“In the realm of collaborative offloading between user devices, edge networks, and cloud data centers, this paper delves into the offloading decision and system parameter optimization. By classifying tasks into delay-sensitive and delay-tolerant categories and applying distinct policies, the authors propose a cloud-edge-device collaborative task offloading scheme. Utilizing a four-dimensional Markov chain and the Gauss-Seidel method, they derive the system's stationary probability distribution, presenting the blocking rate and average delay for both task types. Numerical experiments validate the scheme's effectiveness and optimize the access threshold for minimal system cost.”