https://hal.science/hal-02943776Nguyen, Kim ThangKim ThangNguyenIBISC - Informatique, BioInformatique, Systèmes Complexes - UEVE - Université d'Évry-Val-d'Essonne - Université Paris-SaclayAn improved approximation algorithm for scheduling under arborescence precedence constraintsHAL CCSD2020Lagrangian DualityPrecedence ConstraintsScheduling[INFO.INFO-RO] Computer Science [cs]/Operations Research [cs.RO]Davesne, Frédéric - Algorithmique en ligne au delà des approches traditionnelles - - OATA2015 - ANR-15-CE40-0015 - AAPG2015 - VALID - 2020-09-20 19:20:082021-12-13 09:17:172020-09-20 19:20:08enConference papers10.4230/LIPIcs.MFCS.2020.841We consider a scheduling problem on unrelated machines with precedence constraints. There are m unrelated machines and n jobs and every job has to be processed non-preemptively in some machine. Moreover, jobs have precedence constraints; specifically, a precedence constraint j ≺ j' requires that job j' can only be started whenever job j has been completed. The objective is to minimize the total completion time. The problem has been widely studied in more restricted machine environments such as identical or related machines. However, for unrelated machines, much less is known. In the paper, we study the problem where the precedence constraints form a forest of arborescences. We present a O((log n)2/(log log n)3)-approximation algorithm - that improves the best-known guarantee of O((log n)2/log log n) due to Kumar et al. [12] a decade ago. The analysis relies on a dual-fitting method in analyzing the Lagrangian function of non-convex programs.