Modern manufacturing systems need new types of scheduling methods. While traditional scheduling methods are primarily concerned with sequencing of jobs, modern manufacturing environments provide the additional possibility to process jobs in batches. This adds to the complexity of scheduling. There are two types of batching namely: (i) serial batching (jobs may be batched if they share the same setup on a machine and one job is processed at a time. The machine which processes jobs in this manner is called as discrete processor) and (ii) parallel batching (several jobs can be processed simultaneously on a machine at a time. The machine which processes jobs in this manner is called as batch processor or batch processing machine). Parallel batching environments have attracted wide attention of the researchers working in the field of scheduling. Particularly, taking inspiration from studies of scheduling batch processors in semiconductor manufacturing [Mathirajan and Sivakumar (2006b) and Venkataramana (2006)] and in steel casting industries [Krishnaswamy et al. (1998), Shekar (1998) and Mathirajan (2002)] in the Management Studies Department of Indian Institute of Science, this thesis addresses a special problem on scheduling batch processor, observed in the steel casting manufacturing. A fundamental feature of the steel casting industry is its extreme flexibility, enabling castings to be produced with almost unlimited freedom in design over an extremely wide range of sizes, quantities and materials suited to practically every environment and application. Furthermore, the steel casting industry is capital intensive and highly competitive. From the viewpoint of throughput and utilization of the important and costly resources in the foundry manufacturing, it was felt that the process-controlled furnace operations for the melting and pouring operations as well as the heat-treatment furnace operations are critical for meeting the overall production schedules. The two furnace operations are batch processes that have distinctive constraints on job-mixes in addition to the usual capacity and technical constraints associated with any industrial processes. The benefits of effective scheduling of these batch processes include higher machine utilization, lower work-in-process (WIP) inventory, shorter cycle time and greater customer satisfaction [Pinedo (1995)]. Very few studies address the production planning and scheduling models for a steel foundry, considering the melting furnace of the pre-casting stage as the core foundry operation [Voorhis et al. (2001), Krishnaswamy et al. (1998) and Shekar (1998)]. Even though the melting and pouring operations may be considered as the core of foundry operations and their scheduling is of central importance, the scheduling of heat-treatment furnaces is also of considerable importance. This is because the processing time required at the heat treatment furnace is often longer compared to other operations in the steel-casting foundry and therefore considerably affects the scheduling, overall flow time and WIP inventory. Further, the heat-treatment operation is critical because it determines the final properties that enable components to perform under demanding service conditions such as large mechanical load, high temperature and anti-corrosive processing. It is also important to note that the heat-treatment operation is the only predominantly long process in the entire steel casting manufacturing process, taking up a large part of total processing time (taking up to a few days as against other processes that typically take only a few hours). Because of these, the heat-treatment operation is a major bottleneck operation in the entire steel casting process. The jobs in the WIP inventory in front of heat-treatment furnace vary widely in sizes (few grams to a ton) and dimensions (from 10 mm to 2000 mm). Furthermore, castings are primarily classified into a number of job families based on the alloy type, such as low alloy castings and high alloy castings. These job families are incompatible as the temperature requirement for low alloy and high alloy vary for similar type of heat-treatment operation required. These job families are further classified into various sub-families based on the type of heat treatment operations they undergo. These sub-families are also incompatible as each of these sub-families requires a different combination of heat-treatment operation. The widely varying job sizes, job dimensions and multiple incompatible job family characteristic introduce a high degree of complexity into scheduling heat-treatment furnace. Scheduling of heat-treatment furnace with multiple incompatible job families can have profound effect on the overall production rate as the processing time at heat-treatment operation is very much longer. Considering the complexity of the process and time consumed by the heat treatment operation, it is imperative that efficient scheduling of this operation is required in order to maximize throughput and to enhance productivity of the entire steel casting manufacturing process. This is of importance to the firm. The concerns of the management in increasing the throughput of the bottleneck machine, thereby increasing productivity, motivated us to adopt the scheduling objective of makespan. In a recent observation of heat-treatment operations in a couple of steel casting industries and the research studies reported in the literature, we noticed that the real-life problem of dynamic scheduling of heat-treatment furnace with multiple incompatible job families, non-identical job sizes, non-identical job dimensions, non-agreeable release times and due dates to maximize the throughput, higher utilization and minimize the work-in-process inventory is not at all addressed. However, there are very few studies [Mathirajan et al. (2001, 2002, 2004a, 2007)] which have addressed the problem of scheduling of heat-treatment furnace with incompatible job families and non-identical job sizes to maximize the utilization of the furnace. Due to the difference between the real-life situation on dynamic scheduling of heat-treatment furnace of the steel casting manufacturing and the research reported on the same problem, we identified three new class of batch processor problems, which are applicable to a real-life situation based on the type of heat-treatment operation(s) being carried out and the type of steel casting industry (small, medium and large scale steel casting industry) and this thesis addresses these new class of research problems on scheduling of batch processor. The first part of the thesis addresses our new Research Problem (called Research Problem 1) of minimizing makespan (Cmax) on a batch processor (BP) with single job family (SJF), non-identical job sizes (NIJS), and non-identical job dimensions (NIJD). This problem is of interest to small scale steel casting industries performing only one type of heat treatment operation such as surface hardening. Generally, there would be only a few steel casting industries which offer such type of special heat-treatment operation and thus the customer is willing to accept delay in the completion of his orders. So, the due date issues are not important for these types of industries. We formulate the problem as Mixed Integer Linear Programming (MILP) model and validate the proposed MILP model through a numerical example. In order to understand the computational intractability issue, we carry out a small computational experiment. The results of this experiment indicate that the computational time required, as a function of problem size, for solving the MILP model is non-deterministic and non-polynomial. Due to the computational intractability of the proposed MILP model, we propose five variants of a greedy heuristic algorithm and a genetic algorithm for addressing the Research Problem 1. We carry out computational experiments to obtain the performance of heuristic algorithms based on two perspectives: (i) comparison with optimal solution on small scale instances and (ii) comparison with lower bound for large scale instances. We choose five important problem parameters for the computational experiment and propose a suitable experimental design to generate pseudo problem instances. As there is no lower bound (LB) procedure for the Research Problem1, in this thesis, we develop an LB procedure that provides LB on makespan by considering both NIJS and NIJD characteristics together. Before using the proposed LB procedure for evaluating heuristic algorithms, we conduct a computational experiment to obtain the quality of the LB on makespan in comparison with optimal makespan on number of small scale instances. The results of this experiment indicate that the proposed LB procedure is efficient and could be used to obtain LB on makespan for any large scale problem. In the first perspective of the evaluation of the performance of the heuristic algorithms proposed for Research Problem 1, the proposed heuristic algorithms are run through small scale problem instances and we record the makespan values. We solve the MILP model to obtain optimal solutions for these small scale instances. For comparing the proposed heuristic algorithms we use the performance measures: (a) number of times the proposed heuristic algorithm solution equal to optimal solution and (b) average loss with respect to optimal solution in percentage. In the second perspective of the evaluation of the performance of the heuristic algorithms, the proposed heuristic algorithms are run through large scale problem instances and we record the makespan values. The LB procedure is also run through these problem instances to obtain LB on makespan. For comparing the performance of heuristic algorithms with respect to LB on makespan, we use the performance measures: (a) number of times the proposed heuristic algorithm solution equal to LB on makespan (b) average loss with respect to LB on makespan in percentage, (c) average relative percentage deviation and (d) maximum relative percentage deviation. We extend the Research Problem 1 by including additional job characteristics: job arrival time to WIP inventory area of heat-treatment furnace, due date and additional constraint on non-agreeable release time and due date (NARD). Due date considerations and the constraint on non-agreeable release times and due date (called Research Problem 2) are imperative to small scale steel casting foundries performing traditional but only one type of heat treatment operation such as annealing where due date compliance is important as many steel casting industries offer such type of heat treatment operations. The mathematical model, LB procedure, greedy heuristic algorithm and genetic algorithm proposed for Research Problem 1, including the computational experiments, are appropriately modified and\or extended for addressing Research Problem 2. Finally, we extend the Research Problem 2 is by including an additional real life dimension: multiple incompatible job families (MIJF). This new Research Problem (called Research Problem 3) is more relevant to medium and large scale steel casting foundries performing more than one type of heat treatment operations such as homogenizing and tempering, normalizing and tempering. The solution methodologies, the LB procedure and the computational experiments proposed for Research Problem 2 are further modified and enriched to address the Research Problem 3. From the detailed computational experiments conducted for each of the research problems defined in this study, we observe that: (a) the problem parameters considered in this study have influence on the performance of the heuristic algorithms, (b) the proposed LB procedure is found to be efficient, (c) the proposed genetic algorithm outperforms among the proposed heuristic algorithms (but the computational time required for genetic algorithm increases as problem size keeps increasing), and (d) in case the decision maker wants to choose an heuristic algorithm which is computationally most efficient algorithm among the proposed algorithms, the variants of greedy heuristic algorithms : SWB, SWB(NARD), SWB(NARD&MIJF) is relatively the best algorithm for Research Problem 1, Research Problem 2 and Research Problem 3 respectively.

Distributed systems are everywhere. From large datacenters to mobile devices, an ever richer assortment of applications and services relies on distributed systems, infrastructure, and protocols. Despite their ubiquity, testing and debugging distributed systems remains notoriously hard. Moreover, aside from inherent design challenges posed by partial failure, concurrency, or asynchrony, there remain significant challenges in the implementation of distributed systems. These programming challenges stem from the increasing complexity of the concurrent activities and reactive behaviors in a distributed system on the one hand, and the need to effectively leverage the parallelism offered by modern multi-core hardware, on the other hand. This thesis contributes Kompics, a programming model designed to alleviate some of these challenges. Kompics is a component model and programming framework for building distributed systems by composing message-passing concurrent components. Systems built with Kompics leverage multi-core machines out of the box, and they can be dynamically reconfigured to support hot software upgrades. A simulation framework enables deterministic execution replay for debugging, testing, and reproducible behavior evaluation for largescale Kompics distributed systems. The same system code is used for both simulation and production deployment, greatly simplifying the system development, testing, and debugging cycle. We highlight the architectural patterns and abstractions facilitated by Kompics through a case study of a non-trivial distributed key-value storage system. CATS is a scalable, fault-tolerant, elastic, and self-managing key-value store which trades off service availability for guarantees of atomic data consistency and tolerance to network partitions. We present the composition architecture for the numerous protocols employed by the CATS system, as well as our methodology for testing the correctness of key CATS algorithms using the Kompics simulation framework. Results from a comprehensive performance evaluation attest that CATS achieves its claimed properties and delivers a level of performance competitive with similar systems which provide only weaker consistency guarantees. More importantly, this testifies that Kompics admits efficient system implementations. Its use as a teaching framework as well as its use for rapid prototyping, development, and evaluation of a myriad of scalable distributed systems, both within and outside our research group, confirm the practicality of Kompics. / <p>QC 20130520</p>

L'objectif de cette thèse est l'étude d'un environnement d'exécution pour machines parallèles sans mémoire commune. Elle comprend la définition d'un modèle de programme parallèle, basé sur l'échange de message offrant une forme restreinte de mémoire partagée. La communication est indirecte, via des portes; les processus utilisent les barrières pour la synchronisation. Les entités du système. processus, portes et barrières, sont créées dynamiquement, et placées sur un processeur quelconque du réseau de processeurs de façon explicite. Nous proposons une implantation de ce modèle comme la mise en oeuvre systématique d'une architecture client/serveur. Cette implantation a été efféctuée sur une machine Supemode. La base est un Micro Noyau Parallèle, où le composant principal est un mécanisme d'appel de procédure à distance minimal. / This thesis describes an execution environment for parallel machines without shared memory. A parallel programming model based on message passing, with a special shared memory. In this model, process communication occurs indirectly, via ports, and the processes use barriers for synchronization. All the entities of the system, such as processes, ports and barriers, are created dynamically and loaded on any processor of the network of processors. The implementation architecture of our model is a systematic realization of the client/server model. An implementation is proposed in a Supernode parallel machine as a parallel micro kernel. The principal parallel micro kernel component is a minimal remote procedure call mechanism.

L'objectif de cette thèse est l'étude d'un environnement d'exécution pour machines parallèles sans mémoire commune. Elle comprend la définition d'un modèle de programme parallèle, basé sur l'échange de message offrant une forme restreinte de mémoire partagée. La communication est indirecte, via des portes; les processus utilisent les barrières pour la synchronisation. Les entités du système. processus, portes et barrières, sont créées dynamiquement, et placées sur un processeur quelconque du réseau de processeurs de façon explicite. Nous proposons une implantation de ce modèle comme la mise en oeuvre systématique d'une architecture client/serveur. Cette implantation a été efféctuée sur une machine Supemode. La base est un Micro Noyau Parallèle, où le composant principal est un mécanisme d'appel de procédure à distance minimal. / This thesis describes an execution environment for parallel machines without shared memory. A parallel programming model based on message passing, with a special shared memory. In this model, process communication occurs indirectly, via ports, and the processes use barriers for synchronization. All the entities of the system, such as processes, ports and barriers, are created dynamically and loaded on any processor of the network of processors. The implementation architecture of our model is a systematic realization of the client/server model. An implementation is proposed in a Supernode parallel machine as a parallel micro kernel. The principal parallel micro kernel component is a minimal remote procedure call mechanism.

L'objectif de cette thèse est l'étude d'un environnement d'exécution pour machines parallèles sans mémoire commune. Elle comprend la définition d'un modèle de programme parallèle, basé sur l'échange de message offrant une forme restreinte de mémoire partagée. La communication est indirecte, via des portes; les processus utilisent les barrières pour la synchronisation. Les entités du système. processus, portes et barrières, sont créées dynamiquement, et placées sur un processeur quelconque du réseau de processeurs de façon explicite. Nous proposons une implantation de ce modèle comme la mise en oeuvre systématique d'une architecture client/serveur. Cette implantation a été efféctuée sur une machine Supemode. La base est un Micro Noyau Parallèle, où le composant principal est un mécanisme d'appel de procédure à distance minimal. / This thesis describes an execution environment for parallel machines without shared memory. A parallel programming model based on message passing, with a special shared memory. In this model, process communication occurs indirectly, via ports, and the processes use barriers for synchronization. All the entities of the system, such as processes, ports and barriers, are created dynamically and loaded on any processor of the network of processors. The implementation architecture of our model is a systematic realization of the client/server model. An implementation is proposed in a Supernode parallel machine as a parallel micro kernel. The principal parallel micro kernel component is a minimal remote procedure call mechanism.

Les machines pétascale qui existent aujourd'hui ont un temps moyen entre pannes de plusieurs heures. Il est prévu que dans les futurs systèmes ce temps diminuera. Pour cette raison, les applications qui fonctionneront sur ces systèmes doivent être capables de tolérer des défaillances fréquentes. Aujourd'hui, le moyen le plus commun de le faire est d'utiliser le mécanisme de retour arrière global où l'application fait des sauvegardes périodiques à partir d’un point de reprise. Si un processus s'arrête à cause d'une défaillance, tous les processus reviennent en arrière et se relancent à partir du dernier point de reprise. Cependant, cette solution deviendra infaisable à grande échelle en raison des coûts de l'énergie et de l'utilisation inefficace des ressources. Dans le contexte des applications MPI, les protocoles de journalisation des messages offrent un meilleur confinement des défaillances car ils ne demandent que le redémarrage du processus qui a échoué, ou parfois d’un groupe de processus limité. Par contre, les protocoles existants ont souvent un surcoût important en l’absence de défaillances qui empêchent leur utilisation à grande échelle. Ce surcoût provient de la nécessité de sauvegarder de façon fiable tous les événements non-déterministes afin de pouvoir correctement restaurer l'état du processus en cas de défaillance. Ensuite, comme les journaux de messages sont généralement stockés dans la mémoire volatile, la journalisation risque de nécessiter une large utilisation de la mémoire. Une autre tendance importante dans le domaine des HPC est le passage des applications MPI simples aux nouveaux modèles de programmation hybrides tels que MPI + threads ou MPI + tâches en réponse au nombre croissant de cœurs par noeud. Cela offre l’opportunité de gérer les défaillances au niveau du thread / de la tâche contrairement à l'approche conventionnelle qui traite les défaillances au niveau du processus. Par conséquent, le travail de cette thèse se compose de trois parties. Tout d'abord, nous présentons un protocole de journalisation hiérarchique pour atténuer une défaillance de processus. Le protocole s'appelle Scalable Pattern-Based Checkpointing et il exploite un nouveau modèle déterministe appelé channel-determinism ainsi qu’une nouvelle relation always-happens-before utilisée pour mettre partiellement en ordre les événements de l'application. Le protocole est évolutif, son surcoût pendant l'exécution sans défaillance est limité, il n'exige l'enregistrement d'aucun évènement et, enfin, il a une reprise entièrement distribuée. Deuxièmement, afin de résoudre le problème de la limitation de la mémoire sur les nœuds de calcul, nous proposons d'utiliser des ressources dédiées supplémentaires, appelées logger nodes. Tous les messages qui ne rentrent pas dans la mémoire du nœud de calcul sont envoyés aux logger nodes et sauvegardés dans leur mémoire. À travers de nos expériences nous montrons que cette approche est réalisable et, associée avec un protocole de journalisation hiérarchique comme le SPBC, les logger nodes peuvent être une solution ultime au problème de mémoire limitée sur les nœuds de calcul. Troisièmement, nous présentons un protocole de tolérance aux défaillances pour des applications hybrides qui adoptent le modèle de programmation MPI + tâches. Ce protocole s'utilise pour tolérer des erreurs détectées non corrigées qui se produisent lors de l'exécution d'une tâche. Normalement, une telle erreur provoque une exception du système ce qui provoque un arrêt brutal de l'application. Dans ce cas, l'application doit redémarrer à partir du dernier point de reprise. Nous combinons la sauvegarde des données de la tâche avec une journalisation des messages afin d’aider à la reprise de la tâche qui a subi une défaillance. Ainsi, nous évitons le redémarrage au niveau du processus, plus coûteux. Nous démontrons les avantages de ce protocole avec l'exemple des applications hybrides MPI + OmpSs. / Existing petascale machines have a Mean Time Between Failures (MTBF) in the order of several hours. It is predicted that in the future systems the MTBF will decrease. Therefore, applications that will run on these systems need to be able to tolerate frequent failures. Currently, the most common way to do this is to use global application checkpoint/restart scheme: if some process fails the whole application rolls back the its last checkpointed state and re-executes from that point. This solution will become infeasible at large scale, due to its energy costs and inefficient resource usage. Therefore fine-grained failure containment is a strongly required feature for the fault tolerance techniques that target large-scale executions. In the context of message passing MPI applications, message logging fault tolerance protocols provide good failure containment as they require restart of only one process or, in some cases, a bounded number of processes. However, existing logging protocols experience a number of issues which prevent their usage at large scale. In particular, they tend to have high failure-free overhead because they usually need to store reliably any nondeterministic events happening during the execution of a process in order to correctly restore its state in recovery. Next, as message logs are usually stored in the volatile memory, logging may incur large memory footprint, especially in communication-intensive applications. This is particularly important because the future exascale systems expect to have less memory available per core. Another important trend in HPC is switching from MPI-only applications to hybrid programming models like MPI+threads and MPI+tasks in response to the increasing number of cores per node. This gives opportunities for employing fault tolerance solutions that handle faults on the level of threads/tasks. Such approach has even better failure containment compared to message logging protocols which handle failures on the level of processes. Thus, the work in these dissertation consists of three parts. First, we present a hierarchical log-based fault tolerance solution, called Scalable Pattern-Based Checkpointing (SPBC) for mitigating process fail-stop failures. The protocol leverages a new deterministic model called channel-determinism and a new always-happens-before relation for partial ordering of events in the application. The protocol is scalable, has low overhead in failure-free execution and does not require logging any events, provides perfect failure containment and has a fully distributed recovery. Second, to address the memory limitation problem on compute nodes, we propose to use additional dedicated resources, or logger nodes. All the logs that do not fit in the memory of compute nodes are sent to the logger nodes and kept in their memory. In a series of experiments we show that not only this approach is feasible, but, combined with a hierarchical logging scheme like the SPBC, logger nodes can be an ultimate solution to the problem of memory limitation for logging protocols. Third, we present a log-based fault tolerance protocol for hybrid applications adopting MPI+tasks programming model. The protocol is used to tolerate detected uncorrected errors (DUEs) that happen during execution of a task. Normally, a DUE caused the system to raise an exception which lead to an application crash. Then, the application has to restart from a checkpoint. In the proposed solution, we combine task checkpointing with message logging in order to support task re-execution. Such task-level failure containment can be beneficial in large-scale executions because it avoids the more expensive process-level restart. We demonstrate the advantages of this protocol on the example of hybrid MPI+OmpSs applications.