Improving the Parameter Dependency for High-Multiplicity Scheduling on Uniform Machines

Abstract

We address scheduling problems on uniform machines with high-multiplicity encoding, introducing a divide and conquer approach to assess the feasibility of a general Load Balancing Problem (LBP). Via reductions, our algorithm can also solve the more well-known problems $Q\|C_{\max}$ (makespan minimization), $Q\|C_{\min}$ (santa claus) and $Q\|C_{\text{envy}}$ (envy minimization). State-of-the-art algorithms for these problems, e.g. by Knop et al. (Math.\ Program.\ '23), have running times with parameter dependency $p_{\max}^{O(d^2)}$, where $p_{\max}$ is the largest processing time and $d$ is the number of different processing times. We partially answer the question asked by Kouteck\'y and Zink (ISAAC'20) about whether this quadratic dependency of $d$ can be improved to a linear one: Under the natural assumption that the machines are similar in a way that $s_{\max}/s_{\min} \leq p_{\max}^{O(1)}$ and $\tau\leq p_{\max}^{O(1)}$, our proposed algorithm achieves parameter dependency $p_{\max}^{O(d)}$ for the problems ${Q\|\{C_{\max},C_{\min},C_{\text{envy}}\}}$. Here, $\tau$ is the number of distinct machine speeds. Even without this assumption, our running times achieve a state-of-the-art parameter dependency and do so with an entirely different approach.