Transcript Adversarial Models for Wireless Communication Andrea W. Richa Arizona State University SIROCCO'13, Andrea Richa.

Adversarial Models for
Wireless Communication
Andrea W. Richa
Arizona State University
SIROCCO'13, Andrea Richa
1
Motivation
Channel availability hard to model:
● Mobility
● Packet injection
● Temporary Obstacles
● Background noise
● Physical Interference
● Co-existing networks
● Jammer
SIROCCO'13, Andrea Richa
2
Physical layer jamming
● A physical jammer listens to the open medium and
broadcasts in the same frequency band as network
– can lead to significant disruption of communication at
low cost for the jammer
honest nodes
SIROCCO'13, Andrea Richa
jammer
3
Motivation
Channel availability hard to model:
● Mobility
● Packet injection
● Temporary Obstacles
● Background noise
● Physical Interference
● Co-existing networks
● Jammer
all contribute to some form
of “background noise”
SIROCCO'13, Andrea Richa
4
Motivation
Ideal world:
: noise level
Background
noise
0
time
Usual approach adopted in theory.
SIROCCO'13, Andrea Richa
5
Motivation
Ideal world:
: noise level
Background
noise
OR
0
time
Usual approach adopted in theory.
SIROCCO'13, Andrea Richa
6
Motivation
Ideal world:
: noise level
Background
noise
OR (bounded, predictable)
0
time
Usual approach adopted in theory.
SIROCCO'13, Andrea Richa
7
Motivation
: noise level
Real world:
background
noise
0
time
How to model this???
SIROCCO'13, Andrea Richa
8
Our Approach: Adversarial
Jamming
X
X
X
X
Background noise (microwave, radio signal, etc.)
Intentional jammer
Temporary Obstacles (cars etc.)
Co-existing networks …
X
SIROCCO'13, Andrea Richa
9
Our Approach: Adversarial
Jamming
●Idea: model unpredictable behaviors via adversary
(a.k.a. adversarial jammer)!
X
X
X
X
X
SIROCCO'13, Andrea Richa
10
Overview
● Adaptive adversary
– Single-hop scenario
– Simple (yet powerful) idea
– MAC protocol
● Reactive adversary
– Fairness
●
●
●
●
Adaptive adversary in multi-hop networks
Application: Leader Election
Other adversarial models
Future work
SIROCCO'13, Andrea Richa
11
Wireless communication model
● single frequency: e.g., sensor nodes
● at each time step, a node may decide to transmit a
packet (nodes continuously contend to send packets)
● a node may transmit or sense the channel at any time
step (half-duplex)
● when sensing the channel a node v may
– sense an idle channel
– receive a packet
– sense a busy channel
(due to interference or
adversarial jamming)
SIROCCO'13, Andrea Richa
v
12
Single-hop wireless network
● [Awerbuch, R., Scheideler, PODC’08]
● n reliable honest nodes and a jammer (adversary); all
nodes within transmission range of each other and of
the jammer
jammer
SIROCCO'13, Andrea Richa
13
Adaptive adversary
● knows protocol and entire history
● (T,λ)-bounded adversary, 0 < λ < 1: in any time
window of size w ≥ T, the adversary can jam ≤ λw
time steps
steps jammed by adversary
other steps
w
01…
SIROCCO'13, Andrea Richa
14
Constant-competitive protocol
● a protocol is called constant-competitive against a
(T,λ)-bounded adversary if the nodes manage to
perform successful transmissions in at least a
constant fraction of the steps (w.h.p. or on
expectation), for any sufficiently large number of
steps
successful transmissions
steps jammed by adversary
other steps (idle channel, message collisions)
w
01…
SIROCCO'13, Andrea Richa
15
Our main contribution
● symmetric local-control MAC protocol that is
constant-competitive against any (T,1-ε)-bounded
~
adaptive adversary after Ω (T / ε) steps w.h.p., for
any constant 0<ε<1 and any T.
● energy efficient:
– converges to bounded amount of energy
consumption due to message transmissions by
nodes under continuous adversarial jamming (ε=0)
● fast recovery from any state
SIROCCO'13, Andrea Richa
16
Pros and Cons
Pros:
● no prior knowledge of global parameters
– nodes do not know ε
● no IDs needed
Cons:
● nodes know rough estimate γ=O(1/(log T + loglog n))
– allow for superpolynomial change in n and polynomial
change in T over time
● fair channel use is not guaranteed
– we will see how to fix that later
SIROCCO'13, Andrea Richa
17
Physical Layer Traditional
Jamming Defenses
● spread spectrum & frequency hopping:
– Many references in the literature (specially more
applied work)…
– rely on broad spectrum (large number of available
frequencies). However, sensor nodes or common
wireless devices based on 802.11 have narrow
spreading factors
– Our approach is orthogonal to broad spectrum
techniques, and can be used in conjunction with those.
SIROCCO'13, Andrea Richa
18
MAC Layer Jamming Defenses
● random backoff:
– adaptive adversary too powerful for MAC protocols based on
random backoff\tournaments (including the 802.11 standard
[Bayrataroglu, King, Liu, Noubir, Rajaraman, Thapa, INFOCOM’08])
● [Gilbert, Guerraoui, Newport, OPODIS’06]: cannot handle adaptive
adversaries with high jamming rate
– more general scenario (adversary can also introduce
malicious messages)
– nodes know n
– not energy efficient
● Others (channel surfing, coding strategy,etc.): also cannot
handle adaptive adversary
SIROCCO'13, Andrea Richa
19
Overview
● Adaptive adversary
– Single-hop scenario
– Simple (yet powerful) idea
– MAC protocol
● Reactive adversary
– Fairness
●
●
●
●
Adaptive adversary in multi-hop networks
Application: Leader Election
Other adversarial models
Future work
SIROCCO'13, Andrea Richa
20
Simple (yet powerful) idea
● each node v sends a message at current time step with
probability pv ≤ pmax, for constant 0 < pmax << 1.
p
= ∑ pv (aggregate probability)
qidle = probability the channel is idle
qsucc = probability that only one node is transmitting
(successful transmission)
● Claim. qidle . p ≤ qsucc ≤ (qidle . p)/ (1- pmax)
~
if (number of times the channel is idle) =
(number of
successful transmissions)
p = θ(1)
qsucc = θ(1) !
(what we want!)
SIROCCO'13, Andrea Richa
21
Basic approach
● a node v adapts pv based only on steps when an idle
channel or a successful message transmission are
observed, ignoring all other steps (including all the
blocked steps when the adversary transmits!)
time
idle steps
successful transmissions
steps jammed by adversary
steps where collision occurred but no jamming
SIROCCO'13, Andrea Richa
22
Basic approach
● a node v adapts pv based only on steps when an idle
channel or a successful message transmission are
observed, ignoring all other steps (including all the
blocked steps when the adversary transmits!)!
time
idle steps
successful transmissions
steps jammed by adversary
steps where collision occurred but no jamming
SIROCCO'13, Andrea Richa
23
Naïve protocol
Each time step:
● Node v sends a message with probability pv . If v
does not send a message then
– if the wireless channel is idle then pv = (1+ γ) pv
– if v received a message then pv = pv /(1+ γ)
(Recall that γ = O(1/(log T + loglog n)). )
SIROCCO'13, Andrea Richa
24
Problems
● Basic problem: Aggregate probability p could be too
large.
– all time steps blocked due to message collisions w.h.p.
time
idle steps
successful transmissions
steps jammed by adversary
steps where collision occurred but no jamming
SIROCCO'13, Andrea Richa
25
Problems
● Basic problem: Cumulative probability p could be too
large.
– all time steps blocked due to message collisions w.h.p.
time
idle steps
successful transmissions
steps jammed by adversary
steps where collision occurred but no jamming
SIROCCO'13, Andrea Richa
26
Problems
● Basic problem: Cumulative probability p could be too
large.
– all time steps blocked due to message collisions w.h.p.
● Idea: If more than T consecutive time steps without
successful transmissions (or idle time steps), then
reduce probabilities, which results in fast recovery of p.
● Problem: Nodes do not know T. How to learn a good
time window threshold?
– It turns out that additive-increase additive-decrease is
the right strategy!
SIROCCO'13, Andrea Richa
27
MAC protocol
● each node v maintains
– probability value pv ,
– time window threshold Tv , and
– counter cv
● Initially, Tv = cv = 1 and pv = pmax (< 1/24).
● synchronized time steps (for ease of explanation)
● Intuition: wait for an entire time window (according to
current estimate Tv) until you can increase Tv
SIROCCO'13, Andrea Richa
28
MAC protocol
In each step:
● node v sends a message with probability pv . If v
decides not to send a message then
– if v senses an idle channel, then pv = min{(1+ γ)pv , pmax}
– if v successfully receives a message, then pv = pv /(1+ γ)
and Tv = max{Tv - 1, 1}
● cv = cv + 1. If cv > Tv then
– cv = 1
– if v did not receive a message successfully in the last Tv
steps then pv = pv /(1+ γ) and Tv = Tv +1
SIROCCO'13, Andrea Richa
29
Our results
● Let N = max {T,n}
● Theorem. Our MAC protocol is constant-competitive
under any (T,1-ε)-bounded adversary if the protocol
is executed for Ω(log N . max{T,log3 N/(ε γ2)}/ ε)
steps w.h.p., for any constant 0<ε<1 and any T.
SIROCCO'13, Andrea Richa
30
Proof sketch
● Show competitiveness for time frames of F =
θ((log N . max{T,log3 N/(ε γ2)}/ ε) many steps
If we can show constant competitiveness for any
such time frame of size F, the theorem follows
● Use induction over the number of sufficiently large
time frames seen so far. We subdivide each frame:
I’
I
f = θ(max{T,log3 N/(ε γ2)})
F = (log N / ε) . f
SIROCCO'13, Andrea Richa
31
Proof sketch
● p > 1/(f2(1+γ)2√f) and Tv < √F, in each subframe I’
w.h.p.
● p<12 and p>1/12 within subframe I’ with moderate
probability (so that adaptive adversarial jamming not
successful)
● Constant throughput in I’ with moderate probability
● Over a logarithmic number of subframes, constant
throughput in frame I of size F w.h.p.
SIROCCO'13, Andrea Richa
32
Continuous jamming
● Moreover, under a more powerful adversary
that
~
can perform continuous jamming (after Ω(T)
steps):
● Lemma. The total energy consumption (sending
out messages) during
an entire continuous
~
jamming attack is O(√T), independent of the
length of the attack.
● Exhaust adversary’s energy resources
SIROCCO'13, Andrea Richa
33
Overview
● Adaptive adversary
– Single-hop scenario
– Simple (yet powerful) idea
– MAC protocol
● Reactive adversary
– Fairness
●
●
●
●
Adaptive adversary in multi-hop networks
Application: Leader Election
Other adversarial models
Future work
SIROCCO'13, Andrea Richa
34
Reactive adversary
● [R.,Scheideler, Schmid, Zhang, ICDCS’11]
● Fully adaptive adversary that in addition can quickly
observe the channel at the current time step, before
deciding to jam
– i.e., the adversary has some knowledge about the
random choices at current time step
– Distinguishes between idle and non-idle time steps, but
cannot distinguish between successful transmissions
and collisions (e.g., if packets are encrypted)
SIROCCO'13, Andrea Richa
35
AntiJam: Reactive jammingresistant MAC protocol
● Need to “synchronize” transmission probabilities pv,
as well as counters cv and Tv
– Piggyback pv, cv, Tv to a message sent by v
– Better understanding on how aggregate probability
changes every time step
– Achieve fairness for free (basically all nodes have the
same transmission probability)!
● Once nodes are synchronized (plus some other
smaller changes), one can show that the basic
protocol is also robust against reactive adversaries
SIROCCO'13, Andrea Richa
36
Our results
Theorem. The AntiJam protocol achieves:
● fairness: the channel access probabilities among
nodes do not differ by more than a factor of (1+γ) after
the first message was sent successfully.
● eθ(1/ ε2)-competitiveness w.h.p., under any (T,1-ε)bounded reactive adversary if the protocol is executed
~
for Ω(T/ε) steps w.h.p., for any constant 0<ε<1 and any
T
(constant in throughput now depends on ε )
SIROCCO'13, Andrea Richa
37
Proof sketch: Fairness
● Fact:
– Right after u sends a message successfully along with
the tuple (pu ,cu ,Tu), (pv, cv, Tv) = (pu / (1+ γ), cu,Tu) for
all receiving nodes v, while the sending node values
stay the same. In particular, for any time step t after a
successful transmission by node u, (cv, Tv) = (cw, Tw)
for all nodes v and w V
– This implies that after a successful transmission, the
access probabilities of any two nodes in the network
will never differ by more than a factor
in the
future.
SIROCCO'13, Andrea Richa
38
AntiJam: Throughput for ε = 0.5
ε = 0.5
SIROCCO'13, Andrea Richa
39
AntiJam: Convergence Time
SIROCCO'13, Andrea Richa
40
AntiJam: Fairness
SIROCCO'13, Andrea Richa
41
AntiJam vs. Non-reactive
protocol: Fairness
SIROCCO'13, Andrea Richa
42
AntiJam vs 802.11
SIROCCO'13, Andrea Richa
43
Overview
● Adaptive adversary
– Single-hop scenario
– Simple (yet powerful) idea
– MAC protocol
● Reactive adversary
– Fairness
●
●
●
●
Adaptive adversary in multi-hop networks
Application: Leader Election
Other adversarial models
Future work
SIROCCO'13, Andrea Richa
44
Multihop wireless networks
Physical interference
Jammed
Powerful jammer
SIROCCO'13, Andrea Richa
45
Signal-to-Interference-NoiseRatio (SINR)
● A message sent by node u is received at node v iff
P/d(u,v)
N + w in S P/d(w,v)
>
- N: Gaussian variable for background noise
- S: set of transmitting nodes
- : constant that depends on transmission scheme
- d(x,y): Euclidean distance between x and y
● Well-accepted model (in theory, at least  ) for physical
interference
SIROCCO'13, Andrea Richa
46
Multihop Adversarial Model
● (B,T)-bounded adaptive adversary: has an overall
noise budget of BT that it can use to increase the
noise level at node v and that it can distribute among
the time steps as it likes.
● At any point in time, the adversary makes
independent decisions for each node on whether to
jam it (provided it does not exceed its noise budget
over a window of size T).
● many noise phenomena can be covered under this
model
SIROCCO'13, Andrea Richa
47
Throughput in multihop networks
● [Ogierman, R., Scheideler, Schmid, Zhang, 2013]:
SINR model
● Also achieve constant throughput :
Throughput =
 # successful msgs receivedby v
 # non - jammedtimestepsfor v
v
v
(Earlier, [R., Scheideler, Schmid, Zhang, DISC’10]: unit-disk
graph model.)
SIROCCO'13, Andrea Richa
48
Overview
● Adaptive adversary
– Single-hop scenario
– Simple (yet powerful) idea
– MAC protocol
● Reactive adversary
– Fairness
●
●
●
●
Adaptive adversary in multi-hop networks
Application: Leader Election
Other adversarial models
Future work
SIROCCO'13, Andrea Richa
49
Leader Election in Adversarial
(Single-hop) Networks
●[R., Scheideler, Schmid, Zhang, MobiHoc’11]
●Our goal: select a leader among the nodes
leader
●Challenges: we may start in any state; there is adaptive
adversary
SIROCCO'13, Andrea Richa
50
Self-stabilizing Leader Election
● Goal: design a self-stabilizing protocol that elects a
single node as the leader, irrespective of the jamming
activity
● Challenges:
– a leader node should let the others (followers or other
leaders) know that he is still around
– the followers should be able to notice when there is no
leader in the network
SIROCCO'13, Andrea Richa
51
Leader Election
Why is leader election difficult under an adaptive
jammer?
Example: exponential/polynomial backoff
: jamming activity
channel
activity
(expected)
: messages
0
time
constant success probability
SIROCCO'13, Andrea Richa
52
Why is leader election difficult
under an adaptive jammer?
Example 1: exponential/polynomial backoff
: jamming activity
channel
activity
(expected)
: messages
0
time
constant success probability
SIROCCO'13, Andrea Richa
53
Why is leader election difficult
under an adaptive jammer?
Example 2: reserved leader slot to notify nodes about
leader
: jamming activity
channel
activity
(expected)
: leader message
0
time
SIROCCO'13, Andrea Richa
54
Why is leader election difficult
under an adaptive jammer?
Example 2: reserved leader slot to notify nodes about
leader
: jamming activity
channel
activity
(expected)
: leader message
0
time
SIROCCO'13, Andrea Richa
55
Other adversarial modeling in
wireless networks
● Adversary: used to model external world
● More bening:
– Control packet injection rates
– Control mobility
● Intentionally disruptive:
– jammers
● More disruptive: malicious adversaries
– Undermine security
– Control Byzantine nodes (introduce fake messages)
SIROCCO'13, Andrea Richa
56
Other adversarial modeling in
wireless networks
● Adversarial packet injection/queueing:
– [Chlebus, Kowalski, Rokicki, PODC’06],
[Andrews, Jung, Stolyar, STOC’07],
[Anantharamu, Chlebus, Rokicki, OPODIS’09],
[Chlebus, Kowalski, Rokicki, Distributed Computing’09],
[Lim, Jung, Andrews, INFOCOM’12]
● Multi-channel access with adversarial jamming:
– [Dolev, Gilbert, Guerraoui, Newport, DISC’07],
[Anantharamu, Chlebus, Kowalski, Rokicki, SIROCCO’11],
[Dolev,Gilbert, Khabbazian, Newport, DISC’11],
[Daum, Gilbert, Kuhn, Newport, PODC’12],
[Ghaffari, Gilbert, Newport, Tan, OPODIS’12]
SIROCCO'13, Andrea Richa
57
Other adversarial modeling in
wireless networks
● Broadcasting\Gossiping with adversarial jamming:
– [Dolev, Gilbert, Guerraoui, Newport, DISC’07],
[Dolev,Gilbert, Khabbazian, Newport, DISC’11],
[Daum, Gilbert, Kuhn, Newport, PODC’12],
[Ghaffari, Gilbert, Newport, Tan, OPODIS’12]
● Capacity Maximization with adversarial jamming:
[Dams, Hoefer, Kesselheim, unpublished]
● Malicious adversary:
– [Dolev, Gilbert, Guerraoui, Newport, DISC’07],
[Gilbert, Young, PODC’12]
● Infection spreading with adversarial mobility:
[Wang,Krishnamachari, ]
● Etc.
SIROCCO'13, Andrea Richa
58
Future work: Adversarial
Jamming
● Jamming-resistant protocols with power control:
– Increasing power increases chance that signal will
overcome jamming activity, however…
– Increasing power also generates more interference…
– Also adapt noise threshold level?
● How about reactive jammers in multihop
environments, under SINR?
● Can the protocols be modified so that no rough bound on
n and T are required?
– stochastic/oblivious jammers: Simpler to handle? E.g.,
a constant gamma seems to work fine here.
● Other applications of the MAC protocol (e.g., broadcast)?
SIROCCO'13, Andrea Richa
59
Future Work
● What would be your application of an adversary in
wireless communication? 
SIROCCO'13, Andrea Richa
60
Collaborators
● Baruch Awerbuch (John Hopkins); Stefan Schmid
(TU Berlin\Telekom Labs); Christian Scheideler and
Adrian Ogierman (U. of Paderborn); Jin Zhang
(Google)
● Papers available from my webpage (for recent – and
maybe not so recent  -- submissions, please send
me email) at www.public.asu.edu/~aricha
[email protected]
SIROCCO'13, Andrea Richa
61
Thank you!
Questions?
SIROCCO'13, Andrea Richa
62
SIROCCO'13, Andrea Richa
63
SIROCCO'13, Andrea Richa
64