630 likes | 717 Views
Malicious Motes and Suspicoius Sensors: Byzantine Interference in Wireless Networks. Seth Gilbert February 13, 2006. TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A A A A A A A A A A. Alice, Bob, and Collin. The Basic Problem:.
E N D
Malicious Motes and Suspicoius Sensors:Byzantine Interference in Wireless Networks Seth Gilbert February 13, 2006 TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AAAAAAAAAAA
Alice, Bob, and Collin The Basic Problem: Alice likes chocolate! Bob likes ice cream! Bob likes zebras?? !!?%%? Bob Alice
Sensor Networks • Devices: • Berkeley Motes (TinyOS / TinyDB / etc.) • “Smart Dust”
Sensor Networks • Devices: • Berkeley Motes (TinyOS / TinyDB / etc.) • “Smart Dust” • Properties of Small Devices: • Radio Broadcast • Limited power • Limited computation • Limited storage
Sensor Networks • Proposed Applications: • Environmental Monitoring • e.g., Great Duck Island • System control • e.g., Dam valves at a hydroelectric plant • e.g., Damping vibrations on rockets • Intelligent Highways
Malicious Behavior • Physical Devices: • Attacked by malicious adversary • Hacked • Motes are easy to reprogram • Attacker deploys his own devices • Fake sensors confuse real network • Malfunctioning • Motes are fragile
Malicious Behavior • Communication: • Corrupted by interference • Overwhelmed by attacker • Cannot necessarily distinguish between good/bad messages • Disrupted by attacker • denial-of-service attack
Malicious Behavior Challenges: • Local Communication • Only nearby devices can communicate • Collision prone • Susceptible to contention, EM interference, etc. • Unauthenticated • It may be impossible to identify the sender.
Malicious Behavior Challenges: • Local Communication • Collision prone • Unauthenticated
Malicious Behavior Challenges: • Local Communication • Collision prone • Unauthenticated
Malicious Behavior Challenges: • Local Communication • Collision prone • Unauthenticated
Wireless Ad Hoc Networks • Cryptography is hard: • Public-key crypto: • Computationally intensive • Bandwidth intensive • Symmetric-key crypto: • Slow message dissemination • Energy intensive • Key dissemination??
Today: Overview • How do you cope with malicious devices in wireless networks?
Today: Overview • How do you cope with malicious devices in wireless networks? • How little can we restrict the power of the Byzantine nodes? • What is the trade-off between restricting the power of the Byzantine nodes and the efficiency with which we can computer?
Today: Overview • How do you cope with malicious devices in wireless networks? • Part I: Multi-hop grid wireless networks. Highly restricted adversary. Reliable, authenticated communication. • Part II: Single-hop wireless networks. Bounded-collision adversary. Unreliable communication
Byzantine Generals [LSP’82] • Reliable Broadcast: • Single source s with message m. • n-1 receivers. • Each receiver should receive message m. • Byzantine Adversaries: • Arbitrarily malicious.
Byzantine Generals [LSP’82] • Reliable Broadcast: • Agreement • All nodes receive the same message. • Validity • If the source is correct, then every node receives the message broadcast by the source. • Termination • All nodes eventually receive a message, or null.
Classical Results • Impossibility Results: • If network is asynchronous, then impossible. [FLP] • If t≥n/3 then impossible. [LSP’82] • Algorithms: • If tn/3 then possible in t+1 rounds. [LSP’82] • If 2-cast channel & tn/2 then possible. [FM’00]
Classical Results • Graph Results: • If network is not (t+1)-connected, then impossible. [LSP’82] • If network is (2t+1)-connected, then possible. [D’82]
Part I: Overview • Model • Wireless sensors deployed in a grid. • Lower Bound • Impossible if too many corrupt nodes. • Upper Bound • Flooding-based algorithm. • Bounded Collisions
Grid Net Model • Broadcast Properties: • Synchronous • Each node knows its own location • Radius r broadcast • L1 norm. • Results also hold in L1and L2 norms.
Grid Net Model • Collisions: • If 2 neighbors broadcast, then collision.
Grid Net Model • Collisions: • If 2 neighbors broadcast, then collision. • Assume broadcast schedule. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Grid Net Model • Collisions: • If 2 neighbors broadcast, then collision. • Assume broadcast schedule. • Min size: (2r+1)2 • Not optimally efficient! • Focus on feasibility. • Honest nodes never cause collisions. 1 2 3 4 5 6 7 1 2 3 4 5 6 7 1 8 9 10 11 12 13 14 8 9 10 11 12 13 14 8
Grid Net Model • Byzantine nodes: • Problem 1: Impossible for any t=(n). • Example: Assume tn/100, n¸4800
Grid Net Model • Byzantine Nodes: • Problem 1: Impossible for any bound on t. • Assume locally-bounded adversary. • For every neighborhood of size (2r+1)(2r+1), there are at most t corrupted nodes.
Grid Net Model • Byzantine Nodes: • Problem 1: Impossible for any bound on t. • Assume locally-bounded adversary. • For every neighborhood of size (2r+1)(2r+1), there are at most t corrupted nodes.
Grid Net Model • Byzantine Nodes: • Problem 2: Collisions. • Impossible if t=4.
Grid Net Model • Byzantine Nodes: • Problem 2: Collisions. • Byzantine nodes must follow schedule. • For example, cannot corrupt MAC layer. • We will weaken this assumption later. • Thus, Byzantine nodes cannot cause collisions.
Grid Net Model • Byzantine Nodes: • At most t in every neighborhood. • Cannot cause collisions. • Otherwise, arbitrary behavior.
Main Result Theorem: Reliable broadcast is possible if and only if: ¼r2¼1/4 of a broadcast neighborhood
Mini-Bibliography • Koo, Broadcast in radio networks tolerating Byzantine adversarial behavior. PODC, 2004. • Bhandari, Vaidya, On reliable broadcast in a radio network. PODC 2005. • Bhandari, Vaidya, On reliable broadcast in a radio network: A simplified characteriziation. UIUC-TR 2005. • Koo, Bhandari, Katz, Vaidya, Reliable broadcast in radio networks: The Bounded collision case. PODC 2006.
First Attempt: How many corrupted nodes? ¼ 1/2 in neighborhood Theorem 1: Broadcast is impossible if . Lower Bound
Better Bound: Lower Bound Assume r=6.
Better Bound: How many corrupted nodes? ¼ 1/4 in neighborhood Lower Bound Assume r=6.
Better Bound: How many corrupted nodes? ¼1/4 in neighborhood Theorem 2: Broadcast is impossible if . Lower Bound Assume r= 5.
Algorithm • Reliable Broadcast: • Agreement • Validity • Termination • Assume: • Basic idea:Flooding. • Each node broadcasts everything in each round. • When enough data is received, then decide.
Algorithm • Rule 1:Source sends message m. • If node receives m directly from the source, then it chooses(m).
Algorithm • Rule 2:When a node chooses(m), then it broadcasts COMMITTED(m). • When a node receives t+1COMMITTED(m) messages, then it chooses(m).
Algorithm • Rule 2:t+1COMMITTED(m))choose(m). Example:
Algorithm • Rule 2:t+1COMMITTED(m))choose(m). Example:
Algorithm “relay” • Rule 3: When a node receives COMMITTED(m) from nodei, it broadcasts HEARD(m,i). • When a node receives t+1 • COMMITTED(m) messages and • HEARD(m,i) messages • where all the senders and relays are distinct and in one neighborhood, then choose(m). m HEARD(m) COMMITTED(m)
Proof Assume t= 3. • Agreement: • All neighbors of the source choose the same message m by Rule 1.
Proof • Agreement: • Assume by contradiction that i is the first correct node to choose m’m. • Node i receives m’ from t+1 distinct paths. • Only t can be corrupt. • Hence some correct node sent m’. Contradiction.
Proof Assume t= 3. • Agreement:
Proof • Validity: • If the source is correct and broadcasts m, all neighbors of the source choosem. • By agreement, every node choosesm.
Proof • Termination: • Need to show that every node receives t+1COMMITTED or HEARD messages. • Proof by induction. Directly: Indirectly: