This paper focuses on the edge computing of critical tasks in smart factories. Considering the high latency risk that may occur due to the uncertainty of the channel and the limitation of computing resources in the edge computing

the conditional value at risk (CVaR) is used to capture the tail of the delay distribution. By using the convexity and translational equivariance of CVaR

we introduce the upper bound of CVaR for the delay. Furthermore

we provide an optimal framework design including server selection and computing resource allocation

aiming at jointly optimizing the average delay and the upper bound of CVaR. Simulation results validatethe effectiveness of the algorithm for high-latency distribution characterization. From the simulation results

the proposed strategy not only improves the reliability of the computing

but also reduces the high risk value of the delay.