Dissertations / Theses on the topic 'Workload scheduling'

Thesis (M. Eng.)--Massachusetts Institute of Technology, Dept. of Mechanical Engineering, 2006.
Includes bibliographical references.
This thesis is the result of a six-month internship at the Steel Stock Department of Keppel FELS Singapore, a company which is involved in the design and construction of oil-rigs. The primary objective of this project is to reduce the tardiness of the delivery of steel materials and identify the reasons behind the delay. The initial stage of this attachment is devoted to understanding the process flow of the department. Analysis has been done to pinpoint to the exact causes of the delay, which is at the stages of document processing and dispatching to the storage areas. The workload at each stage of document processing has been analyzed using a queuing model and it has been found that the stage that the issue vouchers have to be generated and printed out is the bottleneck in the entire process flow. Some recommendations have been proposed to alleviate the problem. The second part of this thesis focuses on the reasons why scheduling rules should be utilized when dispatching the issue vouchers to the storage areas. Three scheduling rules have been tested and their performances with regards to tardiness have been studied.
by Chin Jiat Tan.
M.Eng.

The big data era has driven the needs for data analysis in every aspect of our daily lives. With the rapid growth of data size and complexity of data analysis models, modern big data analytic applications face the challenge to provide timely results often with limited resources. Such demand drives the growth of new hardware resources including GPUs and FPGAs, as well as storage devices such as SSDs and NVMs. It is challenging to manage the resources available in a cost restricted environment to best serve the applications with different characteristics. Extant approaches are agnostic to such heterogeneity in both underlying resources and workloads and require user knowledge and manual configuration for best performance. In this dissertation, we design, and implement a series of novel techniques, algorithms, and frameworks, to realize workload-aware resource management and scheduling. We demonstrate our techniques for efficient resource management across memory resource for in-memory data analytic platforms, processing resources for compute-intensive machine learning applications, and finally we design and develop a workload and heterogeneity-aware scheduler for general big data platforms. The dissertation demonstrates that designing an effective resource manager requires efforts from both application and system side. The presented approach makes and joins the efforts on both sides to provide a holistic heterogeneity-aware resource manage and scheduling system. We are able to avoid task failure due to resource unavailability by workload-aware resource management, and improve the performance of data processing frameworks by carefully scheduling tasks according to the task characteristics and utilization and availability of the resources.
Ph. D.
Clusters of multiple computers connected through internet are often deployed in industry for larger scale data processing or computation that cannot be handled by standalone computers. In such a cluster, resources such as CPU, memory, disks are integrated to work together. It is important to manage a pool of such resources in a cluster to efficiently work together to provide better performance for workloads running on top. This role is taken by a software component in the middle layer called resource manager. Resource manager coordinates the resources in the computers and schedule tasks to them for computation. This dissertation reveals that current resource managers often partition resources statically hence cannot capture the dynamic resource needs of workloads as well as the heterogeneous configurations of the underlying resources. For example, some computers in a clsuter might be older than the others with slower CPU, less memory, etc. Workloads can show different resource needs. Watching YouTube require a lot of network resource while playing games demands powerful GPUs. To this end, the disseration proposes novel approaches to manage resources that are able to capture the heterogeneity of resources and dynamic workload needs, based on which, it can achieve efficient resource management, and schedule the right task to the right resource.

A scheduler, as a central components of a computing site, aggregates computing resources and is responsible to distribute the incoming load (jobs) between the resources. Under such an environment, the optimum performance of the system against the service level agreement (SLA) based workloads, can be achieved by calculating the priority of SLA bound jobs using integrated heuristic. The SLA defines the service obligations and expectations to use the computational resources. The integrated heuristic is the combination of different SLA terms. It combines the SLA terms with a specific weight for each term. Theweights are computed by applying parameter sweep technique in order to obtain the best schedule for the optimum performance of the system under the workload. The sweepingof parameters on the integrated heuristic observed to be computationally expensive. The integrated heuristic becomes more expensive if no value of the computed weights result in improvement in performance with the resulting schedule. Hence, instead of obtaining optimum performance it incurs computation cost in such situations. Therefore, there is a need of detection of situations where the integrated heuristic can be exploited beneficially. For that reason, in this thesis we propose a metric based on the concept of utilization, to evaluate the SLA based parallel workloads of independent jobs to detect any impact of integrated heuristic on the workload.

Scheduling algorithm is the algorithm to allocate system resources among processes and data flows. Joint channel-assignment and workload-based (CAWS) is a recently developed algorithm for scheduling in the downlink of multi-channel wireless systems, such as OFDM. Compared to well known algorithms, CAWS algorithm has been proved to throughput optimal with flow-level dynamics. In this master thesis project, we design a system that accounts for power control and for the characteristics of common radio channels. We evaluate the efficiency of the algorithm under a diverse set of conditions. We also do analysis of CAWS algorithm under different traffic density.

The Grid can be seen as a collection of services each of which performs some functionality. Users of the Grid seek to use combinations of these services to perform the overall task they need to achieve. In general this can be seen as a set of services with a workflow document describing how these services should be combined. The user may also have certain constraints on the workflow operations, such as execution time or cost to the user, specified in the form of a Quality of Service (QoS) document. The users submit their workflow to a brokering service along with the QoS document. The brokering service's task is to map any given workflow to a subset of the Grid services taking the QoS and state of the Grid into account -- service availability and performance. We propose an approach for generating constraint equations describing the workflow, the QoS requirements and the state of the Grid. This set of equations may be solved using Mixed-Integer Linear Programming (MILP), which is the traditional method. We further develop a novel 2-stage stochastic MILP which is capable of dealing with the volatile nature of the Grid and adapting the selection of the services during the lifetime of the workflow. We present experimental results comparing our approaches, showing that the 2-stage stochastic programming approach performs consistently better than other traditional approaches. Next we addresses workload allocation techniques for Grid workflows in a multi-cluster Grid. We model individual clusters as MIMIk queues and obtain a numerical solution for missed deadlines (failures) of tasks of Grid workflows. We also present an efficient algorithm for obtaining workload allocations of clusters. Next we model individual cluster resources as G/G/l queues and solve an optimisation problem that minimises QoS requirement violation, provides QoS guarantee and outperforms reservation based scheduling algorithms. Both approaches are evaluated through an experimental simulation and the results confirm that the proposed workload allocation strategies combined with traditional scheduling algorithms performs considerably better in terms of satisfying QoS requirements of Grid workflows than scheduling algorithms that don't employ such workload allocation techniques. Next we develop a novel method for Grid brokers that aims at maximising profit whilst satisfying end-user needs with a sufficient guarantee in a volatile utility Grid. We develop a develop a 2-stage stochastic MILP which is capable of dealing with the volatile nature of the Grid and obtaining cost bounds that ensure that end-user cost is minimised or satisfied and broker's profit is maximised with sufficient guarantee. These bounds help brokers know beforehand whether the budget limits of end-users can be satisfied and, if not, then obtain appropriate future leases from service providers. Experimental results confirm the efficacy of our approach.

Multi-criteria scheduling provides the opportunity to create mathematical optimization models that are applicable to a diverse set of problem domains in the business world. This research addresses two different employee scheduling applications using multi-criteria objectives that present decision makers with trade-offs between global optimality and the level of disruption to current operating resources. Additionally, it investigates a scheduling problem from the product testing domain and proposes a heuristic solution technique for the problem that is shown to produce very high-quality solutions in short amounts of time. Chapter 2 addresses a grant administration workload-to-staff assignment problem that occurs in the Office of Research and Sponsored Programs at land-grant universities. We identify the optimal workload assignment plan which differs considerably due to multiple reassignments from the current state. To achieve the optimal workload reassignment plan we demonstrate a technique to identify the n best reassignments from the current state that provides the greatest progress toward the utopian solution. Solving this problem over several values of n and plotting the results allows the decision maker to visualize the reassignments and the progress achieved toward the utopian balanced workload solution. Chapter 3 identifies a weekly schedule that seeks the most cost-effective set of coach-to-program assignments in a gymnastics facility. We identify the optimal assignment plan using an integer linear programming model. The optimal assignment plan differs greatly from the status quo; therefore, we utilize a similar approach from Chapter 2 and use a multiple objective optimization technique to identify the n best staff reassignments. Again, the decision maker can visualize the trade-off between the number of reassignments and the resulting progress toward the utopian staffing cost solution and make an informed decision about the best number of reassignments. Chapter 4 focuses on product test scheduling in the presence of in-process and at-completion inspection constraints. Such testing arises in the context of the manufacture of products that must perform reliably in extreme environmental conditions. Each product receives a certification at the successful completion of a predetermined series of tests. Operational efficiency is enhanced by determining the optimal order and start times of tests so as to minimize the make span while ensuring that technicians are available when needed to complete in-process and at-completion inspections We first formulate a mixed-integer programming model (MILP) to identify the optimal solution to this problem using IBM ILOG CPLEX Interactive Optimizer 12.7. We also present a genetic algorithm (GA) solution that is implemented and solved in Microsoft Excel. Computational results are presented demonstrating the relative merits of the MILP and GA solution approaches across a number of scenarios.
Ph. D.

The requirements for real-time systems in safety-critical applications typically contain strict timing constraints. The design of such a system must be subject to extensive validation to guarantee that critical timing constraints will never be violated while the system operates. A mathematically rigorous technique to do so is to perform a schedulability analysis for formally verifying models of the computational workload. Different workload models allow to describe task activations at different levels of expressiveness, ranging from traditional periodic models to sophisticated graph-based ones. An inherent conflict arises between the expressiveness and analysis efficiency of task models. The more expressive a task model is, the more accurately it can describe a system design, reducing over-approximations and thus minimizing wasteful over-provisioning of system resources. However, more expressiveness implies higher computational complexity of corresponding analysis methods. Consequently, an ideal model provides the highest possible expressiveness for which efficient exact analysis methods exist. This thesis investigates the trade-off between expressiveness and analysis efficiency. A new digraph-based task model is introduced, which generalizes all previously proposed models that can be analyzed in pseudo-polynomial time without using any analysis-specific over-approximations. We develop methods allowing to efficiently analyze variants of the model despite their strictly increased expressiveness. A key contribution is the notion of path abstraction which enables efficient graph traversal algorithms. We demonstrate tractability borderlines for different classes of schedulers, namely static priority and earliest-deadline first schedulers, by establishing hardness results. These hardness proofs provide insights about the inherent complexity of developing efficient analysis methods and indicate fundamental difficulties of the considered schedulability problems. Finally, we develop a novel abstraction refinement scheme to cope with combinatorial explosion and apply it to schedulability and response-time analysis problems. All methods presented in this thesis are extensively evaluated, demonstrating practical applicability.

A column generation based approach is proposed to solve scheduling problems for fractional airlines efficiently and return near optimal schedules. Crew tours are building blocks of our approach, and our approach is focused on exploring more feasible tours than other approaches. In particular, all elements of a crew tour are optimized during the preparation and tour generation procedures. Moreover, time windows of customer-requested flights are handled exactly, and generalized to time window and crew time window of duties and tours. Furthermore, time windows of tours are contained in the MIP formulation to ensure more feasible connections between tours. In the pricing subproblem, an efficient constrained shortest path algorithm is proposed, which is necessary for our model and also provides extensibility for incorporating more complex constraints in the future. Computational results of our model show very small optimality gaps and consistent improvements over the model used in practice. Moreover, restricted versions of our model that have fast running time are provided, thus very desired in the case that running time has more priority than solution quality. In order to understand the demand, data mining of demand data is presented and analyzed. Moreover, a recovery model is proposed to deal with unscheduled maintenance in practice, by reserving airplanes and crews in the model. Computational experiments show the advantage of the recovery model, in the case of simulated unscheduled maintenance and comparing to models without recovery considerations.

Time series forecasting and modeling are challenging problems during the past decades, because of its plenty of properties and underlying correlated relationships. As a result, researchers proposed a lot of models to deal with the time series. However, the proposed models such as Autoregressive integrated moving average (ARIMA) and artificial neural networks (ANNs) only describe part of the properties of time series. In this thesis, we introduce a new hybrid model integrated filter structure to improve the prediction accuracy. Case studies with real data from University of Kentucky HealthCare are carried out to examine the superiority of our model. Also, we applied our model to operating room (OR) to reduce the inefficiency cost. The experiment results indicate that our model always outperforms compared with other models in different conditions.

Emerging architecture designs include tens of processing cores on a single chip die; it is believed that the number of cores will reach the hundreds in not so many years from now. However, most common workloads cannot expose fluctuating parallelism, insufficient to utilize such systems. The combination of these issues suggests that large-scale systems will be either multiprogrammed or have their unneeded resources powered off. To achieve these features, workloads must be able to provide a metric on their parallelism which the system can use to dynamically adapt per-application resource allotments.Adaptive resource management requires scheduling abstractions to be split into two cooperating layers. The system layer that is aware of the availability of resources and the application layer which can accurately and iteratively estimate the workload's true resource requirements.This thesis addresses these issues and provides a self-adapting work-stealing scheduling method that can achieve expected performance while conserving resources. This method is based on deterministic victim selection (DVS) that controls the concentration of the load among the worker threads. It allows to use the number of spawned but not yet processed tasks as a metric for the requirements. Because this metric measures work to be executed in the future instead of past behavior, DVS is versatile to handlevery irregular workloads.

QC 20130910

Emerging architecture designs include tens of processing cores on a single chip die; it is believed that the number of cores will reach the hundreds in not so many years from now. However, most common parallel workloads cannot fully utilize such systems. They expose fluctuating parallelism, and do not scale up indefinitely as there is usually a point after which synchronization costs outweigh the gains of parallelism. The combination of these issues suggests that large-scale systems will be either multiprogrammed or have their unneeded resources powered off.Multiprogramming leads to hardware resource contention and as a result application performance degradation, even when there are enough resources, due to negative share effects and increased bus traffic. Most often this degradation is quite unbalanced between co-runners, as some applications dominate the hardware over others. Current Operating Systems blindly provide applications with access to as many resources they ask for. This leads to over-committing the system with too many threads, memory contention and increased bus traffic. Due to the inability of the application to have any insight on system-wide resource demands, most parallel workloads will create as many threads as there are available cores. If every co-running application does the same, the system ends up with threads $N$ times the amount of cores. Threads then need to time-share cores, so the continuous context-switching and cache line evictions generate considerable overhead.This thesis proposes a novel solution across all software layers that achieves throughput optimization and uniform performance degradation of co-running applications. Through a novel fully automated approach (DVS and Palirria), task-parallel applications can accurately quantify their available parallelism online, generating a meaningful metric as parallelism feedback to the Operating System. A second component in the Operating System scheduler (Pond) uses such feedback from all co-runners to effectively partition available resources.The proposed two-level scheduling scheme ultimately achieves having each co-runner degrade its performance by the same factor, relative to how it would execute with unrestricted isolated access to the same hardware. We call this fair scheduling, departing from the traditional notion of equal opportunity which causes uneven degradation, with some experiments showing at least one application degrading its performance 10 times less than its co-runners.

QC 20151016

We develop a detailed, large-scale, airspace planning and collaborative decision-making model (APCDM), that is part of an $11.5B, 10-year, Federal Aviation Administration (FAA)-sponsored effort to increase U.S. National Airspace (NAS) capacity by 30 percent. Given a set of flights that must be scheduled during some planning horizon, we use a mixed-integer programming formulation to select a set of flight plans from among alternatives subject to flight safety, air traffic control workload, and airline equity constraints. Novel contributions of this research include three-dimensional probabilistic conflict analyses, the derivation of valid inequalities to tighten the conflict safety representation constraints, the development of workload metrics based on average (and its variance from) peak load measures, and the consideration of equity among airline carriers in absorbing the costs related to re-routing, delays, and cancellations. We also propose an improved set of flight plan cost factors for representing system costs and investigating fairness issues by addressing flight dependencies occurring in hubbed operations, as well as market factors such as schedule convenience, reliability, and the timeliness of connections. The APCDM model has potential use for both tactical and strategic applications, such as air traffic control in response to severe weather phenomenon or spacecraft launches, FAA policy evaluation, Homeland Defense contingency planning, and military air campaign planning. The model is tested to consider various airspace restriction scenarios imposed by dynamic severe weather systems and space launch Special Use Airspace (SUA) impositions. The results from this model can also serve to augment the FAAâ s National Playbook of standardized flight profiles in different disruption-prone regions of the National Airspace.
Ph. D.

Los cl'usters de computadores representan una soluci'on alternativa a los supercomputadores. En este tipo de sistemas, se suele restringir el espacio de direccionamiento de memoria de un procesador dado a la placa madre local. Restringir el sistema de esta manera es mucho m'as barato que usar una implementaci'on de memoria compartida entre las placas. Sin embargo, las diferentes necesidades de memoria de las aplicaciones que se ejecutan en cada placa pueden dar lugar a un desequilibrio en el uso de memoria entre las placas. Esta situaci'on puede desencadenar intercambios de datos con el disco, los cuales degradan notablemente las prestaciones del sistema, a pesar de que pueda haber memoria no utilizada en otras placas. Una soluci'on directa consiste en aumentar la cantidad de memoria disponible en cada placa, pero el coste de esta soluci'on puede ser prohibitivo. Por otra parte, el hardware de acceso a memoria remota (RMA) es una forma de facilitar interconexiones r'apidas entre las placas de un cl'uster de computadores. En trabajos recientes, esta caracter'¿stica se ha usado para aumentar el espacio de direccionamiento en ciertas placas. En este trabajo, la m'aquina base usa esta capacidad como mecanismo r'apido para permitir al sistema operativo local acceder a la memoria DRAM instalada en una placa remota. En este contexto, una plani¿caci'on de memoria e¿ciente constituye una cuesti'on cr'¿tica, ya que las latencias de memoria tienen un impacto importante sobre el tiempo de ejecuci'on global de las aplicaciones, debido a que las latencias de memoria remota pueden ser varios 'ordenes de magnitud m'as altas que los accesos locales. Adem'as, el hecho de cambiar la distribuci'on de memoria es un proceso lento que puede involucrar a varias placas, as'¿ pues, el plani¿cador de memoria ha de asegurarse de que la distribuci'on objetivo proporciona mejores prestaciones que la actual. La presente disertaci'on pretende abordar los asuntos mencionados anteriormente mediante la propuesta de varias pol'¿ticas de plani¿caci'on de memoria. En primer lugar, se presenta un algoritmo ideal y una estrategia heur'¿stica para asignar memoria principal ubicada en las diferentes regiones de memoria. Adicionalmente, se ha dise¿nado un mecanismo de control de Calidad de Servicio para evitar que las prestaciones de las aplicaciones en ejecuci'on se degraden de forma inadmisible. El algoritmo ideal encuentra la distribuci'on de memoria 'optima pero su complejidad computacional es prohibitiva dado un alto n'umero de aplicaciones. De este inconveniente se encarga la estrategia heur'¿stica, la cual se aproxima a la mejor distribuci'on de memoria local y remota con un coste computacional aceptable. Los algoritmos anteriores se basan en pro¿ling. Para tratar este defecto potencial, nos centramos en soluciones anal'¿ticas. Esta disertaci'on propone un modelo anal'¿tico que estima el tiempo de ejecuci'on de una aplicaci'on dada para cierta distribuci'on de memoria. Dicha t'ecnica se usa como un predictor de prestaciones que proporciona la informaci'on de entrada a un plani¿cador de memoria. El plani¿cador de memoria usa las estimaciones para elegir din'amicamente la distribuci'on de memoria objetivo 'optima para cada aplicaci'on que se est'e ejecutando en el sistema, de forma que se alcancen las mejores prestaciones globales. La plani¿caci'on a granularidad m'as alta permite pol'¿ticas de plani¿caci'on m'as simples. Este trabajo estudia la viabilidad de plani¿car a nivel de granularidad de p'agina del sistema operativo. Un entrelazado convencional basado en hardware a nivel de bloque y un entrelazado a nivel de p'agina de sistema operativo se han tomado como esquemas de referencia. De la comparaci'on de ambos esquemas de referencia, hemos concluido que solo algunas aplicaciones se ven afectadas de forma signi¿cativa por el uso del entrelazado a nivel de p'agina. Las razones que causan este impacto en las prestaciones han sido estudiadas y han de¿nido la base para el dise¿no de dos pol'¿ticas de distribuci'on de memoria basadas en sistema operativo. La primera se denomina on-demand (OD), y es una estrategia simple que funciona colocando las p'aginas nuevas en memoria local hasta que dicha regi'on se llena, de manera que se bene¿cia de la premisa de que las p'aginas m'as accedidas se piden y se ubican antes que las menos accedidas para mejorar las prestaciones. Sin embargo, ante la ausencia de dicha premisa para algunos de los benchmarks, OD funciona peor. La segunda pol'¿tica, denominada Most-accessed in-local (Mail), se propone con el objetivo de evitar este problema.
Cluster computers represent a cost-effective alternative solution to supercomputers. In these systems, it is common to constrain the memory address space of a given processor to the local motherboard. Constraining the system in this way is much cheaper than using a full-fledged shared memory implementation among motherboards. However, memory usage among motherboards may be unfairly balanced depending on the memory requirements of the applications running on each motherboard. This situation can lead to disk-swapping, which severely degrades system performance, although there may be unused memory on other motherboards. A straightforward solution is to increase the amount of available memory in each motherboard, but the cost of this solution may become prohibitive. On the other hand, remote memory access (RMA) hardware provides fast interconnects among the motherboards of a cluster computer. In recent works, this characteristic has been used to extend the addressable memory space of selected motherboards. In this work, the baseline machine uses this capability as a fast mechanism to allow the local OS to access to DRAM memory installed in a remote motherboard. In this context, efficient memory scheduling becomes a major concern since main memory latencies have a strong impact on the overall execution time of the applications, provided that remote memory accesses may be several orders of magnitude higher than local accesses. Additionally, changing the memory distribution is a slow process which may involve several motherboards, hence the memory scheduler needs to make sure that the target distribution provides better performance than the current one. This dissertation aims to address the aforementioned issues by proposing several memory scheduling policies. First, an ideal algorithm and a heuristic strategy to assign main memory from the different memory regions are presented. Additionally, a Quality of Service control mechanism has been devised in order to prevent unacceptable performance degradation for the running applications. The ideal algorithm finds the optimal memory distribution but its computational cost is prohibitive for a high number of applications. This drawback is handled by the heuristic strategy, which approximates the best local and remote memory distribution among applications at an acceptable computational cost. The previous algorithms are based on profiling. To deal with this potential shortcoming we focus on analytical solutions. This dissertation proposes an analytical model that estimates the execution time of a given application for a given memory distribution. This technique is used as a performance predictor that provides the input to a memory scheduler. The estimates are used by the memory scheduler to dynamically choose the optimal target memory distribution for each application running in the system in order to achieve the best overall performance. Scheduling at a higher granularity allows simpler scheduler policies. This work studies the feasibility of scheduling at OS page granularity. A conventional hardware-based block interleaving and an OS-based page interleaving have been assumed as the baseline schemes. From the comparison of the two baseline schemes, we have concluded that only the performance of some applications is significantly affected by page-based interleaving. The reasons that cause this impact on performance have been studied and have provided the basis for the design of two OS-based memory allocation policies. The first one, namely on-demand (OD), is a simple strategy that works by placing new pages in local memory until this region is full, thus benefiting from the premise that most of the accessed pages are requested and allocated before than the least accessed ones to improve the performance. Nevertheless, in the absence of this premise for some benchmarks, OD performs worse. The second policy, namely Most-accessed in-local (Mail), is proposed to avoid this problem
Serrano Gómez, M. (2013). Scheduling Local and Remote Memory in Cluster Computers [Tesis doctoral]. Editorial Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/31639
Alfresco

Cloud Computing presents a novel way for businesses to procure their IT needs. Its elasticity and on-demand provisioning enables a shift from capital expenditures to operating expenses, giving businesses the technological agility they need to respond to an ever-changing marketplace. The rapid adoption of Cloud Computing, however, poses a unique challenge to Cloud providers—their already very large electricity bill and carbon footprint will get larger as they expand; managing both costs is therefore essential to their growth. This thesis squarely addresses the above challenge. Recognizing the presence of Cloud data centers in multiple locations and the differences in electricity price and emission intensity among these locations and over time, we develop an optimization framework that couples workload distribution with time-varying signals on electricity price and emission intensity for financial and environmental benefits. The framework is comprised of an optimization model, an aggregate cost function, and 6 scheduling heuristics. To evaluate cost savings, we run simulations with 5 data centers located across North America over a period of 81 days. We use historical data on electricity price, emission intensity, and workload collected from market operators and research data archives. We find that our framework can produce substantial cost savings, especially when workloads are distributed both geographically and temporally—up to 53.35% on electricity cost, or 29.13% on carbon cost, or 51.44% on electricity cost and 13.14% on carbon cost simultaneously.

The continuous shrink of transistor sizes has allowed more complex and powerful devices to be implemented in the same area, which provides new capabilities and functionalities. However, this complexity increase comes with a considerable rise in power consumption. This situation is critical in portable devices where the energy budget is limited and, hence, battery lifetime defines the usefulness of the system. Therefore, power consumption has become a major concern in the design of real-time multicore embedded systems. This dissertation proposes several techniques aimed to save energy without sacrifying real-time schedulability in this type of systems. The proposed techniques deal with different main components of the system. In particular, the techniques affect the task partitioner and the scheduler, as well as the memory controller. Some of the techniques are especially tailored for multicores with shared Dynamic Voltage and Frequency Scaling (DVFS) domains. Workload balancing among cores in a given domain has a strong impact on power consumption, since all the cores sharing a DVFS domain must run at the speed required by the most loaded core. In this thesis, a novel workload partitioning algorithm is proposed, namely Loadbounded Resource Balancing (LRB). The proposal allocates tasks to cores to balance a given resource (processor or memory) consumption among cores, improving real-time schedulability by increasing overlapping between processor and memory. However, distributing tasks in this way regardless the individual core utilizations could lead to unfair load distributions. That is, one of the cores could become much loaded than the others. To avoid this scenario, when a given utilization threshold is exceeded, tasks are assigned to the least loaded core. Unfortunately, workload partitioning alone is sometimes not able to achieve a good workload balance among cores. Therefore, this work also explores novel task migration approaches. Two task migration heuristics are proposed. The first heuristic, referred to as Single Option Migration (SOM ), attempts to perform only one migration when the workload changes to improve utilization balance. Three variants of the SOM algorithm have been devised, depending on the point of time the migration attempt is performed: when a task arrives to the system (SOMin), when a task leaves the system (SOMout), and in both cases (SOMin−out). The second heuristic, referred to as Multiple Option Migration (MOM ) explores an additional alternative workload partitioning before performing the migration attempt. Regarding the memory controller, memory controller scheduling policies are devised. Conventional policies used in Non Real-Time (NRT) systems are not appropriate for systems providing support for both Hard Real-Time (HRT) and Soft Real-Time (SRT) tasks. Those policies can introduce variability in the latencies of the memory requests and, hence, cause an HRT deadline miss that could lead to a critical failure of the real-time system. To deal with this drawback, a simple policy, referred to as HR- first, which prioritizes requests of HRT tasks, is proposed. In addition, a more advanced approach, namely ATR-first, is presented. ATR-first prioritizes only those requests of HRT tasks that are necessary to ensure real-time schedulability, improving the Quality of Service (QoS) of SRT tasks. Finally, this thesis also tackles dynamic execution time estimation. The accuracy of this estimation is important to avoid deadline misses of HRT tasks but also to increase QoS in SRT systems. Besides, it can also help to improve the schedulability of the systems and reduce power consumption. The Processor-Memory (Proc-Mem) model, that dynamically predicts the execution time of real-time application for each frequency level, is proposed. This model measures at the first hyperperiod, making use of Performance Monitoring Counters (PMCs) at run-time, the portion of time that each core is performing computation (CPU ), waiting for memory (MEM ), or both (OVERLAP). This information will be used to estimate the execution time at any other working frequency
March Cabrelles, JL. (2014). Dynamic Power-Aware Techniques for Real-Time Multicore Embedded Systems [Tesis doctoral]. Editorial Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/48464
TESIS

Changes in reimbursement make it imperative for nurse managers to develop tools and methods to assist them to stay within budget. Disparity between planned staffing and required staffing often requires supplemental staffing and overtime. In addition, many states are now mandating staffing committees to demonstrate effective staff planning. This retrospective quantitative study developed an empirical method for building nursing unit staffing plans through the incorporation of patient acuity and patient turnover as adjustments towards planning nursing workload. The theoretical framework used to guide this study was structural contingency theory (SCT). Patient turnover was measured by Unit Activity Index (UAI). Patient acuity was measured using case mix index (CMI). Nursing workload was measured as hours per patient day (HPPD). The adjustment to HPPD was made through the derivation of a weight factor based on UAI and CMI. The study consisted of fourteen medical, surgical, and mixed medical-surgical units within a large academic healthcare center. Data from 3 fiscal years were used. This study found that there were significant, but generally weak correlations between UAI and CMI and HPPD. The method of deriving a weight factor for adjusting HPPD was not as important as the decision-making relative to when to adjust planned HPPD. In addition, the measure of unit activity index was simplified which will assist researchers to more easily calculate patient turnover. As a result of this study, nurse managers and will be better able to adjust and predict HPPD in cases where benchmarking has been problematic. Data-driven adjustments to HPPD based on UAI and CMI will assist the nurse manager to plan and budget resources more effectively.

Esta tese apresenta análises de dois problemas de máquina única relacionados à programação da produção com seqüência predefinida. Para ambos os problemas são sugeridas modelagens via programação matemática e algoritmos que encontram a solução ótima em tempo polinomial e pseudo-polinomial. O primeiro problema é o de inserção de espera no problema com função-objetivo que considera s soma de funções convexas do horário de término independentes para cada ordem. O segundo considera custos distintos de adiantamento e atraso para cada ordem e custos de utilização de capacidade adicional ponderados distintamente para cada período de capacidade adicional que possa ser utilizado. Sugere-se adicionalmente um procedimento onde o mesmo avalia a melhor opção entre se utilizar tempo de espera, horas-extras e criar ou eliminar turnos de trabalho. São feitas análises e algumas generalizações como a utilização de diversos intervalos de tempo com diferentes custos concatenados e uma sugestão para a utilização dos procedimentos num ambiente de múltiplas máquinas
This thesis analyses two cases of one-machine problem regarding to production scheduling with fixed sequence. In both problems, modeling with mathematical programming, and (pseudo)polynomial-time algorithms are suggested. The first problem deals with idle time insertion in the problem where the objective function (represented by a sum of costs) considers that each job has costs described as any convex function of its completion time. The second problem considers earliness and tardiness with distinct costs for each job considering the possible use of additional capacity. For the additional capacity we assume that there are distinct costs for each time period where jobs can be processed. A procedure dealing with options of either to change the number of shifts or to utilize overtime considering total costs is suggested. Analysis and generalizations based on the utilization of several contiguous time periods with distinct costs and a heuristic extension for the multiple-machine case are also presented

In recent years there has been an extraordinary growth of large-scale data processing and related technologies in both, industry and academic communities. This trend is mostly driven by the need to explore the increasingly large amounts of information that global companies and communities are able to gather, and has lead the introduction of new tools and models, most of which are designed around the idea of handling huge amounts of data. A good example of this trend towards improved large-scale data processing is MapReduce, a programming model intended to ease the development of massively parallel applications, and which has been widely adopted to process large datasets thanks to its simplicity. While the MapReduce model was originally used primarily for batch data processing in large static clusters, nowadays it is mostly deployed along with other kinds of workloads in shared environments in which multiple users may be submitting concurrent jobs with completely different priorities and needs: from small, almost interactive, executions, to very long applications that take hours to complete. Scheduling and selecting tasks for execution is extremely relevant in MapReduce environments since it governs a job's opportunity to make progress and determines its performance. However, only basic primitives to prioritize between jobs are available at the moment, constantly causing either under or over-provisioning, as the amount of resources needed to complete a particular job are not obvious a priori. This thesis aims to address both, the lack of management capabilities and the increased complexity of the environments in which MapReduce is executed. To that end, new models and techniques are introduced in order to improve the scheduling of MapReduce in the presence of different constraints found in real-world scenarios, such as completion time goals, data locality, hardware heterogeneity, or availability of resources. The focus is on improving the integration of MapReduce with the computing infrastructures in which it usually runs, allowing alternative techniques for dynamic management and provisioning of resources. More specifically, it is focused in three scenarios that are incremental in its scope. First, it studies the prospects of using high-level performance criteria to manage and drive the performance of MapReduce applications, taking advantage of the fact that MapReduce is executed in controlled environments in which the status of the cluster is known. Second, it examines the feasibility and benefits of making the MapReduce runtime more aware of the underlying hardware and the characteristics of applications. And finally, it also considers the interaction between MapReduce and other kinds of workloads, proposing new techniques to handle these increasingly complex environments. Following these three items described above, this thesis contributes to the management of MapReduce workloads by 1) proposing a performance model for MapReduce workloads and a scheduling algorithm that leverages the proposed model and is able to adapt depending on the various needs of its users in the presence of completion time constraints; 2) proposing a new resource model for MapReduce and a placement algorithm aware of the underlying hardware as well as the characteristics of the applications, capable of improving cluster utilization while still being guided by job performance metrics; and 3) proposing a model for shared environments in which MapReduce is executed along with other kinds of workloads such as transactional applications, and a scheduler aware of these workloads and its expected demand of resources, capable of improving resource utilization across machines while observing completion time goals.

In recent years there has been an extraordinary growth of large-scale data processing and related technologies in both, industry and academic communities. This trend is mostly driven by the need to explore the increasingly large amounts of information that global companies and communities are able to gather, and has lead the introduction of new tools and models, most of which are designed around the idea of handling huge amounts of data. A good example of this trend towards improved large-scale data processing is MapReduce, a programming model intended to ease the development of massively parallel applications, and which has been widely adopted to process large datasets thanks to its simplicity. While the MapReduce model was originally used primarily for batch data processing in large static clusters, nowadays it is mostly deployed along with other kinds of workloads in shared environments in which multiple users may be submitting concurrent jobs with completely different priorities and needs: from small, almost interactive, executions, to very long applications that take hours to complete. Scheduling and selecting tasks for execution is extremely relevant in MapReduce environments since it governs a job's opportunity to make progress and determines its performance. However, only basic primitives to prioritize between jobs are available at the moment, constantly causing either under or over-provisioning, as the amount of resources needed to complete a particular job are not obvious a priori. This thesis aims to address both, the lack of management capabilities and the increased complexity of the environments in which MapReduce is executed. To that end, new models and techniques are introduced in order to improve the scheduling of MapReduce in the presence of different constraints found in real-world scenarios, such as completion time goals, data locality, hardware heterogeneity, or availability of resources. The focus is on improving the integration of MapReduce with the computing infrastructures in which it usually runs, allowing alternative techniques for dynamic management and provisioning of resources. More specifically, it is focused in three scenarios that are incremental in its scope. First, it studies the prospects of using high-level performance criteria to manage and drive the performance of MapReduce applications, taking advantage of the fact that MapReduce is executed in controlled environments in which the status of the cluster is known. Second, it examines the feasibility and benefits of making the MapReduce runtime more aware of the underlying hardware and the characteristics of applications. And finally, it also considers the interaction between MapReduce and other kinds of workloads, proposing new techniques to handle these increasingly complex environments. Following these three items described above, this thesis contributes to the management of MapReduce workloads by 1) proposing a performance model for MapReduce workloads and a scheduling algorithm that leverages the proposed model and is able to adapt depending on the various needs of its users in the presence of completion time constraints; 2) proposing a new resource model for MapReduce and a placement algorithm aware of the underlying hardware as well as the characteristics of the applications, capable of improving cluster utilization while still being guided by job performance metrics; and 3) proposing a model for shared environments in which MapReduce is executed along with other kinds of workloads such as transactional applications, and a scheduler aware of these workloads and its expected demand of resources, capable of improving resource utilization across machines while observing completion time goals.

Nell’ambito dei sistemi High-Performance Computing, l'uso di euristiche di dispatching efficaci, per lo scheduling e l'allocazione dei jobs in arrivo, è fondamentale al fine di ottenere buoni livelli di Quality of Service. In questo elaborato ci concentreremo sul design e l’analisi di euristiche di allocazione delle risorse, che saranno progettate per sistemi HPC eterogenei, nei quali i nodi possono essere equipaggiati con diverse tipologie di unità di elaborazione. Impiegheremo poi euristiche data-driven per la predizione della durata dei jobs, e valuteremo il tutto dal punto di vista del throughput di sistema. Considereremo in particolare Eurora, un sistema HPC eterogeneo realizzato da CINECA, oltre che un workload catturato dal relativo log di sistema, contenente jobs reali inviati dagli utenti. Tutto ciò è stato possibile grazie ad AccaSim, un simulatore di sistemi HPC sviluppato nel Dipartimento di Informatica - Scienza e Ingegneria (DISI) dell’Università di Bologna, ed al quale si è contribuito in modo sostanziale. Quest’elaborato mostra che l’impatto di diverse euristiche di allocazione sul throughput di un sistema HPC eterogeneo non è trascurabile, con variazioni in grado di raggiungere picchi di un ordine di grandezza, e più pronunciate considerando brevi intervalli temporali, dell'ordine dei mesi. Abbiamo inoltre osservato che l’impiego di euristiche per la predizione della durata dei jobs è di grande beneficio al throughput su tutte le euristiche di allocazione, e specialmente su quelle che integrano in maniera più profonda tali elementi data-driven. Infine, l’analisi effettuata ha permesso di caratterizzare integralmente il sistema Eurora ed il relativo workload, permettendoci di comprendere al meglio gli effetti su di esso dei diversi metodi di dispatching, nonché di estendere le nostre considerazioni anche ad altre classi di sistemi.

This dissertation presents and evaluates a methodology for scheduling medical application workloads in virtualized computing environments. Such environments are being widely adopted by providers of “cloud computing” services. In the context of provisioning resources for medical applications, such environments allow users to deploy applications on distributed computing resources while keeping their data secure. Furthermore, higher level services that further abstract the infrastructure-related issues can be built on top of such infrastructures. For example, a medical imaging service can allow medical professionals to process their data in the cloud, easing them from the burden of having to deploy and manage these resources themselves. In this work, we focus on issues related to scheduling scientific workloads on virtualized environments. We build upon the knowledge base of traditional parallel job scheduling to address the specific case of medical applications while harnessing the benefits afforded by virtualization technology. To this end, we provide the following contributions: An in-depth analysis of the execution characteristics of the target applications when run in virtualized environments. A performance prediction methodology applicable to the target environment. A scheduling algorithm that harnesses application knowledge and virtualization-related benefits to provide strong scheduling performance and quality of service guarantees. In the process of addressing these pertinent issues for our target user base (i.e. medical professionals and researchers), we provide insight that benefits a large community of scientific application users in industry and academia. Our execution time prediction and scheduling methodologies are implemented and evaluated on a real system running popular scientific applications. We find that we are able to predict the execution time of a number of these applications with an average error of 15%. Our scheduling methodology, which is tested with medical image processing workloads, is compared to that of two baseline scheduling solutions and we find that it outperforms them in terms of both the number of jobs processed and resource utilization by 20-30%, without violating any deadlines. We conclude that our solution is a viable approach to supporting the computational needs of medical users, even if the cloud computing paradigm is not widely adopted in its current form.

In recent years there has been an extraordinary growth of the Internet of Things (IoT) and its protocols. The increasing diffusion of electronic devices with identification, computing and communication capabilities is laying ground for the emergence of a highly distributed service and networking environment. The above mentioned situation implies that there is an increasing demand for advanced IoT data management and processing platforms. Such platforms require support for multiple protocols at the edge for extended connectivity with the objects, but also need to exhibit uniform internal data organization and advanced data processing capabilities to fulfill the demands of the application and services that consume IoT data. One of the initial approaches to address this demand is the integration between IoT and the Cloud computing paradigm. There are many benefits of integrating IoT with Cloud computing. The IoT generates massive amounts of data, and Cloud computing provides a pathway for that data to travel to its destination. But today’s Cloud computing models do not quite fit for the volume, variety, and velocity of data that the IoT generates. Among the new technologies emerging around the Internet of Things to provide a new whole scenario, the Fog Computing paradigm has become the most relevant. Fog computing was introduced a few years ago in response to challenges posed by many IoT applications, including requirements such as very low latency, real-time operation, large geo-distribution, and mobility. Also this low latency, geo-distributed and mobility environments are covered by the network architecture MEC (Mobile Edge Computing) that provides an IT service environment and Cloud-computing capabilities at the edge of the mobile network, within the Radio Access Network (RAN) and in close proximity to mobile subscribers. Fog computing addresses use cases with requirements far beyond Cloud-only solution capabilities. The interplay between Cloud and Fog computing is crucial for the evolution of the so-called IoT, but the reach and specification of such interplay is an open problem. This thesis aims to find the right techniques and design decisions to build a scalable distributed system for the IoT under the Fog Computing paradigm to ingest and process data. The final goal is to explore the trade-offs and challenges in the design of a solution from Edge to Cloud to address opportunities that current and future technologies will bring in an integrated way. This thesis describes an architectural approach that addresses some of the technical challenges behind the convergence between IoT, Cloud and Fog with special focus on bridging the gap between Cloud and Fog. To that end, new models and techniques are introduced in order to explore solutions for IoT environments. This thesis contributes to the architectural proposals for IoT ingestion and data processing by 1) proposing the characterization of a platform for hosting IoT workloads in the Cloud providing multi-tenant data stream processing capabilities, the interfaces over an advanced data-centric technology, including the building of a state-of-the-art infrastructure to evaluate the performance and to validate the proposed solution. 2) studying an architectural approach following the Fog paradigm that addresses some of the technical challenges found in the first contribution. The idea is to study an extension of the model that addresses some of the central challenges behind the converge of Fog and IoT. 3) Design a distributed and scalable platform to perform IoT operations in a moving data environment. The idea after study data processing in Cloud, and after study the convenience of the Fog paradigm to solve the IoT close to the Edge challenges, is to define the protocols, the interfaces and the data management to solve the ingestion and processing of data in a distributed and orchestrated manner for the Fog Computing paradigm for IoT in a moving data environment.
En els últims anys hi ha hagut un gran creixement del Internet of Things (IoT) i els seus protocols. La creixent difusió de dispositius electrònics amb capacitats d'identificació, computació i comunicació esta establint les bases de l’aparició de serveis altament distribuïts i del seu entorn de xarxa. L’esmentada situació implica que hi ha una creixent demanda de plataformes de processament i gestió avançada de dades per IoT. Aquestes plataformes requereixen suport per a múltiples protocols al Edge per connectivitat amb el objectes, però també necessiten d’una organització de dades interna i capacitats avançades de processament de dades per satisfer les demandes de les aplicacions i els serveis que consumeixen dades IoT. Una de les aproximacions inicials per abordar aquesta demanda és la integració entre IoT i el paradigma del Cloud computing. Hi ha molts avantatges d'integrar IoT amb el Cloud. IoT genera quantitats massives de dades i el Cloud proporciona una via perquè aquestes dades viatgin a la seva destinació. Però els models actuals del Cloud no s'ajusten del tot al volum, varietat i velocitat de les dades que genera l'IoT. Entre les noves tecnologies que sorgeixen al voltant del IoT per proporcionar un escenari nou, el paradigma del Fog Computing s'ha convertit en la més rellevant. Fog Computing es va introduir fa uns anys com a resposta als desafiaments que plantegen moltes aplicacions IoT, incloent requisits com baixa latència, operacions en temps real, distribució geogràfica extensa i mobilitat. També aquest entorn està cobert per l'arquitectura de xarxa MEC (Mobile Edge Computing) que proporciona serveis de TI i capacitats Cloud al edge per la xarxa mòbil dins la Radio Access Network (RAN) i a prop dels subscriptors mòbils. El Fog aborda casos d’us amb requisits que van més enllà de les capacitats de solucions només Cloud. La interacció entre Cloud i Fog és crucial per a l'evolució de l'anomenat IoT, però l'abast i especificació d'aquesta interacció és un problema obert. Aquesta tesi té com objectiu trobar les decisions de disseny i les tècniques adequades per construir un sistema distribuït escalable per IoT sota el paradigma del Fog Computing per a ingerir i processar dades. L'objectiu final és explorar els avantatges/desavantatges i els desafiaments en el disseny d'una solució des del Edge al Cloud per abordar les oportunitats que les tecnologies actuals i futures portaran d'una manera integrada. Aquesta tesi descriu un enfocament arquitectònic que aborda alguns dels reptes tècnics que hi ha darrere de la convergència entre IoT, Cloud i Fog amb especial atenció a reduir la bretxa entre el Cloud i el Fog. Amb aquesta finalitat, s'introdueixen nous models i tècniques per explorar solucions per entorns IoT. Aquesta tesi contribueix a les propostes arquitectòniques per a la ingesta i el processament de dades IoT mitjançant 1) proposant la caracterització d'una plataforma per a l'allotjament de workloads IoT en el Cloud que proporcioni capacitats de processament de flux de dades multi-tenant, les interfícies a través d'una tecnologia centrada en dades incloent la construcció d'una infraestructura avançada per avaluar el rendiment i validar la solució proposada. 2) estudiar un enfocament arquitectònic seguint el paradigma Fog que aborda alguns dels reptes tècnics que es troben en la primera contribució. La idea és estudiar una extensió del model que abordi alguns dels reptes centrals que hi ha darrere de la convergència de Fog i IoT. 3) Dissenyar una plataforma distribuïda i escalable per a realitzar operacions IoT en un entorn de dades en moviment. La idea després d'estudiar el processament de dades en el Cloud, i després d'estudiar la conveniència del paradigma Fog per resoldre els desafiaments de IoT a prop del Edge, és definir els protocols, les interfícies i la gestió de dades per resoldre la ingestió i processament de dades d’una manera més eficient

Le domaine du Calcul à Haute Performance (HPC) évolue étroitement avec les dernières avancées technologiques des architectures informatiques et des besoins toujours croissants en demande de puissance de calcul. Cette thèse s'intéresse à l'étude d'un type d'intergiciel particulier appelé gestionnaire de tâches et ressources (RJMS) qui est chargé de distribuer la puissance de calcul aux applications dans les plateformes pour le HPC. Le RJMS joue un rôle central du fait de sa position dans la pile logicielle. Les dernières évolutions dans les couches matérielles et dans les applications ont largement augmenté le niveau de complexité auquel doit faire face ce type d'intergiciel. Des problématiques telles que le passage à l'échelle, la prise en compte d'un taux d'activité irrégulier, la gestion des contraintes liées à la topologie du matériel, l'efficacité énergétique et la tolérance aux pannes doivent être particulièrement pris en considération, afin, entre autres, de fournir une meilleure exploitation des ressources à la fois du point de vue global du système ainsi que de celui des utilisateurs. La première contribution de cette thèse est un état de l'art sur la gestion des tâches et des ressources ainsi qu'une analyse comparative des principaux intergiciels actuels et des différentes problématiques de recherche associées. Une métrique importante pour évaluer l'apport d'un RJMS sur une plate-forme est le niveau d'utilisation de l'ensemble du système. On constate parmi les traces d'activité de plusieurs plateformes qu'un grand nombre d'entre elles présentent un taux d'utilisation significativement inférieure à une pleine utilisation. Ce constat est la principale motivation des autres contributions de cette thèse qui portent sur les méthodes d'exploitations de ces périodes de sous-utilisation au profit de la gestion globale du système ou des applications en court d'exécution. Plus particulièrement cette thèse explore premièrement, les moyens d'accroître le taux de calculs utiles dans le contexte des grilles légères en présence d'une forte variabilité de la disponibilité des ressources de calcul. Deuxièmement, nous avons étudié le cas des tâches dynamiques et proposé différentes techniques s'intégrant au RJMS OAR et troisièmement nous évalués plusieurs modes d'exploitation des ressources en prenant en compte la consommation énergétique. Finalement, les évaluations de cette thèse reposent sur une approche expérimentale pour laquelle nous avons proposés des outils et une méthodologie permettant d'améliorer significativement la maîtrise et la reproductibilité d'expériences complexes propre à ce domaine d'étude
High Performance Computing is characterized by the latest technological evolutions in computing architectures and by the increasing needs of applications for computing power. A particular middleware called Resource and Job Management System (RJMS), is responsible for delivering computing power to applications. The RJMS plays an important role in HPC since it has a strategic place in the whole software stack because it stands between the above two layers. However, the latest evolutions in hardware and applications layers have provided new levels of complexities to this middleware. Issues like scalability, management of topological constraints, energy efficiency and fault tolerance have to be particularly considered, among others, in order to provide a better system exploitation from both the system and user point of view. This dissertation provides a state of the art upon the fundamental concepts and research issues of Resources and Jobs Management Systems. It provides a multi-level comparison (concepts, functionalities, performance) of some Resource and Jobs Management Systems in High Performance Computing. An important metric to evaluate the work of a RJMS on a platform is the observed system utilization. However, studies and logs of production platforms show that HPC systems in general suffer of significant un-utilization rates. Our study deals with these clusters' un-utilization periods by proposing methods to aggregate otherwise un-utilized resources for the benefit of the system or the application. More particularly this thesis explores RJMS level mechanisms: 1) for increasing the jobs valuable computation rates in the high volatile environments of a lightweight grid context, 2) for improving system utilization with malleability techniques and 3) providing energy efficient system management through the exploitation of idle computing machines. The experimentation and evaluation in this type of contexts provide important complexities due to the inter-dependency of multiple parameters that have to be taken into control. In this thesis we have developed a methodology based upon real-scale controlled experimentation with submission of synthetic or real workload traces

Many breakthroughs in scientific and industrial research are supported by simulations and calculations performed on high performance computing (HPC) systems. These systems typically consist of uniform, largely parallel compute resources and high bandwidth concurrent file systems interconnected by low latency synchronous networks. HPC systems are managed by batch schedulers that order the execution of application jobs to maximize utilization while steering turnaround time. In the past, demands for greater capacity were met by building more powerful systems with more compute nodes, greater transistor densities, and higher processor operating frequencies. Unfortunately, the scope for further increases in processor frequency is restricted by the limitations of semiconductor technology. Instead, parallelism within processors and in numbers of compute nodes is increasing, while the capacity of single processing units remains unchanged. In addition, HPC systems’ memory and I/O hierarchies are becoming deeper and more complex to keep up with the systems’ processing power. HPC applications are also changing: the need to analyze large data sets and simulation results is increasing the importance of data processing and data-intensive applications. Moreover, composition of applications through workflows within HPC centers is becoming increasingly important. This thesis addresses the HPC scheduling challenges created by such new systems and applications. It begins with a detailed analysis of the evolution of the workloads of three reference HPC systems at the National Energy Research Supercomputing Center (NERSC), with a focus on job heterogeneity and scheduler performance. This is followed by an analysis and improvement of a fairshare prioritization mechanism for HPC schedulers. The thesis then surveys the current state of the art and expected near-future developments in HPC hardware and applications, and identifies unaddressed scheduling challenges that they will introduce. These challenges include application diversity and issues with workflow scheduling or the scheduling of I/O resources to support applications. Next, a cloud-inspired HPC scheduling model is presented that can accommodate application diversity, takes advantage of malleable applications, and enables short wait times for applications. Finally, to support ongoing scheduling research, an open source scheduling simulation framework is proposed that allows new scheduling algorithms to be implemented and evaluated in a production scheduler using workloads modeled on those of a real system. The thesis concludes with the presentation of a workflow scheduling algorithm to minimize workflows’ turnaround time without over-allocating resources.

Work also supported by the U.S. Department of Energy, Office of Science, Office of Advanced Scientific Computing Research (ASCR) and we used resources at the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility, supported by the Officece of Science of the U.S. Department of Energy, both under Contract No. DE-AC02-05CH11231.

Heavy-tailed workload distributions are commonly experienced in many areas of distributed computing. Such workloads are highly variable, where a small number of very large tasks make up a large proportion of the workload, making the load very hard to distribute effectively. Traditional task assignment policies are ineffective under these conditions as they were formulated based on the assumption of an exponentially distributed workload. Size-based task assignment policies have been proposed to handle heavy-tailed workloads, but their applications are limited by their static nature and assumption of prior knowledge of a task's service requirement. This thesis analyses existing approaches to load distribution under heavy-tailed workloads, and presents a new generalised task assignment policy that significantly improves performance for many distributed applications, by intelligently addressing the negative effects on performance that highly variable workloads cause. Many problems associated with the modelling and optimisations of systems under highly variable workloads were then addressed by a novel technique that approximated these workloads with simpler mathematical representations, without losing any of their pertinent original properties. Finally, we obtain advance queuing metrics (such as the variance of key measurements like waiting time and slowdown that are difficult to obtain analytically) through rigorous simulation.

碩士
國立屏東大學
資訊工程學系碩士班
105
Virtualization is the most popular technology in recent years. It shares the physical resources of the physical machine to multiple virtual machines. The cost and space of server management can be reduced significantly if the physical resources are allocated wisely. The Xen Project is a well-know virtualization platform created by Keir Fraser and Ian Pratt. In this thesis, we shall target the physical resources allocation problem for xen virtualization systems. In particular, we are interested in dynamic CPU scheduling for virtual machines such that the performance of the entire system could be improved. At the current stage, existing CPU scheduler for xen are all in static manner. In other words, it allocation physical CPUs to virtual CPUs offline. However, we can not predict workload (i.e, number of tasks of each virtual machine at the runtime.In this thesis, an RTDS-based CPU scheduler, called enhanced real-time deferrable server(ERTDS), is proposed an additional capacity to virtual CPUs when their run-time requirements are higher than expected. The proposed ERTDS has been implemented in Xen 4.7 and a series of experiments has been conducted for which we have some encouraging results.

碩士
國立臺中教育大學
資訊工程學系
107
Workflow scheduling is an NP-complete problem, and has always been an important research topic on parallel job scheduling. Different types of scheduling heuristics have been developed for tackling the challenging workflow scheduling problem. List-based scheduling is one of the most widely used categories for different workflow structures and optimization goals. Among various list-based workflow scheduling algorithms, HEFT and PEFT are two typical and well-known representatives with important innovations. However, none of them can consistently retain advantage over the other across various workflow properties and system scales. That motivated our research work in this thesis, focusing on the problem of task-parallel workflow scheduling on homogeneous parallel systems which have become feasible and common on current cloud computing platforms with virtualization technology. Based on thorough experimental analysis of HEFT and PEFT, we found drawbacks of them, and then developed a new workload- and resource-aware list-based workflow scheduling approach featuring three new mechanisms, including structural task ranking, task ranking based on allocated critical path, and adaptable task allocation. The proposed approach was evaluated with a series of simulation experiments based on both synthetic workflow structures and real-world workflow models, and compared to HEFT and PEFT. The experimental results demonstrate that our approach can achieve significantly superior performance in most circumstances in terms of both average makespan and best ratio.

碩士
國立中正大學
資訊工程研究所
101
Due to the diversity in computing capabilities of processors in heterogeneous multicore systems, it is difficult to come up with a perfect task scheduling algorithm that can avoid all processors from becoming idle at some point in time during the whole schedule. The situation becomes worse when the capabilities of processors differ by a large margin or the ratio of communication time between tasks to the computation time of tasks is very large. Nevertheless, it is this imperfection that motivates this Thesis to propose a re-scheduling scheme that leverages the characteristics of divisible tasks by partitioning the workload across two different processors so as to fill the holes (idle time slots) in the schedule. Based on the type of hole, constant or varying, different strategies are proposed, including a profiling-based partitioning and an on-the-fly partitioning. Re-scheduling based on a combination of these two strategies results in a decrease in the makespan and total amount of idle time of processors. Experiment results show that the makespan can be decreased by 14% and the total amount of idle time by 50%.

碩士
逢甲大學
工業工程學系
88
In this study, issues of vehicle routing and scheduling problems arising in local distribution centers are addressed. Due to the evolution of local economy from production-oriented markets to customer-oriented markets, a large number of distribution centers and the associated convenience stores are established in major metropolitan areas. Since the distribution services of commodities provided by distribution centers to convenience stores have tremendous impacts on the cost-effective performance of a firm and the level of customer services, issues of planning and management of commodity distribution have been received great concerns in practice. Theoretically, the planning of commodity distribution services is one type of vehicle routing and scheduling problems and can be improved through the application of optimization techniques. The vehicle routing and scheduling problem can be treated as a combinatorial optimization problem in nature and is NP-hard problem. First, we analyze the vehicle routing and scheduling problem arising in local distribution centers. The vehicle routing and scheduling problem under consideration is analyzed in terms of the following criteria: (1) to minimize the number of vehicles needed to provide the required distribution service, (2) to minimize the total travel distance needed to complete the total distribution service, (3) to ensure that every route is balanced in terms of workload, and (4) to ensure that the delivery time of every route is within the required duration. Next, an optimization model for describing this problem is explored and formulated. A heuristic-type solution method is developed by exploiting the property of the developed model. The developed heuristic consists of three stages. First, we generate an initial solution using savings method. Second we devise five procedures using one-point movement, two-point exchange, 2-opt improvement, reinitialization, and infeasibility improvement. Third, we ensure that every route is balanced in terms of workload. A real-world data is collected from a local distribution center and its associated convenience stores. The collected data is used for testing and implementation of the proposed approach. Results suggest that the proposed approach can achieve the foregoing objectives. The developed heuristic is fast and has improved upon previous best-known solution. It can obtain solution of higher accuracy and provide more scheduling information that has substantial contribution to operate the planning and management of commodity distribution.

碩士
東海大學
資訊工程學系
103
In this paper, we will examine how to improve workload balancing on a computing cluster by a parallel loop self-scheduling scheme. We use hybrid MPI and OpenMP parallel programming in C language. The block partition loop is according to the performance weighting of compute nodes. This study implements parallel loop self-scheduling use Xeon Phi, with its characteristics to improve workload balancing between heterogeneous nodes. The parallel loop self-scheduling is composed of the static and dynamic allocation. A weighting algorithm is adopted in the static part while the well-known loop self-scheduling scheme is adopted in the dynamic part. In recent years, Intel promotes its new product Xeon Phi coprocessor, which is similar to the x86 architecture coprocessor. It has about 60 cores and can be regarded as a single computing node, with the computing power that cannot be ignored. In our experiment, we will use a plurality of computing nodes. We compute four applications, i.e., matrix multiplication, sparse matrix multiplication, Mandelbrot set computation, and the circuit satisfiability problem. Our results will show how to do the weight allocation and how to choose a scheduling scheme to achieve the best performance in the parallel loop self-scheduling.

碩士
國立交通大學
網路工程研究所
96
Memory is an important shared resource in a multi-core real-time embedded system. The memory contentions between cores will lengthen the total execution time due to waiting for memory requests being served. In this thesis, we focus on the tasks partition scheduling problem while considering memory contentions in multi-core real-time embedded systems. We propose an energy efficient scheduling mechanism with consideration to the memory workload of tasks, called WAS-DVS (workload-aware scheduling-dynamic voltage scaling), which is an improvement of an existing method, LTF-MES (Largest-Task-First-Minimize-Energy-Scheduling). The main difference between ours and LTF-MES is that we consider the execution order of tasks that may reduce the frequency of memory contentions. Simulation results show that by reducing memory contentions between tasks, the slack time will increase and the proposed WAS-DVS can use it to lower total execution time and total energy consumption on a variety of workloads in multi-core systems. The proposed WAS-DVS can lower the total execution time from 2% to 10.3% before applying DVS and improve the total energy consumption from 3.85% to 19% compared to LTF-MES, under various numbers of tasks and 2 to 16 cores after applying DVS.

In hosting environments such as Infrastructure as a Service (IaaS) clouds, desirable application performance is typically guaranteed through the use of Service Level Agreements (SLAs), which specify minimal fractions of resource capacities that must be allocated by a service provider for unencumbered use by customers to ensure proper operation of their workloads. Most IaaS offerings are presented to customers as fixed-size and fixed-price SLAs, that do not match well the needs of specific applications. Furthermore, arbitrary colocation of applications with different SLAs may result in inefficient utilization of hosts' resources, resulting in economically undesirable customer behavior. In this thesis, we propose the design and architecture of a Colocation as a Service (CaaS) framework: a set of strategic and operational services that allow the efficient colocation of customer workloads. CaaS strategic services provide customers the means to specify their application workload using an SLA language that provides them the opportunity and incentive to take advantage of any tolerances they may have regarding the scheduling of their workloads. CaaS operational services provide the information necessary for, and carry out the reconfigurations mandated by strategic services. We recognize that it could be the case that there are multiple, yet functionally equivalent ways to express an SLA. Thus, towards that end, we present a service that allows the provably-safe transformation of SLAs from one form to another for the purpose of achieving more efficient colocation. Our CaaS framework could be incorporated into an IaaS offering by providers or it could be implemented as a value added proposition by IaaS resellers. To establish the practicality of such offerings, we present a prototype implementation of our proposed CaaS framework.

Shared memory multiprocessor systems are becoming increasingly important and common. Multiprocessor environments are significantly different from uniproces­sor environments, raising new scheduling issues that need to be considered. A funda­mental scheduling issue arises in situations in which a unit of work may be processed more efficiently on one processor than on any other, due to factors such as the rate at which the required data can be accessed from the given processor. The unit of work is said to have an "affinity" for the given processor, in such a case. The scheduling issue that has to be considered is the trade off between the goals of respecting proces­sor affinities (so as to obtain improved efficiencies in execution) and of dynamically assigning each unit of work to whichever processor happens to be, at the time, least loaded (so as to obtain better load balance and decreased processor idle times). A specific context in which the above scheduling issue arises is that of shared memory multiprocessors with large, per-processor caches or cached main memories. The shared-memory programming paradigm of such machines permits the dynamic scheduling of work. The data required by a unit of work may, however, often reside' mostly in the cache of one particular processor, to which that unit of work thus has affinity. In this thesis, the design of "affinity scheduling" algorithms, in which both affinity and load balancing considerations play major roles in the scheduling policy, is explored. Two new affinity scheduling algorithms are proposed for a context in which the units of work have widely varying execution times. An experimental study of these algorithms finds them to be superior to the previously proposed algorithms 'in this context.'

Dissertação de mestrado em Engenharia Informática
The goal of this dissertation is to explore techniques to improve the efficiency and performance level of scientific applications on computing platforms that are equipped with multiple multi-core devices and at least one many-core device, such as Intel MIC and/or NVidia GPU devices. These platforms are known as heterogeneous servers, which are becoming increasingly popular both in research environments as in our daily gadgets. To fully exploit the performance capabilities of the heterogeneous servers, it is crucial to have an efficient workload distribution among the available devices; however the heterogeneity of the server and the workload irregularity dramatically increases the challenge. Most state of the art schedulers efficiently balance regular workloads among heterogeneous devices, although some lack adequate mechanisms for irregular workloads. Scheduling these type of workloads is particularly complex due to their unpredictability, namely on their execution time. To overcome this issue, this dissertation presents an efficient dynamic adaptive scheduler that efficiently balances irregular workloads among multiple devices in a heterogeneous environment. To validate the scheduling mechanism, the case study used in this thesis is an irregular scientific application that has a set of independent embarrassingly parallel tasks applied to a very large number of input datasets, whose tasks durations have an unpredictable range larger than 1:100. By dynamically adapting the size of the workloads that were distributed among the multiple devices in run-time, the scheduler featured in this dissertation had an occupancy rate of every computing resources over 97% of the application’s run-time while generating an overhead well below 0.001%.
O objetivo desta dissertação é o de explorar técnicas que possam melhorar a eficiência e o nível de performance de aplicações cientificas em plataformas de computação que estão equipadas com vários dispositivos multi-core e pelo menos um dispositivo many-core, como por exemplo um Intel MIC e/ou um GPU da NVidia. Estas plataformas são conhecidas como servidores heterogéneos e estão a se tornar cada vez mais populares, tanto em ambientes de investigação como em nossos gadgets diários. Para explorar completamente as capacidades de desempenho dos servidores heterogéneos, é crucial ter uma distribuição eficiente da carga de trabalho entre os vários dispositivos disponíveis; no entanto a heterogeneidade do servidor e a irregularidade das cargas de trabalho aumentam drasticamente o desafio. A maioria dos escalonadores mais avançados são capazes de equilibrar eficientemente cargas de trabalho regulares entre dispositivos heterogéneos, embora alguns deles não disponham de mecanismos adequados para cargas de trabalho irregulares. O escalonamento desse tipo de cargas de trabalho é particularmente complexo devido à sua imprevisibilidade, nomeadamente ao seu tempo de execução. Para superar este problema, esta dissertação apresenta um escalonador dinâmico e adaptativo que equilibra de forma eficiente cargas de trabalho irregulares entre vários dispositivos de uma plataforma heterogénea. Para validar o escalonador, o caso de estudo utilizado nesta tese é uma aplicação científica irregular que possui um conjunto de tarefas independentes, que são embaraçosamente paralelas, aplicadas a um grande número de conjuntos de dados, cujas tarefas têm durações com um n´nível de imprevisibilidade maior do que 1:100. Ao adaptar dinamicamente o tamanho das cargas de trabalho, que são distribuídas entre os vários dispositivos, em tempo de execução, o escalonador apresentado nesta dissertação apresenta uma taxa de ocupação de cada dispositivo acima de 97 % do tempo total de execução da aplicação e tem um peso que é bem abaixo dos 0,001 %.

In much of classical queueing theory, workloads are assumed to be light-tailed, with job sizes being described using exponential or phase type distributions. However, over the past two decades, studies have shown that several real-world workloads exhibit heavy-tailed characteristics. As a result, there has been a strong interest in studying queues with heavy-tailed workloads. So at this stage, there is a large body of literature on queues with light-tailed workloads, and a large body of literature on queues with heavy-tailed workloads. However, heavy-tailed workloads and light-tailed workloads differ considerably in their behavior, and these two types of workloads are rarely studied jointly.

In this thesis, we design scheduling policies for queueing systems, considering both heavy-tailed as well as light-tailed workloads. The motivation for this line of work is twofold. First, since real world workloads can be heavy-tailed or light-tailed, it is desirable to design schedulers that are robust in their performance to distributional assumptions on the workload. Second, there might be scenarios where a heavy-tailed and a light-tailed workload interact in a queueing system. In such cases, it is desirable to design schedulers that guarantee fairness in resource allocation for both workload types.

In this thesis, we study three models involving the design of scheduling disciplines for both heavy-tailed as well as light-tailed workloads. In Chapters 3 and 4, we design schedulers that guarantee robust performance across heavy-tailed and light-tailed workloads. In Chapter 5, we consider a setting in which a heavy-tailed and a light-tailed workload complete for service. In this setting, we design scheduling policies that guarantee good response time tail performance for both workloads, while also maintaining throughput optimality.