Selfish MAC Layer Misbehavior in Wireless Networks Pradeep Kyasanur and Nitin H. Vaidya 2005 IEEE Reviewed by Dean Chiang

How is network contention resolved? Distributed Contention Function (DCF)  Relies on protocol adherence from all of the network. I’m transmitting guys! Ok everyone better wait the same amount! You got it! …What if I don’t want to…?

Problem Definition Selfish Host Backoff = rand[1,1] Normal Host Backoff = rand[0,CW] Normal Channel Usage Channel Usage with a Selfish Host t t

Problem Challenges Difference between a small backoff and differing usage by each node? Node channel conditions are different. A node can claim an idle channel.

Solution Modified MAC Protocol Goal: retain performance of while ensuring detection of misbehavior in a short interval Assumptions:  Misbehaving senders only. Receivers are assumed to be trusted.  No collusion between sender and receiver

Protocol Modifications - Backoff Receiver assigns random backoff and sends it in the CTS and ACK. Receiver is in control.

Protocol Modifications - Retransmission Sender adds attempt number to RTS Receiver estimates expected backoff from attempt number.

Penalty Scheme Penalize if:  B_actual < Alpha * B_expected (Alpha = 0.8) Penalty  Add to next backoff by an amount proportional to deviation and then some. Still not adhering? Identify as a selfish host. (Last W packets that deviated over total threshold T slots) Receiver is in control. Misbehavior with backoff cannot be spoofed by senders.

What happens after identification? Possibilities MAC Layer Receiver exiles sender by not responding to RTS. Network Layer Network routes around host.

Misdiagnosis Receiver senses a busy channel from 2 hops away when sender does not. Sender sends to receiver, receiver thinks sender did not adhere to assigned backoff. Solution: receiver does not count a busy channel for transmission, only during overheard RTS/CTS.

Results: Diagnostic Accuracy Notes: Low misdiagnosis rate Does not fully resolve selfish behavior problem. Only alleviates high end of deviation. Notice there is a threshold of misbehavior that can go undetected.

Results: Throughput Throughput comparison to Adheres closely to throughput of

Results: Fairness Fairness Index comparison to Fairness decreases with # of nodes Fairness deviates more when background traffic is high (TWO-FLOW)

Results: What about crafty clients? Correct Diagnosis % Protocol is worse at detecting adaptive misbehavior Throughput gain is not large because of penalty scheme HOWEVER

Pros and Cons Pros  Fast diagnosis, no need to collect tons of data  Low Misdiagnosis rate  Retains fairness and throughput of  Reduces workaround potential Cons  Additional overhead  Backwards compatibility  Spoofed MAC addresses

Addendum: Optimal Threshold Authors show a proof for an optimal threshold Threshold T is a function of assigned back-off value and # of slots. However, authors chose a static threshold in simulations. This suggests the data could be better.

Discussion/Consideration What does the police do? Why did the government create a police force? At what intrinsic layer does the government’s control end?

Discussion/Consideration Is this really the best place to handle selfish behavior? PC, Driver Wireless Network Adapter Regulated by IEEE Not Regulated by IEEE Base Station