arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2203.01741cs.DS
收藏

一致机器上高重数调度的结构性结果

Structural Results for High-Multiplicity Scheduling on Uniform Machines

Hauke Brinkop, David Fischer, Klaus Jansen

AI总结:

针对一致机器高重数调度问题,提出基于子实例分数解的邻近技术,获得新结构性结果,并给出匹配已知最优的算法及嫉妒最小化的fpt算法。

摘要:

以最大处理时间 $p_{max}$ 和不同作业处理时间数量 $d$ 为参数,我们针对一致机器上的高重数调度问题,在最大完工时间最小化 ($C_{max}$) 和 Santa Claus ($C_{min}$) 目标下提出了一种邻近技术,以获得这些问题的新结构性结果。我们方法的新颖之处在于,我们仅处理某个子实例的分数解,而该子实例本身事先未知。尽管与通常的邻近技术不同,分数解的构造和计算并非在多项式时间内完成,但这使得我们能够提出一个相对较强且通用的邻近性陈述。最终,这使我们能够根据分数解预先分配作业,将每台机器和每种作业类型需要分配的作业数量减少至 $p_{max}$ 的多项式,本质上返回有限数量(至多 $O(p_{max}^{O(d^2)})$)的核,每个核对应一个(猜测的)子实例。我们可以利用结构性结果获得运行时间为 $p_{max}^{O(d^2)}poly|I|$ 的算法,与 Knop 等人 (Oper. Res. Lett. '21) 迄今已知的最佳结果相匹配。此外,我们针对一致机器上高重数设置下的嫉妒最小化 $C_{envy}$ 提出了运行时间为 $p_{max}^{O(d^2)} poly |I|$ 的算法,证明该问题在 $p_{max}$ 中是 fpt。最后,我们还提出了一种通用机制,将所谓负载均衡问题的 Configuration ILP 中最大系数限制为 $(dp_{max})^{O(d)}$,我们希望这对算法开发有所助益。

英文摘要:

Parameterizing by the largest processing time $p_{max}$ and the number of different job processing times $d$, we propose a proximity technique for High-Multiplicity Scheduling on Uniform Machines for the objectives Makespan Minimization ($C_{max}$) and Santa Claus ($C_{min}$) to obtain new structural results for these problems. The novelty in our approach is that we deal with a fractional solution for only a sub-instance, where the sub-instance itself is not known a priori. While the construction and computation of the fractional solution -- in contrast to usual proximity techniques -- is not done in polynomial time, this also allows us to formulate a comparably strong and general proximity statement. Eventually, this allows us to reduce the number of jobs that need to be distributed to a polynomial in $p_{max}$ for each machine and job type, by preassigning jobs according to the fractional solution, essentially returning a bounded number (at most $O(p_{max}^{O(d^2)})$) of kernels, one for each (guessed) sub-instance. We can use our structural results to obtain an algorithm with running time is $p_{max}^{O(d^2)}poly|I|$, matching the best-known so far by Knop et al. (Oper. Res. Lett. '21). Moreover, we propose an $p_{max}^{O(d^2)} poly |I|$ time algorithm for Envy Minimization $C_{envy}$ in the High-Multiplicity Setting on Uniform Machines, showing that this problem is \textsc{fpt} in $p_{max}$. Eventually, we also propose a general mechanism to bound the largest coefficient in the Configuration ILP for so called \emph{Load Balancing Problems} by $(dp_{max})^{O(d)}$, which we hope to be of interest for the development of algorithms.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑