for energy cost
Universal tree for planar, range limited wireless
communication
Fault-tolerant network structures
32
Conclusion

We live in an exciting era:




Wireless physical layer is capable of providing high
data rates
Software flexibility
Computation power
This provides the building blocks to enable
ubiquitous networking



Creates new threats
Need smart adaptive control of the physical layer
Need to deal with security and robustness in a
scalable way
33
Universal Tree for the Euclidian Space

Results:



Polynomial time algorithm to build a universal tree with stretch
O(log k) [where k is the size of selected subgroup]
Hardness result: no algorithm can build a tree with stretch lower
O(log n/loglog n)
Definition:

Level i of v: Liv = {u: 2i-1 < d(u, v) 2i}
L4r

Algorithm:


L3r
Divide V –{r} into L1r, L2r, …, LlogDr,
Run A(Lir, r) in parallel
34

Algorithm A(U, r)

L = {r}

Repeat




For every uU, let Iu denote the level of u to its nearest
neighbor in L;
Let I = max {Iu : u U}
Let H = {u U : Iu = I}
Let H’  H s.t.




u, v H’ d(u,v)  2I-1,
u H\H’ v H’ s.t. d(u,v) < 2I-1
u H’ output edge (u, nearest-neighbor(u))
L = L  H’; U = U\H’;
Until no edge output;
35
Universal Tree Algorithm
H
H’
36
Universal Tree Algorithm
H
H’
37
Universal Tree Algorithm
H
H’
38
Universal Tree Algorithm
H
H’
39