Nearly-Linear Time Approximate Scheduling Algorithms

We study nearly-linear time approximation algorithms for non-preemptive scheduling problems in two settings: the unrelated machine setting, and the identical machine with job precedence constraints setting. The objectives we study include makespan, weighted completion time, and $L_q$ norm of machine loads. We develop nearly-linear time approximation algorithms for the studied problems with $O(1)$-approximation ratios, many of which match the correspondent best known ratios achievable in polynomial time. Our main technique is linear programming relaxation. For problems in the unrelated machine setting, we formulate mixed packing and covering LP relaxations of nearly-linear size, and solve them approximately using the nearly-linear time solver of Young. We show the LP solutions can be rounded within $O(1)$-factor loss. For problems in the identical machine with precedence constraints setting, the precedence constraints can not be formulated as packing or covering constraints. To achieve the claimed running time, we define a polytope for the constraints, and leverage the multiplicative weight update (MWU) method with an oracle which always returns solutions in the polytope. Along the way of designing the oracle, we encounter the single-commodity maximum flow problem over a directed acyclic graph $G = (V, E)$, where sources and sinks have limited supplies and demands, but edges have infinite capacities. We develop a $\frac{1}{1+\epsilon}$-approximation for the problem in time $O\left(\frac{|E|}{\epsilon}\log |V|\right)$, which may be of independent interest.