Scheduling jobs to two machines subject to batch arrival ordering

The problem considered is to assign n jobs to two processors so as to minimize the total flow time, with the constraint that a predetermined partial ordering (induced by batch arrivals) must be preserved within the subset of jobs assigned to each processor. An efficient algorithm of time 0(n⁵) is developed, and computational experience is reported.     

A note on allocating jobs to two machines

A recent article in this journal by Mehta, Chandrasekaran, and Emmons [1] described a dynamic programming algorithm for assigning jobs to two identical parallel processors in a way that minimizes the average delay of these jobs. Their problem has a constraint on the sequence of the jobs such that any group of jobs assigned to a processor must be processed in the order of the sequence. This note has two purposes. First, we wish to point out a relationship between this work and some prior work [2]. Second, we wish to point out that Mehta, Chandrasekaran, and Emmons formulation, slightly generalized, can be used to find the optimum assignment of jobs to two machines in a more general class of problems than they considered including a subclass in which the jobs are not constrained to be processed in a given sequence.     

Index policies for the routing of background jobs

Arriving (generic) jobs may be processed at one of several service stations, but only when no other (dedicated) jobs are waiting there. We consider the problem of how to route these incoming background jobs to make best use of the spare service capacity available at the stations. We develop an approximative approach to Whittle's proposal for restless bandits to obtain an index policy for routing. The indices concerned are increasing and nonlinear in the station workload. A numerical study testifies to the strong performance of the index policies developed. © 2004 Wiley Periodicals, Inc. Naval Research Logistics, 2004     

Optimal scheduling of reader—systems

We consider a reader—writer system consisting of a single server and a fixed number of jobs (or customers) belonging to two classes. Class one jobs are called readers and any number of them can be processed simultaneously. Class two jobs are called writers and they have to be processed one at a time. When a writer is being processed no other writer or readers can be processed. A fixed number of readers and writers are ready for processing at time 0. Their processing times are independent random variables. Each reader and writer has a fixed waiting cost rate. We find optimal scheduling rules that minimize the expected total waiting cost (expected total weighted flowtime). We consider both nonpreemptive and preemptive scheduling. The optimal nonpreemptive schedule is derived by a variation of the usual interchange argument, while the optimal schedule in the preemptive case is given by a Gittins index policy. These index policies continue to be optimal for systems in which new writers enter the system in a Poisson fashion. © 1998 John Wiley & Sons, Inc. Naval Research Logistics 45: 483–495, 1998     

Determining optimal sizes of bounded batches with rejection via quadratic min‐cost flow

In this article, we consider a single machine scheduling problem, in which identical jobs are split into batches of bounded sizes. For each batch, it is allowed to produce less jobs than a given upper bound, that is, some jobs in a batch can be rejected, in which case a penalty is paid for each rejected job. The objective function is the sum of several components, including the sum of the completion times, total delivery cost, and total rejection cost. We reduce this problem to a min‐cost flow problem with a convex quadratic function and adapt Tamir's algorithm for its solution. © 2017 Wiley Periodicals, Inc. Naval Research Logistics 64: 217–224, 2017     

On single-machine sequencing with order constraints

A collection of jobs is to be processed by a single machine. Each job has a cost function associated with it which may be either linear or exponential, costs accruing when a job is completed. The machine may be allocated to the jobs according to a precedence relation. The problem is to find a strategy for allocating the machine which minimizes the total cost and which is consistent with the precedence relation. The paper extends and simplifies some previous work done by Sidney.     

An optimal critical level policy for inventory systems with two demand classes

Traditional inventory systems treat all demands of a given item equally. This approach is optimal if the penalty costs of all customers are the same, but it is not optimal if the penalty costs are different for different customer classes. Then, demands of customers with high penalty costs must be filled before demands of customers with low penalty costs. A commonly used inventory policy for dealing with demands with different penalty costs is the critical level inventory policy. Under this policy demands with low penalty costs are filled as long as inventory is above a certain critical level. If the inventory reaches the critical level, only demands with high penalty costs are filled and demands with low penalty costs are backordered. In this article, we consider a critical level policy for a periodic review inventory system with two demand classes. Because traditional approaches cannot be used to find the optimal parameters of the policy, we use a multidimensional Markov chain to model the inventory system. We use a sample path approach to prove several properties of this inventory system. Although the cost function is not convex, we can build on these properties to develop an optimization approach that finds the optimal solution. We also present some numerical results. © 2008 Wiley Periodicals, Inc. Naval Research Logistics, 2008     

Scheduling to minimize the total compression and late costs

We consider a single-machine scheduling model in which the job processing times are controllable variables with linear costs. The objective is to minimize the sum of the cost incurred in compressing job processing times and the cost associated with the number of late jobs. The problem is shown to be NP-hard even when the due dates of all jobs are identical. We present a dynamic programming solution algorithm and a fully polynomial approximation scheme for the problem. Several efficient heuristics are proposed for solving the problem. Computational experiments demonstrate that the heuristics are capable of producing near-optimal solutions quickly. © 1998 John Wiley & Sons, Inc. Naval Research Logistics 45: 67–82, 1998     

Minimizing the average deviation of job completion times about a common due date

This paper considers a single-machine scheduling problem in which penalities occur when a job is completed early or late. The objective is to minimize the total penalty subject to restrictive assumptions on the due dates and penalty functions for jobs. A procedure is presented for finding an optimal schedule.     

Scheduling stochastic jobs with asymmetric earliness and tardiness penalties

We consider a stochastic counterpart of the well-known earliness-tardiness scheduling problem with a common due date, in which n stochastic jobs are to be processed on a single machine. The processing times of the jobs are independent and normally distributed random variables with known means and known variances that are proportional to the means. The due dates of the jobs are random variables following a common probability distribution. The objective is to minimize the expectation of a weighted combination of the earliness penalty, the tardiness penalty, and the flow-time penalty. One of our main results is that an optimal sequence for the problem must be V-shaped with respect to the mean processing times. Other characterizations of the optimal solution are also established. Two algorithms are proposed, which can generate optimal or near-optimal solutions in pseudopolynomial time. The proposed algorithms are also extended to problems where processing times do not satisfy the assumption in the model above, and are evaluated when processing times follow different probability distributions, including general normal (without the proportional relation between variances and means), uniform, Laplace, and exponential. © 1997 John Wiley & Sons, Inc. Naval Research Logistics 44, 531–557, 1997.     

Bounds on parallel-processing of queues with multiple-phase jobs

Currently, both the hardware and software designs of many large computing systems aim at improved system performance through exploitation of parallelism in multiprocessor systems. In studying these systems, mathematical modelling and analysis constitute an important step towards providing design tools that can be used in building such systems. With this view the present paper describes a queueing model of a multiprocessor system operating in a job-shop environment in which arriving jobs consist of a random number of segments (sub-jobs). Two service disciplines are considered: one assumes that the sub-jobs of a given job are capable of parallel operation on different processors while the other assumes that the same sub-jobs must be operated in a strictly serial sequ'snce. The results (in particular, the mean number in the system and waiting time in queue) obtained for these two disciplines are shown to be bounds for more general job structures.     

Analysis of algorithms for two‐stage flowshops with multi‐processor task flexibility

In this article we introduce a 2‐machine flowshop with processing flexibility. Two processing modes are available for each task: namely, processing by the designated processor, and processing simultaneously by both processors. The objective studied is makespan minimization. This production environment is encountered in repetitive manufacturing shops equipped with processors that have the flexibility to execute orders either individually or in coordination. In the latter case, the product designer exploits processing synergies between two processors so as to execute a particular task much faster than a dedicated processor. This type of flowshop environment is also encountered in labor‐intensive assembly lines where products moving downstream can be processed either in the designated assembly stations or by pulling together the work teams of adjacent stations. This scheduling problem requires determining the mode of operation of each task, and the subsequent scheduling that preserves the flowshop constraints. We show that the problem is ordinary NP‐complete and obtain an optimal solution using a dynamic programming algorithm with considerable computational requirements for medium and large problems. Then, we present a number of dynamic programming relaxations and analyze their worst‐case error performance. Finally, we present a polynomial time heuristic with worst‐case error performance comparable to that of the dynamic programming relaxations. © 2003 Wiley Periodicals, Inc. Naval Research Logistics, 2004.     

一种基于MPSoC的网络处理器模型及其应用

从MPSoC系统设计角度出发提出了网络处理器的参数化分析模型,称为NePlat。该模型采用数据流进程网络(DPN,Dataflow Process Network)描述网络应用,构造参数化异构硬件资源,并将应用模型映射到体系结构资源上评价网络处理器性能。     

Scheduling with variable time slot costs

In this article, we study a class of new scheduling models where time slot costs have to be taken into consideration. In such models, processing a job will incur certain cost which is determined by the time slots occupied by the job in a schedule. The models apply when operational costs vary over time. The objective of the scheduling models is to minimize the total time slot costs plus a traditional scheduling performance measure. We consider the following performance measures: total completion time, maximum lateness/tardiness, total weighted number of tardy jobs, and total tardiness. We prove the intractability of the models under general parameters and provide polynomial‐time algorithms for special cases with non‐increasing time slot costs.© 2010 Wiley Periodicals, Inc. Naval Research Logistics, 2010     

基于计数器的处理器核性能分析与优化

随着处理器微体系结构日益复杂,性能分析在处理器研制过程中的作用越来越重要。常用的性能分析方法是建立性能模型,该方法主要用于研制初期的设计空间探索,如果用于微体系结构级的分析和优化,速度和精度都会成为限制因素。提出了一种基于计数器的性能分析方法,该方法以项目组已经完成的一款处理器核的硬件实现代码为基础,在处理器核外部添加一个专用性能监测单元,收集微体系结构分析和优化需要的各种事件,并通过结果分析器对统计的事件进行分析,得到微体系结构实现的性能受限因素。采用此方法,在FPGA原型系统上对SPEC CPU2000测试程序运行时的性能受限因素进行了分析,并根据分析结果采取了相应的优化措施,优化后的处理器核性能得到了明显提升。     

Capacitated selection problem

Optimizing the selection of resources to accomplish a set of tasks involves evaluating the tradeoffs between the cost of maintaining the resources necessary to accomplish the tasks and the penalty cost associated with unfinished tasks. We consider the case where resources are categorized into types, and limits (capacity) are imposed on the number of each type that can be selected. The objective is to minimize the sum of penalty costs and resource costs. This problem has several practical applications including production planning, new product design, menu selection and inventory management. We develop a branch‐and‐bound algorithm to find exact solutions to the problem. To generate bounds, we utilize a dual ascent procedure which exploits the special structure of the problem. Information from the dual and recovered primal solutions are used to select branching variables. We generate strong valid inequalities and use them to fix other variables at each branching step. Results of tests performed on reasonably sized problems are presented. © 1999 John Wiley & Sons, Inc. Naval Research Logistics 46: 19–37, 1999     

A note on the optimal scheduling of two parallel processors

The problem of minimizing mean flow time of two parallel processors is discussed. Prior results are briefly reviewed. A dynamic programming algorithm is presented which minimizes mean flow time for a set of n preordered jobs on two nonequivalent parallel processors. The algorithm is illustrated with an example problem. The computational experience is presented which illustrates the efficiency of the algorithm.     

多核环境下负载均衡的并行离散事件全局调度机制

分析了多核环境下传统的离散事件时间弯曲并行系统的性能,针对其事件调度开销小和负载均衡能力强难以兼得的问题,提出了一种基于分布式队列的全局调度机制,设计了相应的数据结构和调度算法,大大减少了锁开销.通过大量实验对多核环境下几种典型离散事件系统并行策略的性能分析表明,本文提出的全局调度策略不仅事件调度开销小,而且回滚率大大降低,有效克服了传统策略回滚量较大或难以实现动态负载平衡的情况,并具备良好的可扩展性.     

任务划分与处理机调度的递归-分治方法

本文介绍并实现了一种如何把一个顺序执行的任务集,根据其子任务之间潜在的并行性,划分成若干个可并发执行的任务子集,并把每个子集分配给一个处理机,使各处理机之间的数据通信量尽可能地少,同时兼顾各处理机之间负载平衡的算法。最后给出了几个典型例题的试算结果,为了满足用户的不同要求,文章还提出了几点改进方法。     

长向量处理器高效RNN推理方法

模型深度的不断增加和处理序列长度的不一致对循环神经网络在不同处理器上的性能优化提出巨大挑战。针对自主研制的长向量处理器FT-M7032,实现了一个高效的循环神经网络加速引擎。该引擎采用行优先矩阵向量乘算法和数据感知的多核并行方式,提高矩阵向量乘的计算效率;采用两级内核融合优化方法降低临时数据传输的开销;采用手写汇编优化多种算子,进一步挖掘长向量处理器的性能潜力。实验表明,长向量处理器循环神经网络推理引擎可获得较高性能,相较于多核ARM CPU以及Intel Golden CPU,类循环神经网络模型长短记忆网络可获得最高62.68倍和3.12倍的性能加速。