1. Simulation Evaluation of Peer Link Management Protocol
September 2007
Simulation Evaluation of Peer Link Management Protocol
Date:
Authors:
Meiyuan Zhao, Intel Corp.

2. September 2007
Abstract
This submission reports status of ongoing work on evaluating performance of PLM protocol. The preliminary simulation model and experimental results are reported.
Feedbacks and collaboration are welcome.
Meiyuan Zhao, Intel Corp.

3. September 2007
PLM Protocol Overview
Unique link instance: <myMAC, peerMAC, myLinkID, peerLinkID>
Messages binds with the instance identifier
Open/Confirm messages to establish a link
Confirm sent in response to Open
Link established when
Each MP send Open and Confirm messages
Each MP receive Open and Confirm messages
Close messages to tear down a link
Holding state for grace period of close
Holding timer to control
Retry mechanism to manage failures
Retry timer and counter
Additional confirm timer to prevent deadlock and livelock
Fully specified using finite state machine (FSM)
Unique link instance identifiers enable simultaneous link instance handling and avoid deadlock situations.
Meiyuan Zhao, Intel Corp.

4. Performance Evaluation Goals
September 2007
Performance Evaluation Goals
Intensive design consideration is still not fully satisfying
Need performance analysis to validate – protocol behaviors match design goals
Questions
How long does it take to establish a link under normal/abnormal conditions?
How long does it take to tear down a link under normal/abnormal conditions?
What timeout value and retry counter limit can enable robust protocol execution?
Timeout value relationships? (retryTimeout, confirmTimeout, holdingTimeout)
Network overhead of establishing/tearing down a peer link?
Meiyuan Zhao, Intel Corp.

5. PLM Finite State Machine in OPNET
September 2007
PLM Finite State Machine in OPNET
Meiyuan Zhao, Intel Corp.

6. Timers Retry Timer Confirm Timer Holding Timer September 2007
Set to a default value of 40 ms (for a 11Mbps link) after a open message is sent. The timer is reset every time till the dot11MeshMaxRetries expires.
Retrytimeout = return Retrytimeout + (getRandom mod timeout)
Confirm Timer
Theoretically, it can be governed by the equation
Confirmtimeout = (dot11MeshMaxRetries + 1) * Retrytimeout
Need simulation results to validate this setting
Holding Timer
Confirmtimeout = (dot11MeshMaxRetries ) * Retrytimeout
Meiyuan Zhao, Intel Corp.

7. Performance Evaluation Criterion
September 2007
Performance Evaluation Criterion
System model
Two nodes, one link (11Mbps)
Default beaconing and retransmission
Performance metrics
Latency of establish/tear down a link
Success rate of establishing a link
Scenario A
MP1 and MP2 work properly
Examine performance with various radio channel condition
Default: 10% drop rate
Scenario B
MP2 respond significantly slower
Slow processing delay: exponential distribution with  = 5 ms (default)
Meiyuan Zhao, Intel Corp.

8. Latency for Establishing a Link
September 2007
61 %
Mean latency Variance
Case A with 82 Simulation runs, default timeout values, default Maxretries=3
One retry slows down the link establishment ~ 30 ms
Retry 0 – 61% , Retry 1 – 33%, Retry 2 – 6%
Meiyuan Zhao, Intel Corp.

9. September 2007
Success Rate
Case A: Two mesh nodes, 82 Simulation runs, Retry Timeout = 40 ms, Confirm Timeout = 160 ms, Holding Timeout = 120 ms, Maxretries=3
The Success rate was 100% for establishing the links.
The Retry_Counter value is 3 at the Peer Link Management level apart from the MAC level retransmissions.
Success Rate for Scenario B (Abnormal Conditions) has to be analyzed to understand protocol robustness.
Meiyuan Zhao, Intel Corp.

10. Mean Latency for Closing a link
September 2007
Mean Latency for Closing a link
Two setups with same parameters as in Scenario A:
Case 1: Both MPs randomly close the peer link
Case 2: One MP closes the peer link
Mean Latency for closing a link (X state -> Idle state)
The last conclusion suggest an additional experiment to measure how effective the holding timer is:
Measure how many close messages sent by the MP while in the holding state
If the number > 0, this means holding state can really help to speed up the link tear down procedure.\
Mean Latency Variance
Closing a link is ~ 5X slower than establishing a link
Bound failure variance
Meiyuan Zhao, Intel Corp.

11. Confirm Timeout Experiment 1: Experiment 2: Conclusions September 2007
Default radio channel noise (10%)
Vary ConfirmTimeout value [80,200] ms
Evaluate establishment success rate
Result: The confirm timer never expires. 7 retransmission is sufficient to ensure Open frame delivery
Experiment 2:
Vary drop rate [0.1, 0.8]
Fix ConfirmTimeout value: 120 ms
Result: The confirm timer never expires. MAC retransmission reports error and drops the link before confirm timer expires
Conclusions
Value of confirm timeout is not sensitive to network condition
Conjecture: ConfirmTimer used for worst case when no Open frame is caused by peer MP
Meiyuan Zhao, Intel Corp.

12. Next Steps Enhancement and thorough validation of simulation model
September 2007
Next Steps
Enhancement and thorough validation of simulation model
OPNET model
Experiment configurations (channel, processing delay, interference, background traffic…)
More experiments
Experiment 1: Latency and Success Rate in Scenario B
Experiment 2: Understand Retry timeout value’s impact on Success rate
Experiment 3: Identify the tradeoffs on Max Retries and Success rate
Experiment 4: Parameter Negotiation failure handling
Experiment 5: Latency and the Overhead of establishing secure link
Various topology
More than two nodes and heterogeneous nodes
Bootstrapping evaluation
Link establishment behavior
Stability of topology
Meiyuan Zhao, Intel Corp.

13. September 2007
Summary
We have started effort on evaluating PLM and AbbrHS using simulation
Have some interesting preliminary results
Welcome feedbacks and collaboration
Meiyuan Zhao, Intel Corp.

14. September 2007
Questions?
Meiyuan Zhao, Intel Corp.