Search

Container based virtualization is an innovative technology that accelerates software development by providing portability and maintainability of applications. Recently, a growing number of workloads such as high performance computing (HPC) and Deep Learning(DL) are deployed in the container based environment. However, GPU resource management issues especially the GPU memory over subscription issue in container-based clusters, which brings substantial performance loss, is still challenging. This paper proposes an adaptive fair-share method to share effectively in container-based virtualization environment as well as an execution rescheduling method to manage the execution order of each container for acquiring maximum performance gain. We also proposed a checkpoint based mechanism especially for DL workload running with TensorFlow, which can efficiently solve the GPU memory over subscription problem. We demonstrate that our approach contributes to overall performance improvement as well as higher resource utilization compared to default and static fair-share methods with homogeneous and heterogeneous workloads. Compared to two other conditions, their results show that the proposed method reduces by 16.37%, 15.61% in average execution time and boosts approximately by 52.46%, 10.3% in average GPU memory utilization, respectively. We also evaluated our checkpoint based mechanism by running multiple CNN workloads with TensorFlow at the same time and the result shows our proposed mechanism can ensure each workload executing safely without out of memory (OOM) error occurs.

As the rapid advancement and diversity in the computing systems, it is demanding to take the most robust scheduling algorithm that guarantee an optimized performance to execute manifold applications on large-scale heterogeneous computing environments. This paper present an adaptive application-aware job scheduling optimization strategy for large-scale high throughput computing in heterogeneous infrastructures. The proposed scheduling optimization method is built on two main concepts. First, it provides application-aware job distribution weights through empirical data in large-scale heterogeneous infrastructures. Here we adopt the concept of weight, which represent the ratio of tasks that will be computed on each resource. The weights can vary in terms of application type and are optimized until the system get in steady status. Secondly, it offers an adaptive control phase that is invoked by the weight adjustment and resource scaling feature. The feedback data from monitoring module is forwarded to the control phase in order to adjust weights and over-provisioning ratio, and result in enhancing overall balance between performances and utilization of system. The experimental evaluation with the four realistic workload patterns demonstrates that, when compared to the core-based scheme which distributes tasks in proportions of each resource's number of cores, the use of our optimization method can achieve 62 % better average throughput, 43 % shorter average queueing time, and 38 % better average utilization of the entire resources in diverse infrastructure environments by harnessing our adaptive module.

This paper presents a MOHA(Many-Task Computing on Hadoop) framework which aims to effectively apply the Many-Task Computing(MTC) technologies originally developed for high-performance processing of many tasks, to the existing Big Data processing platform Hadoop. We present basic concepts, motivation, preliminary results of PoC based on distributed message queue, and future research directions of MOHA. MTC applications may have relatively low I/O requirements per task. However, a very large number of tasks should be efficiently processed with potentially heavy inter-communications based on files. Therefore, MTC applications can show another pattern of data-intensive workloads compared to existing Hadoop applications, typically based on relatively large data block sizes. Through an effective convergence of MTC and Big Data technologies, we can introduce a new MOHA framework which can support the large-scale scientific applications along with the Hadoop ecosystem, which is evolving into a multi-application platform.

Virtualization is an innovative technology that accelerates software development by providing portability and maintainability of applications. However, it often leads underperformance especially caused by overheads from managing virtual machines. To address the limitation of virtual machines, container technology has emerged to deploy and operate distributed applications without launching entire virtual machines. Unfortunately, resources contention issues in container-based clusters, bringing substantial performance loss are still challenging. This paper proposes an adaptive fair-share method to share effectively in container-based virtualization environment. In particular, we focus on enabling GPU sharing between multiple concurrent containers without lack of GPU memory. We demonstrate that our approach contributes to overall performance improvement as well as higher resource utilization compared to default and static fair-share methods with homogeneous and heterogeneous workloads. Compared to two other conditions, their results show that the proposed method reduces by 16.37%, 15.61% in average execution time and boosts approximately by 52.46%, 10.3% in average GPU memory utilization, respectively.

Cloud computing provides on-demand resource provisioning and scalable resources dynamically for the efficient use of computing resources. Scientific applications recently need a very large number of loosely coupled tasks to be handled efficiently. In response, current computing environments often consist of heterogeneous resources such as cloud computing. To effectively use cloud resources, auto-scaling methods that consider diverse metrics such as CPU utilization and costs of resource usage have been studied widely. However it still remains a challenge to automatically and timely allocate resources such that deadline violation and application types are considered. In this paper, we propose auto-scaling methods that consider specific conditions such as application types, task dependency, user-defined deadlines and data transfer times within a hybrid computing infrastructure. Our hybrid computing infrastructure consists of local cluster and cloud resources using HTCaaS. We observe noticeable improvements in performance when our auto-scaling methods for bag-of-tasks and workflow applications is applied.

Nationwide supercomputing infrastructures in Korea consist of geographically distributed supercomputing clusters. We developed High-Throughput Computing as a Service(HTCaaS) based on these distributed national supecomputing clusters to facilitate the ease at which scientists can explore large-scale and complex scientific problems. In this paper, we present our mechanism for dynamically managing computing resources and show its effectiveness through a case study of a real scientific application called drug repositioning. Specifically, we show that the resource utilization, accuracy, reliability, and usability can be improved by applying our resource management mechanism. The mechanism is based on the concepts of waiting time and success rate in order to identify valid computing resources. The results show a reduction in the total job completion time and improvement of the overall system throughput.

In this paper, we present a case study for a drug repositioning simulation based on distributed supercomputing technology that requires highly efficient processing of large-scale computations. Drug repositioning is the application of known drugs and compounds to new indications (i.e., new diseases), and this process requires efficient processing of a large number of docking tasks with relatively short per-task execution times. This mechanism shows the main characteristics of a Many-Task Computing (MTC) application, and as a representative case of MTC applications, we have applied a drug repositioning simulation in our HTCaaS system which can leverage distributed supercomputing infrastructure, and show that efficient task dispatching, dynamic resource allocation and load balancing, reliability, and seamless integration of multiple computing resources are crucial to support these challenging scientific applications.

Cloud computing platform is becoming one of the most desired environments for the emerging applications as well as traditional application domains. In this regards, there have been lots of efforts to develop and redesign the framework to deliver solutions that meet a variety of requirements including resilience, performance, and agility. This paper presents a design and implementation of iSDF, an integrated software-defined computing framework for scientific experiments. The iSDF aims at reinforcing traditional cloud-based scheme by supporting application-aware scheduling that avoids resource starvation and contentions for software-defined computing. To demonstrate the effectiveness of the iSDF, we evaluate it by performing several experiments based on three scenarios. The results show that our framework can achieve a 51% increase on resource utilization and average makespan time a 19% improvement compared to the conditions that do not include the policy. Thus, we proved that the framework can contribute to increase utilization for shared resources and to optimize overall makespan time, eventually affecting overall throughput of jobs.

The science cloud paradigm has been actively developed and investigated, but still requires a suitable model for science cloud system in order to support increasing scientific computation needs with high performance. This paper presents an effective provisioning model of science cloud, particularly for large-scale high throughput computing applications. In this model, we utilize job traces where a statistical method is applied to pick the most influential features to improve application performance. With these features, a system determines where VM is deployed (allocation) and which instance type is proper (provisioning). An adaptive evaluation step which is subsequent to the job execution enables our model to adapt to dynamical computing environments. We show performance achievements by comparing the proposed model with other policies through experiments and expect noticeable improvements on performance as well as reduction of cost from resource consumption through our model.

Traditional High-Throughput Computing (HTC) consists of running many loosely-coupled tasks that are independent but requires a large amount of computing power during significant period of time. However, recent emerging applications requiring millions or even billions of tasks to be processed within a relatively short period of time have expanded the traditional HTC into Many-Task Computing (MTC).In silico drug discovery offers an efficient alternative to reduce the cost of drug development and discovery process. For this purpose, virtual screening is used to select the most promising candidate drugs for in vitro testing from millions of chemical compounds. This process requires a substantial amount of computing resources and high-performance processing of docking simulations, which shows the typical characteristics of MTC applications. As the number of users performing this virtual screening process increases with limited available computing resources, it becomes crucial to devise an effective scheduling policy that can ensure a certain degree of fairness and user satisfaction.In this paper, we present a comparative analysis of scheduling mechanisms for the virtual screening workflow where multiple users in the system are sharing a common service infrastructure. To effectively support these multiple users, the underlying system should be able to consider fairness, user response time and overall system throughput. We have implemented two different scheduling algorithms which can address fairness and user response time respectively in a common middleware stack called HTCaaS which is a pilot-job based multi-level scheduling system running on top of a dedicated production-level cluster. Through our comparative analysis of two different scheduling mechanisms targeting different metrics on top of a single H/W and S/W system, we can give an insight to the research community on the design and implementation of a scheduling mechanism that can trade-off user fairness and overall system performance which is crucial to support challenging MTC applications.

Summary In this paper, we conduct a comparative study of relatively simple yet effective scheduling policies for many-task applications where multiple users with varying numbers of tasks are actively sharing a common system infrastructure. We have implemented three different scheduling mechanisms that can address fairness and user response time respectively in a common middleware stack called HTCaaS, which is a pilot-job–based multi-level scheduling system running on top of production-level clusters. As a representative case of our many-task applications, we have leveraged the virtual screening application, which is a computational technique used in drug discovery process to select the most promising candidate drugs for in vitro testing from millions of chemical compounds, which typically requires a substantial amount of computing resources and efficient processing of docking simulations. Our comparative experimental results of different scheduling policies show how we can effectively support multiple users in a shared resource environment by balancing between user satisfaction and overall system performance, provide guidelines to improve system utilization, and address additional technical challenges to support various many-task applications.

We discuss the aspect of the MC simulation performance on the agent-based multi-level job submission system, HTCaaS, which offers efficiencies by dispatching pilot jobs across multiple and heterogeneous resources. We show that the MC tools such as MadGraph5 and PYTHIA can be integrated well in the HTCaaS system without the modification of the codes. We also show that the idle cycles of the distributed computing environment can be best utilize to greatly reduce the run times of the MC event generations under HTCaaS system.

Recent cloud computing enables numerous scientists to earn advantages by serving on-demand and elastic resources whenever they desire computing resources. This science cloud paradigm has been actively developed and investigated to satisfy requirements of the scientists such as performance, feasibility and so on. However, effective allocation and provisioning virtual machines on clouds are still considered as a challenging issue in scientists using high throughput computing, since it determines whether they can earn benefits from economy of scale in clouds or not. Moreover, allocating the "right" provisioned cloud resources on an optimal data center is very important as performance can vary widely depending on where and under what circumstances it actually runs. In these reasons, it is required that an appropriate and suitable model for science cloud to support increasing scientists and computations.In this paper, we present an allocation and provisioning model of science cloud, especially for high throughput computing applications. In this model, we utilize job traces where statistical method is applied to pick the most influential features for improving application performance. With the feature, the system determines where VM is deployed (allocation) and which instance type is proper (provisioning). An adaptive evaluation step which is subsequent to the job execution enables our model to adapt to dynamical computing environments. We show performance achievements as comparing the proposed model with other policies through experiments. Finally, we expect that improvement on performance as well as reduction of cost from resource consumption through our model.

We present the HTCaaS (High-Throughput Computing as a Service) system which aims to provide researchers with ease of exploring large-scale and complex HTC problems by leveraging Supercomputers, Grids, and Cloud. HTCaaS can hide heterogeneity and complexity of harnessing different types of computing infrastructures from users, and efficiently submit a large number of jobs at once by effectively managing and exploiting of all available computing resources. Our system has been effectively integrated with national Supercomputers in Korea, international computational Grids, and Amazon EC2 resulting in combining a vast amount of computing resources to support most challenging scientific problems.

As computing technologies develop, various studies for e-Science have been actively developed for many years. The advent of Cloud computing, for example, enables scientists to expand their research environments over supercomputers to on-demand and scalable resources. However, several performance drawbacks on the cloud computing can cause considerable obstacles in scientific domains, despite its strong merits. A hybrid infrastructure which consists of the existing and cloud architecture thus can produce a synergy effect for utilizing resources efficiently. In this paper, we proposed hybrid infrastructure-supported a scientific workflow environment for Aerodynamics design and demonstrated its superiority. Hybrid infrastructure covers grid and private cloud computing in this paper. Especially, we focused on improving performance by supporting hybrid infrastructure and efficient usages of physical resources. Since it offers diverse types of computing infrastructures including cloud computing, it can serve elastic resources regardless of the number of tasks for experiments or limitations of space and can reduce costs of time as well as budget during simulations.

The rapid advent of computing technology enables scientists to expand their research infrastructure over supercomputer-on-demand using a resource leasing service on pay-per-use basis. This infrastructure service is called as Science cloud which provides uniform user interface to scientific experiments over large-scale heterogeneous resources. In spite of the strength of conveniences, it is difficult to manage the experiments to guarantee optimal performance of jobs since the execution environment is based on virtualization technology which has additional performance overheads. This paper proposes a cloud resource provisioning model using statistical analysis of job profiles for computational science. In this model, we utilize job profiles which are generated from application executions and identify features of application by applying statistical analysis. The analysis is performed using PCA and the most effective factors are inferred from profiles. The effective factors are used for picking a job profile being referred and then VM type is determined using the most effective job's property. An application is executed on VM of the chosen cluster and its performance result is incorporated into job profile set with purpose of evaluating profile's credit. Our simulation results show this model enables scientists to take advantages of cloud computing without performance degradation.

The advent of cloud computing makes scientists to extend their research environments over supercomputers to on-demand and dynamically scalable resources. Science cloud becomes a trend in various scientific domains these days. However, it is difficult to provide optimal job execution environment rapidly and dynamically depending on user's demands. Therefore, it is very important to predict user's requirements and to prepare execution environment in advance. In addition, it needs scheduling mechanisms for virtual machines to provide some level of guaranteed performance of a user application. In this paper, we propose a cloud resource provisioning model using statistical analysis of job history. In this model, we use job history which is generated from many application executions and identifies characteristics of an application by applying statistical analysis. We utilize a statistical technique, PCA (Principal Component Analysis), to analyze execution history of applications and to extract the factors which contribute much to execution time. The effective factors are used for selecting reference job profile and then VM is deployed on the selected node based on the reference profile. An application is executed on chosen nodes and its performance result is incorporated into job history with the purpose of evaluating profile's credit. As a result, this model can provide efficient management of cloud resource for a service provider and reduce management overhead on cloud.

As one of IaaS (Infrastructure-as-a-Service), it is beneficial to arrange virtual machines dynamically to applications based on resource provisioning mechanism. However, it is challenging to apply scheduling scheme to utilize resources efficiently when many tasks require a lot of resources at the same time. Especially, scientific applications, which require large-scale computing resource for long term execution period, need more dynamic scheduling to occupy resource appropriately. Resource virtualization, manipulating several virtual machines (VMs) over physical resource gives good opportunities to enhance the performance of applications and resources. In this paper, we conducted experiments on adaptable scheduling schemes with scientific applications which need a lot of resources for long execution time period on cloud computing environment by distributing jobs to VMs. We provide cloud computing infrastructure by using OpenNebula virtual infrastructure engine and Haziea scheduler. Moreover, we verified the improvement of the execution time of applications and whole resource usage by scheduling VMs according to their priorities.