Note on a class of admission control policies for the stochastic knapsack problem

In this note we discuss a class of exponential penalty function policies recently proposed by Iyengar and Sigman for controlling a stochastic knapsack. These policies are based on the optimal solution of some related deterministic linear programs. By finding explicitly their optimal solution, we reinterpret the exponential penalty function policies and show that they belong to the class of threshold policies. This explains their good practical behavior, facilitates the comparison with the thinning policy, simplifies considerably their analysis and improves the bounds previously proposed