In the areas of operations research and industrial engineering, the objective function in manyoptimization and decision-making problems involves complicated system measures that need to be evaluated by simulation. The procedures of evaluating and optimizing stochastic systems via simulation present several critical challenges: i) The stochastic systems typically involve uncertainties that are captured by a set of input distributions to be estimated from real-world input data, and the estimation error from the input distributions may propagate and cause uncertainties in the evaluation of the objective function; ii) Under complicated system logic, the exact simulation of the objective function can be computationally expensive or impossible; iii) Simulations need to be conducted on multiple values of the decision variable or across various systems to facilitate optimization and selection. These challenges exacerbate issues about the allocation and management of computational resources. This thesis aims to propose effective algorithm design for simulation-based decision making, with the goal of optimizing the stochastic systems and saving computational resources at the same time. In particular, our contributions are threefold: i) With the objective of selecting the system with the best performance, we propose a general framework to analyze the joint resource allocation problem for collecting input data and generating simulation replications; ii) For continuous optimization via simulation (COvS) problems, we propose gradient-based algorithms that sequentially utilizes multi-resolution approximations to optimize systems for which exact simulation is costly or impossible. iii) We provide new time-parallel simulation algorithms, so that simulation can be executed quickly to inform time-sensitive decisions.