Recent Advances on Modeling, Control, and Optimization for Complex Engineering SystemsView this Special Issue
Optimal Rules for Single Machine Scheduling with Stochastic Breakdowns
This paper studies the problem of scheduling a set of jobs on a single machine subject to stochastic breakdowns, where jobs have to be restarted if preemptions occur because of breakdowns. The breakdown process of the machine is independent of the jobs processed on the machine. The processing times required to complete the jobs are constants if no breakdown occurs. The machine uptimes are independently and identically distributed (i.i.d.) and are subject to a uniform distribution. It is proved that the Longest Processing Time first (LPT) rule minimizes the expected makespan. For the large-scale problem, it is also showed that the Shortest Processing Time first (SPT) rule is optimal to minimize the expected total completion times of all jobs.
Machine scheduling problems belong to the classic combinational optimization problems. These problems deal with the model where decision maker needs to arrange jobs to process on a limited number of machines or processors. Machine scheduling problems play an important role in manufacturing, parallel computing, or compiler optimization. These problems have been studied since the 1950s and a lot of results have been achieved until now. We refer to the books by Brucker  and Pinedo  for a general overview of literature in scheduling problems.
In the environment of classical scheduling problems, the machine is assumed to be workable continuously until the completion of all jobs. Nevertheless, some unexpected events during production (e.g., equipment damaged, error operation, and instrument breakdown) often occur in manufacturing environment. Therefore, it is very common that a machine breakdown happens during the processing of a job. Moreover, the information about the breakdowns may be uncertain. In the realistic situation, the decision maker has to consider how to utilize the available information to give a more effective scheduling plan in order to increase the output and reduce the cost. In this way, it is necessary and valuable to research the stochastic scheduling problems with random machine breakdowns.
According to the impact a machine breakdown exerts to the job being processed, the machine breakdowns could be categorized into two models: preemptive-resume model and preemptive-repeat. In the preemptive-resume model, if a breakdown happens during the processing of a job, the work done prior to the breakdown is not lost, and the job could be resumed when the machine becomes available again. In the preemptive-repeat model, the job has to be reprocessed in its entirety if the machine breakdown occurs before the job is completed.
The main purpose of this paper is to study the problem with machine breakdowns of preemptive-repeat model. There are many results on preemptive-resume model. Such as Glazebrook , Birge et al. , Mittenthal and Raghavachari , Cai and Zhou [6, 7], and Qi et al. . Regarding the preemptive-repeat model, many authors have contributed remarkable achievements. Adiri et al. [9, 10] studied the problems with single breakdown; Cai et al. [11–13] studied the problems in which the realizations of processing times for a job between breakdowns are the same. They referred to this model as the case of without resampling. Frostig  considered the resampling mechanism in which the repeated processing times of a job are i.i.d. between breakdowns. Khalil and Dimitrov  studied the flow time of a job under the preemptive-repeat and preemptive-resume models. Lee and Lin  considered the problem where the decision maker can decide whether to activate a maintenance or to leave the machine to run at a slower speed. Kasap et al.  studied the uptime distributions to ensure that the LPT rule minimizes the expected makespan. Tang and Zhao  designed an optimal algorithm for the problem with early and tardy penalties. Lee and Yu  gave algorithms to the problem with the objective to minimize the expected weighted completion times and expected maximum tardiness.
However, all the papers reviewed above (except Kasap et al. ) carry the implicit assumption that the breakdown process of the machine is dependent on the job it is processing. With this assumption, the problem with machine breakdowns could be converted to the traditional scheduling problem without breakdowns; see papers [11, 12, 18].
In this paper, the machine breakdowns are subject to preemptive-repeat model and are independent of job it is processing. The objective is to minimize the expected makespan or expected total completion times of all jobs. For this problem, Adiri et al.  firstly studied a special case of a single machine scheduling with only one breakdown, and the machine is assumed to be continuous workable after the breakdown. Subject to this restriction, Adiri et al. concluded that the LPT (SPT) rule minimizes the expected makespan when the distribution function of the uptime is convex (concave). This paper considers the general problem where the downtimes (repairing time) are i.i.d., and the uptimes are independently subject to a common uniform distribution. Under the assumptions above, it is proved that LPT rule is optimal to achieve the minimal expected makespan, and SPT rule minimizes the expected total completion times for large-scale problem, where the number of jobs is large enough.
The remainder of the paper is organized as follows. In Section 2, the model with stochastic preemptive-repeat breakdowns is formulated. Then, for a given processing order, we present a formulation of the expected completion time of a job. In Section 3, we show that the LPT rule minimizes the expected makespan. Section 4 demonstrates that SPT rule minimizes the expected total completion times for large-scale problems. Finally, some concluding remarks are made in Section 5.
2. Formulation of Problem
Suppose there are jobs available at time 0 and these jobs are to be processed on a single machine. Denote by a constant the time needed to process job if no breakdown occurs during its processing. Due to the breakdowns, the actual time needed to process job may be more than , and the time may vary in different processing orders. It is assumed that the machine could process one and only one job at a time, and once a job begins to be processed on the machine, it could not be preempted by other jobs (except by machine breakdowns) until its completion.
The machine is subject to stochastic breakdowns, and, after each maintenance, the machine will start anew. The breakdown process is characterized by a sequence of positive random vectors , where are, respectively, the durations of the th uptime and downtime of the machine. The uptimes and downtimes are defined to be independent of the jobs. If the machine breaks down during the processing of job , the work done on the job will be lost and the job has to be restarted. are defined to be i.i.d. and follow the uniform distribution with the distribution function with the support in . Hence, . are also i.i.d. with an arbitrary distribution with . The objective functions in this paper are the expected makespan and the expected total completion times. Our work focuses on the scheduling order of all jobs so as to minimize the objective functions.
Define a jobs processing order , and . Given a processing order , assume the machine begins to process only the th, thth jobs at time zero; then define that denotes the time to complete the jobs. So the makespan , and the completion time of th job is . Let be an indicator variable such that if event A occurs; otherwise . Based on the notations defined above, the expected completion time of job could be expressed as which implies Therefore, we obtain the expected makespan
3. LPT Minimizes Expected Makespan
In this section, we will prove the optimality of LPT to minimize the expected value of makespan. Define the processing order , which is obtained by interchanging the processing order of the two jobs in . The following lemma shows that the processing of first two jobs is subject to LPT rule.
Lemma 1. Consider the processing order . If the uptimes are i.i.d and uniformly distributed and , then the expected makespan can be reduced by interchanging the first two jobs ; that is,
Proof. According to the definition of and by (3), it is obtained that By replacing (5) in (3), That is, By the same method, we could get the expression of . Since we obtain The conclusion follows by .
Corollary 2. Assume the uptimes are i.i.d and uniformly distributed. If , then
Proof. The proof is the same as that in Lemma 1.
Next we will give another expression for . Assume the machine begins to process jobs at time zero and . Let ; then we consider the following three cases.
Case 1 (). With the possibility , in this case, we have Case 2 (). With the possibility , we have Case 3 (). We have From the three cases above, we get Known from Corollary 2, we know . Hence That is, The following lemma shows that the processing of any two consecutive jobs is subject to LPT rule.
Lemma 3. Assume the uptimes are i.i.d and uniformly distributed. If , then
Proof. Let be the uptime where the job is completed, and assume the machine has been continuously processing jobs for time units when is finished at time ; that is, we have . Let , and let be the distribution function of . We have
which implies that is uniformly distributed for any given . According to the definition of , we know . We now consider four possibilities depending on the value obtains.
Case 1 (). We have Case 2 (). We have Case 3 (). We have Case 4 (). We have Hence, Because By replacing (24) in (23), we obtain By (16) and , the conclusion in this lemma holds.
Theorem 4. Assume the uptimes are i.i.d and uniformly distributed with support in , where ; then the LPT rule is optimal to minimize the expected makespan.
4. SPT Minimizes Expected Total Completion Times
This section considers the single machine problem to minimize the expected value of total completion times, that is, . Assume there exist two constants such that for ; that is, the processing times of all jobs are uniformly bounded. For a given processing order , if machine begins to process the th, thth jobs at time zero, define that denotes the sum of the completion times of the jobs. So we have In this section, we focus on the large-scale scheduling problems; that is, the number of jobs is large enough. With the assumption, the SPT rule will be proved to be optimal.
Lemma 5. Assume the uptimes are i.i.d and uniformly distributed. If , then for large number .
By Lemma 5, the corollary below follows immediately.
Corollary 6. Assume the uptimes are i.i.d and uniformly distributed. If , then as long as is large enough. Also, we have
Proof. The proof is the same as that in Lemma 5.
In order to prove the main conclusion in this section, we will give another expression for . Assume the machine begins to process jobs at time zero and . Let , and ( for ). We consider the following three cases.
Case 1 (). In this case, we have We obtain Case 2 (). In this case, we have Therefore, we get Case 3 (). We have And we obtain Based on the three cases above, we have for large number . The inequality holds by the conclusion in Corollary 6.
Lemma 7. Assume the uptimes are i.i.d and uniformly distributed. If (), then one has for large number .
Proof. We let be the uptime where the job is completed and let be the downtime after . Assume the machine has been continuously processing for time units when is finished at time ; that is, we have . Let , and let be the distribution function of . Here, we have
Case 1 (). In this case, we have
Therefore, we obtain
Case 2 (). The case is similar to the case . So we get
Case 3 (). The case is similar to the case . We have
Case 4 (). The case is similar to the case . We obtain
Based on the four cases above, we have
By (35), we have
We discuss the following cases..
Case 1 (). Because , we have That is, So we obtain Therefore, in this case we obtain That is, Case 2 (). In this case, we define , and . Next two cases are discussed for the value obtains.
Case 2.1 (). In this case, we have That is, Known from Case 1 above, we obtain Case 2.2 (). According to the definition of , we have , so Note that there exists a number such that , so we get for all . Therefore, for all . So we obtain Either or not, we always have for large number .
Based on Lemmas 5 and 7, the following theorem is immediate.
Theorem 8. Assume the uptimes are i.i.d. and uniformly distributed with support , where ; then the SPT rule is optimal to minimize the expected value of total completion times if the number of jobs is large enough.
5. Concluding Remarks
The stochastic scheduling problem on a single machine with random breakdowns has been investigated in this paper. We consider the situation where the uptimes are uniformly distributed and i.i.d; the downtimes are also assumed to be i.i.d and follow an arbitrary distribution. The machine breakdowns are defined to be independent of the job it is processing. Under the assumptions above, we prove that the LPT rule could achieve the minimal expected makespan; the SPT rule is optimal to minimize the expected value of total completion times for large scale problems. For the scheduling with stochastic breakdowns independent of job it is processing, the result obtained in this paper is the foundation in this area.
Some problems may be considered for the future research: (a) whether optimal rule exists when the uptimes are subject to other probability distributions; (b) problems with other objective functions are worth investigation; (c) the multimachine version will also be an interesting but difficult problem in the future.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
We are grateful to the anonymous referees for their valuable comments and suggestions. This research is supported by the Young Scientists Fund of the National Natural Science Foundation of China (61304209 and 11201282), the Fund of scheme for training young teachers in Colleges and universities in Shanghai (ZZCD12006), the Ministry of Education of Humanities and Social Science Fund Project (10YJCZH032), and Innovation Program of Shanghai Municipal Education Commission (14YZ127).
M. Pinedo, Scheduling: Theory, Algorithms, and Systems, Springer, Heidelberg, Germany, 3rd edition, 2002.View at: MathSciNet
X. Cai, X. Sun, and X. Zhou, “Stochastic scheduling with preemptive-repeat machine breakdowns to minimize the expected weighted flow time,” Probability in the Engineering and Informational Sciences, vol. 17, no. 4, pp. 467–485, 2003.View at: Publisher Site | Google Scholar | Zentralblatt MATH | MathSciNet