Research On Multi-Terminal Oriented Computation Offloading In Edge Computing

Computation offloading is one of the main research hotspots in edge computing,which has received extensive attention from scholars at home and abroad.Computation offloading in edge computing has a series of important academic achievements while it still faces critical challenges such as service demand diversity,information absence,limited edge resources,and unstable edge node status with the mature commercial use of 5G and a large number and a wide variety of terminals with different capabilities accessing the network.Considering the diversity of terminal amounts,types,and capabilities,this thesis systematically studies multiterminal oriented computation offloading in edge computing in terms of statistical delay guarantee,distributed cooperation,and reliability enhancement.The main contributions are as follows:(1)In practice,the wireless network dynamics makes the deterministic latency guarantee at the high cost of strict resource provisioning,incurring high terminal energy consumption.To address this issue,this thesis proposes a statistical latency guarantee based computational offloading approach.First,this approach proposes a statistical latency guarantee based computational offloading framework,where the correlation between the statistical latency guarantee and the computation offloading strategy is quantified based on probability analysis and queue theory.Second,this approach formulates the task offloading problem as a mixed-integer nonlinear programming problem.Third,to reduce the approach complexity,the problem is decomposed and solved based on convex optimization theory and the Gibbs sampling method,which works iteratively.Finally,simulation results show that the proposed approach greatly reduces the energy consumption of mobile terminals.(2)The unpredictability of wireless network status and the absence of competing terminals’ strategies incurs low decision-making efficiency.To address this issue,this thesis proposes a distributed cooperation based online computation offloading approach.First,this approach formulates the resource competition among terminals as a game.Second,this approach introduces the reinforcement learning mechanism into the game model.In the game,each terminal learns its strategy from historical feedback through trial-and-error interactions with other terminals and the dynamic wireless environment.Third,to verify the applicability of the method in practical scenarios,this approach analyzes the influence of feedback delay and multi-frequency-band access.Finally,simulation results show that the proposed algorithms greatly improve the decisionmaking efficiency of terminals.(3)To enhance the computation offloading reliability under unstable edge nodes status,this thesis proposes a multi-armed-bandit based online computation offloading approach.First,this approach constructs a joint model of latency and reliability based on probability analysis and extreme event theory.Second,since the task completion latency is unpredicted under unstable edge node status,this approach adopts the multi-armed bandit framework to analyze the sequential decision-making problem.Third,this approach makes computation offloading decisions based on the dual tradeoff(exploitation-exploration tradeoff and performance-cost tradeoff)with upper-bounded learning loss.Finally,simulation results show that the proposed algorithm significantly reduces latency with the reliability guarantee.