Cross-Layer Optimization for State Update in Mobile Gaming Yang Yu*, Zhu Li*, Larry Shi*, Yi-Chiun Chen+, Hua Xu+ *Application Research Center, Motorola Labs +Motorola Networks & Enterprise Oct. 16 2007 Wayne State University
Motivation Application trend: Large scale MMOG on mobile devices Gaming requirements: Efficient state update is crucial for satisfactory gaming experience Network constraints: Limited bandwidth, variable network delay and channel condition Query Privacy in Wireless Sensor Networks November 15, 2018 2/21
Problem Scenario Down-link state update from one WiMAX access point to all clients Dead-reckoning algorithm for state update Pre-specified bandwidth limitation Real-time channel quality and network delay feedback Query Privacy in Wireless Sensor Networks November 15, 2018 3/21
Goal and Contributions Goal: Minimize gaming state distortion with an efficient state update mechanism that adapts to network states: Limited bandwidth Network delay Contributions: Characterize the traffic-distortion tradeoffs of gaming behavior Off-line optimization and a history-based prediction method for on-line adaptation Validation and evaluation using real game traces Query Privacy in Wireless Sensor Networks November 15, 2018 4/21
WiMAX Link Model OFDM Symbol Number FCH Burst Burst #4 #1 Burst #5 1 2 3 4 5 6 7 8 … . N-1 1 … . M-1 1 FCH Burst 2 Burst #4 #1 ACK Burst #5 Sub-Channel Logical Number DL MAP Preamble Burst #6 Burst #2 CQI Burst UL MAP #7 Burst #3 Ns Downlink Subframe TTG Uplink Subframe Query Privacy in Wireless Sensor Networks November 15, 2018 5/21
Dead-Reckoning Algorithm Client A Client B Client C Actual move @ A δ Predicted move @ B & C Location update time Updates from A to server Updates from server to B & C time Distance difference >= δ Query Privacy in Wireless Sensor Networks November 15, 2018 6/21
Traffic-Distortion Tradeoffs – Theoretical Intuition Actual location Location function Estimated location Location difference Update triggered when Query Privacy in Wireless Sensor Networks November 15, 2018 7/21
Traffic-Distortion Tradeoffs – Real Game Traces Query Privacy in Wireless Sensor Networks November 15, 2018 8/21
User Diversity Fixed update threshold large variations in user distortion and update traffic Query Privacy in Wireless Sensor Networks November 15, 2018 9/21
Off-Line Problem Formulation Assumption Game traces at the t-th second are known a priori Given For all n clients, distortion function, Di(δi), and traffic function, Ri(δi), The constellation size for each client i, αi, and OFDMA parameters, Q (frame rate) and h (number of sub-carriers per sub-channel) Bandwidth constraint, B, in terms of total available clusters per frame Find distortion threshold vector δ = {δ1, δ2, …, δn} and cluster allocation vector b = {b1, b2, …, bn}, so as to minimize Subjec to Query Privacy in Wireless Sensor Networks November 15, 2018 10/21
Lagrangian Relaxation λ = 0.08 λ = 0.16 λ: Lagrangian multiplier separate λ = 0.08 λ = 0.16 Time complexity: Λ: domain of λ Δ: domain of δ Query Privacy in Wireless Sensor Networks November 15, 2018 11/21
On-Line Adaptation Explore temporal locality of gaming behavior Historical data-based prediction Our simulation results show one second history performs the best for a driving game Query Privacy in Wireless Sensor Networks November 15, 2018 12/21
Evaluation Setup Baselines: Off-line optimal allocation Uniform policy: same bandwidth for all clients Proportional policy: same δ for all clients bandwidth allocation proportional to extent of state changes Real 40 second traces for a driving game with 32 vehicles Δ: [0.2, 10] meters Update packet size: 200 bytes Query Privacy in Wireless Sensor Networks November 15, 2018 13/21
WiMAX Link Quality and Adaptive Coding Query Privacy in Wireless Sensor Networks November 15, 2018 14/21
Main Results 200 total clusters (peak 4.3 Mbps), 10 ms network delay, 200 frames per second, 24 sub-carriers per sub-channel NABA performed close to Optimal Both NABA and Optimal were able to efficiently utilize the bandwidth constraint Query Privacy in Wireless Sensor Networks November 15, 2018 15/21
Impact of Bandwidth Constraint 100 to 500 clusters (peak 2.2 – 10.8 Mbps) Distortion dropped with BW constraint NABA approaches to Optimal Both NABA and Optimal were able to efficiently utilize the bandwidth constraint Query Privacy in Wireless Sensor Networks November 15, 2018 16/21
Impact of Network Delay Network delay: 10 – 100 ms Distortion increased with network delay NABA performed close to Optimal Both NABA and Optimal were able to efficiently utilize the bandwidth constraint Query Privacy in Wireless Sensor Networks November 15, 2018 17/21
Impact of Rounding On average, <2% increase in distortion Query Privacy in Wireless Sensor Networks November 15, 2018 18/21
Related Works Study of the impact of network delay and packet loss on state consistency Zhou 2004 (ACM trans. , Yasui 2005 (NetGames) Study of dead-reckoning Suitability, Pantel 2002 (NetGames) Accuracy, Aggarwal 2005 (NetGames) Our paper is the first effort to model the traffic-distortion tradeoffs to facilitate bandwidth allocation in a wireless environment Query Privacy in Wireless Sensor Networks November 15, 2018 19/21
Conclusion Revealed the traffic-distortion tradeoffs in dead-reckoning algorithm Formulation & off-line optimization of the bandwidth allocation problem History-based prediction for on-line adaptation Validation and evaluation via real game traces Query Privacy in Wireless Sensor Networks November 15, 2018 20/21
Q & A Query Privacy in Wireless Sensor Networks November 15, 2018 21/21