Performance-Cost Optimization of Moldable Scientific Workflows

In a shared virtual computing environment, dynamic load changes as well as different quality requirements of applications in their lifetime give rise to dynamic and various capacity demands, which results in lower resource utilization and application quality using the existing static resource allocation. Furthermore, the total required capacities of all the hosted applications in current enterprise data centers, for example, Google, may surpass the capacities of the platform. In this paper, we argue that the existing techniques by turning on or off servers with the help of virtual machine (VM) migration is not enough. Instead, finding an optimized dynamic resource allocation method to solve the problem of on-demand resource provision for VMs is the key to improve the efficiency of data centers. However, the existing dynamic resource allocation methods only focus on either the local optimization within a server or central global optimization, limiting the efficiency of data centers. We propose a two-tiered on-demand resource allocation mechanism consisting of the local and global resource allocation with feedback to provide on-demand capacities to the concurrent applications. We model the on-demand resource allocation using optimization theory. Based on the proposed dynamic resource allocation mechanism and model, we propose a set of on-demand resource allocation algorithms. Our algorithms preferentially ensure performance of critical applications named by the data center manager when resource competition arises according to the time-varying capacity demands and the quality of applications. Using Rainbow, a Xen-based prototype we implemented, we evaluate the VM-based shared platform as well as the two-tiered on-demand resource allocation mechanism and algorithms. The experimental results show that Rainbow without dynamic resource allocation (Rainbow-NDA) provides 26 to 324 percent improvements in the application performance, as well as 26 percent higher average CPU utilization than traditional service computing framework, in which applications use exclusive servers. The two-tiered on-demand resource allocation further improves performance by 9 to 16 percent for those critical applications, 75 percent of the maximum performance improvement, introducing up to 5 percent performance degradations to others, with 1 to 5 percent improvements in the resource utilization in comparison with Rainbow-NDA.

Scientific workflows are one of the well-established pillars of modern large-scale computational science. More recently, scientists have started to leverage machine learning (ML) capabilities in their workflows, leading to a new category of scientific workflows, denoted as scientific ML workflows. ML is not only about training and inference, modern ML workflows also involve complex data processing steps before the training can start, which are not often accounted for in most performance studies. In this work, we consider scientific ML workflows, from data pre-processing to training, inference, and model evaluation. We aim to explore (i) how scientific ML workflows differ from more traditional scientific workflows and; (ii) how we can characterize ML workflows both in terms of execution time and data movements when executing on a target cloud platform. We select three representative workflows, ranging from image classification to natural language processing and image segmentation, which have been executed using the academic cloud platform, Chameleon. We build four realistic deployment scenarios for each workflow, which stress data movements during workflow executions. Then, we compare the performance observed when utilizing these different configurations and study how different settings impact overall workflows performance and efficiency when running on cloud infrastructures. Finally, we summarize our findings and discuss performance impacts when augmenting scientific workflows with ML techniques and how traditional workflow management systems can improve their support for such workflows.

In this work, optimization methodologies of the kinematics of flapping wings in forward and hover flights are considered. Particularly, local and global optimization algorithms are combined with the unsteady vortex lattice method (UVLM) to determine the most efficient kinematics. In the first problem, the kinematic optimization of a three-dimensional flapping wing in forward flight with active shape morphing is aimed at maximizing the propulsive efficiency under lift and thrust constraints. Results show that due to the quasi-concavity of the objective function associated with the flapping kinematics employed in forward flight, local gradient-based optimizers perform very well in terms of accuracy and computational cost. In the second problem, the objective is to identify the optimized kinematics of a twodimensional hovering wing that minimize the aerodynamic power under lift constraint. The results show that using a hybrid optimization algorithm (combination of local and global schemes) accelerates the convergence to optimal points and avoids being trapped at local minimum points. While the efficiency of using a local optimizer, a global optimizer, or a hybrid of the two depends on the nature of the problem and associated design space, it is determined that hybrid optimization is best suited for determining local minima.

Workflow technology has been adopted in scientific domains to orchestrate and automate scientific processes in order to facilitate experimentation. Such scientific workflows often involve large data sets and intensive computation that necessitate the use of the Grid, which offers supercomputing power through shared distributed resources. To execute a scientific workflow in the Grid, tasks within the workflow that represent steps in the scientific process are assigned to Grid resources for execution. To ensure efficient execution of the workflow, Grid workflow scheduling is required to manage the allocation of Grid resources. Although many Grid workflow scheduling techniques exist, they are mainly designed for the execution of a single workflow. This is not the case with parameter sweep workflows which focus on parametric study and parameter optimisation. A parameter sweep workflow is executed numerous times with different input parameters in order to determine the effect of each parameter combination on the experiment. While executing multiple instances of a parameter sweep workflow in parallel can reduce the time required for the overall execution, this execution introduces new challenges to Grid workflow scheduling. Not only is a scheduling algorithm that is able to manage multiple workflow instances required, but this algorithm also needs the ability to schedule tasks across multiple workflow instances judiciously, as tasks may require the same set of Grid resources. Without appropriate resource allocation, resource competition problem could arise. In the thesis, we propose a new Grid workflow scheduling technique for parameter sweep workflow called the Besom scheduling algorithm. The scheduling decision of our algorithm is based on the resource dependencies of tasks in the workflow, as well as conventional Grid resource-performance metrics. In addition, the proposed technique is extended to handle loop structures in scientific workflows without using existing loop-unrolling techniques. We evaluate the Besom algorithm under a variety of conditions. A comparison between the simulation results of the Besom algorithm and of the three existing Grid workflow scheduling algorithms shows that the Besom algorithm is able to perform better than the existing algorithms for workflows that have complex structures and that involve overlapping resource dependencies of tasks. The Besom scheduling algorithm advances the ability to schedule parallel execution of parameter sweep workflows in the Grid. The outcomes of this thesis justify continuing research in this area to increase our understanding of scheduling multiple Grid workflow instances and to provide support to those involved in parametric study and scientific workflow management.

Workflow is becoming a more and more important tool for business operations, scientific research and engineering. Cloud computing provides an elastic, on-demand and high cost-efficient resource allocation model for workflow executions. During workflow execution, the load will change from time to time and therefore, it becomes an interesting topic to optimize resource utilization of workflows in the cloud computing environment. In this paper, a workflow framework is proposed that can adaptively use cloud resources. In detail, after users specify the desired goal to achieve, the proposed workflow framework then monitors the workflow execution, and utilizes different pricing models to acquire cloud resources according to the change of workflow load. In this way, the cost of workflow execution is reduced. .

In the allocation of resources to activities, each activity is described by a concave return function and the sum of the returns is maximized. There are linear constraints on the available quantity of each essential resource. Existing methods for the incremental generation of almost-optimal allocations and for the evaluation of such allocations are extended to include allocations involving "noise" constraints in addition to the linear constraints on the available quantity of each resource. A noise constraint, which is defined in the paper, may be expressed by any relationship, not necessarily linear. An example is given in which the noise constraints take the form of constraints on additional resources.

In many clusters and data centers, application frameworks are used that offer programming models such as Dryad and MapReduce, and jobs submitted to the clusters or data centers may be targeted at specific instances of these frameworks, for example because of the presence of certain data. An important question that then arises is how to allocate resources to framework instances that may have highly fluctuating workloads over their lifetimes. Static resource allocation, a traditional approach for scheduling jobs, may result in inefficient resource allocation because of poor resource utilization during off-peak hours. We address this issue with a strategy for the dynamic deployment of a component-based framework by extending a resource manager responsible for scheduling jobs in multi-cluster environments. This extension allows scheduling multiple concurrent instances of the framework as long-running utility jobs that share computational resources of the cluster. In order to accommodate the fluctuating resource demands of frameworks, we consider two provisioning policies for dynamic resource allocation: OnDemand and Proactive provisioning. We evaluate the effectiveness of both policies by comparing them with static resource allocation on the das4 multi-cluster system. Our results show that dynamic resource allocation gives at least 30 % improvement over the static resource allocation with respect to both the utilization of the resources and the reject rate of the applications within the framework.

Executing scientific workflows in high-performance cloud computing platforms requires the use of scheduling algorithms that allow workflows execution as fast as possible, while minimizing the monetary cost of such executions. In this work we study how the use of supporting information can offer guidance to scheduling algorithms, helping them to devise more efficient execution plans in terms of the total execution time (makespan) and the total monetary cost. Using two large-scale scientific workflows, our experiments showed that simple modifications on a classical scheduling algorithm (HEFT), in conjunction with the appropriate supporting information, could reduce the monetary cost of an execution in up to 59% and reduce the makespan in up to 8.6%.

*† ‡ § An efficient and high-fidelity design approach is proposed by combining global and local optimization methods for wing planform and surface design. For enhanced design results, aerodynamic shape optimization process is carried out via 2-stage with different optimization strategy. In the first stage, global optimization techniques are applied to planform design with a few geometric design variables. In the second stage, local optimization techniques are used for wing surface design with a lot of design variables to maintain a sufficient design space with high DOF (Degree of Freedom) geometric change. For global optimization, meta-modeling techniques such as RS (Response Surface) and Kriging methods are used in conjunction with Genetic Algorithm (GA). For local optimization, a discrete adjoint variable method is used. By the successive combination of global and local optimization techniques, drag minimization is performed for a multi-body aircraft configuration while maintaining the baseline lift and the wing weight at the same time. Through the design process, performances of the test models are remarkably improved in comparison with the single stage design approach. The capability of proposed design framework including wing planform design variables can be evaluated by the drag decomposition method which can provide improvement of induced drag and wave drag, respectively.

Many computation-intensive scientific applications feature complex workflows of distributed computing modules with intricate execution dependencies. Such scientific workflows must be mapped and executed in shared environments to support distributed scientific collaborations. We formulate workflow mapping as an optimization problem for latency minimization, whose difficulty essentially arises from the topological matching nature in the spatial domain, which is further compounded by the resource sharing complicacy in the temporal dimension. We conduct a rigorous analysis of the resource sharing dynamics in workflow executions, which constitutes the base for a workflow mapping algorithm to minimize the end-to-end delay. The correctness of the dynamics analysis is verified in comparison with an approximate solution, a dynamic system simulation program, and a real network deployment, and the performance superiority of the proposed mapping solution is illustrated by extensive comparisons with existing methods using both simulations and experiments.

The scale of scientific applications becomes increasingly large not only in computation, but also in data. Many of these applications also concern inter-related tasks with data dependencies, hence, they are scientific workflows. The efficient coordination of executing/running scientific workflows is of great practical importance. The core of such coordination is scheduling and resource allocation. In this paper, we present three scheduling heuristics for running large-scale, data-intensive scientific workflows in clouds. In particular, the three heuristic algorithms are designed to leverage slot queue threshold, data locality and data prefetching, respectively. We also demonstrate how these heuristics can be collectively used to tackle different issues in running "data-intensive" workflows in clouds although each of these heuristics can be used independently. The practicality of our algorithms has been realized by actually implementing and incorporating them into our workflow execution system (DEWE). Using Montage, an astronomical image mosaic engine, as an example workflow, and Amazon EC2 as the cloud environment, we evaluate the performance of our heuristics in terms primarily of completion time (make span). We also scrutinize workflow execution showing different execution phases to identify their impact on performance. Our algorithms scale well and reduce make span by up to 27%.

Scientific workflow technology has become a de facto standard in e-Science. It provides a powerful unifying platform that allows scientists to build arbitrarily complicated applications by combining predefined components. With a wide range of components, with the functions ranging from acquiring data inputs from sensors, querying databases, performing data-mining, performing execution, through to visualising the results, this paradigm allows scientists to automate, manage and execute various steps in scientific research. Although powerful, scientific workflow systems still require some programming knowledge, which might not be optimal for many scientists. Recently, scientific workflows have been integrated into the Web, creating workflow-based science gateways. These gateways allow scientists to launch large-scale computational experiments without concern for the underlying complexity of the infrastructure. This approach delivers scientific workflows to more scientists and is becoming increasingly popular. To date, much workflow-based science gateway development effort has been invested in creating user-friendly and secure interfaces to setup parameters, execute workflows and collect results. However, interaction between users and executing workflows has not received a lot of attention. Scientific workflows have typically been executed in batch mode. This model does not work with human-in-the-loop workflows, which require continuous interaction between users and executing workflows. A fundamental contribution of this thesis is to show that gateways can support complex human- workflow interactions, facilitating monitoring and steering of the workflow execution as it progresses. This is important because scientific workflows increasingly require user decisions and interactions at various steps; human decision-making is an important component of real-world computational science. In order to achieve this goal, the thesis presents the design, implementation and demonstration of WorkWays, an interactive workflow-based science gateway for supporting development and execution of human-in-the-loop scientific workflows. WorkWays consists of two distinct frameworks with well defined APIs and bindings: a framework that enables human- workflow interaction and a framework for workflow development and execution.

Previous work was done to compare the static resource allocation (StRA) scheme and scalable resource allocation (ScRA) scheme within an integrated GSM and GPRS network (see Tang, S. and Tafazolli, R., Vehicular Technology Conference - Spring, 2003). In order to maintain the QoS for both, the ScRA scheme assumed that the load history database was a credible estimate of the present state of traffic intensity in the cell area. In cases when there is a sudden change in traffic conditions, the ScRA scheme may not be able to perform well. Hence, in order to cope with sudden changes of traffic in the network, we propose to incorporate prediction schemes into the ScRA algorithm. For comparison purposes, we present and evaluate three prediction strategies, namely linear regression prediction, exponential regression prediction, and current value prediction, in six various traffic conditions. In addition, resource utilisation gains between the StRA scheme and the proposed ScRA schemes are also studied. The findings show that ScRA exponential regression prediction scheme can cope best in all the tested traffic scenarios and experiences an average of 0.5 percent of resource utilisation gain over the StRA scheme.

Software reliability growth model with debugging efforts

In this paper, we propose cumulative distribution function (CDF)-based multiuser scheduling (CMS) for downlink non-orthogonal multiple access (NOMA) to accommodate a massive number of IoT devices over limited radio resources and statistically analyze the ergodic capacity of devices. With the proposed CMS algorithm, multiple devices having relatively better channel gain to exploit multiuser diversity are selected while meeting QoS requirements by controlling the amount of resources. We first provide the user selection probability in terms of the weight of each device, which determines the amount of time resources. Based on some selected results, it is shown that equal weights ensure fair resource allocation. Further, for fair resource allocation scenario, the distribution of signal-to-interference-plus-noise ratio (SINR) when a specific device is selected is derived. With this derived statistical result of SINR, we analyze the ergodic capacity of the specific device. Some selected results show that the proposed CMS algorithm allocates the time resource more fairly than proportional fair scheduling. Further, the average throughput and amount of resources can be controlled by adjusting the weights of each device, so different QoS requirements can be guaranteed.