Priority-driven spatial resource sharing scheduling for embedded graphics processing units

Many visual tasks in modern personal devices such smartphones resort heavily to graphics processing units (GPUs) for their fluent user experiences. Because most GPUs for embedded systems are non-preemptive by nature, it is important to schedule GPU resources efficiently across multiple GPU tasks. We present a novel spatial resource sharing (SRS) technique for GPU tasks, called a budget-reservation spatial resource sharing (BR-SRS) scheduling, which limits the number of GPU processing cores for a job based on the priority of the job. Such a priority-driven resource assignment can prevent a high-priority foreground GPU task from being delayed by background GPU tasks. The BR-SRS scheduler is invoked only twice at the arrival and completion of jobs, and thus, the scheduling overhead is minimized as well. We evaluated the performance of our scheduling scheme in an Android-based smartphone, and found that the proposed technique significantly improved the performance of high-priority tasks in comparison to the previous temporal budget-based multi-task scheduling.     

Online algorithms for advance resource reservations

We consider the problem of providing QoS guarantees to Grid users through advance reservation of resources. Advance reservation mechanisms provide the ability to allocate resources to users based on agreed-upon QoS requirements and increase the predictability of a Grid system, yet incorporating such mechanisms into current Grid environments has proven to be a challenging task due to the resulting resource fragmentation. We use concepts from computational geometry to present a framework for tackling the resource fragmentation, and for formulating a suite of scheduling strategies. We also develop efficient implementations of the scheduling algorithms that scale to large Grids. We conduct a comprehensive performance evaluation study using simulation, and we present numerical results to demonstrate that our strategies perform well across several metrics that reflect both user- and system-specific goals. Our main contribution is a timely, practical, and efficient solution to the problem of scheduling resources in emerging on-demand computing environments.     

不规则任务在图形处理器集群上的调度策略

针对大量的资源需求少且并行度高的不规则任务集合,利用图形处理器（GPU）来加速处理是目前的主流。然而现有的不规则任务调度策略要么采用独占GPU的方式,要么使用传统的优化方法将任务映射到GPU设备上。前者导致GPU资源的闲置,后者不能最大限度利用GPU计算资源。在分析了现存问题的基础上,采用多背包优化思想,使更多的不规则任务以最佳的方式共享GPU设备。首先,针对GPU集群的特点,给出了由调度器、执行器组成的分布式GPU作业调度框架;然后,以GPU显存为代价,设计了一种基于GPU计算资源的扩展贪心调度（EGS）算法,该算法将尽可能多的不规则任务调度到多个可用的GPU上,以最大限度地利用GPU计算资源,并解决了GPU资源的闲置问题;最后,使用实际基准程序随机生成目标任务集来验证所提调度策略的有效性。实验结果表明,与传统的贪心算法、最早完成时间（MCT）算法和Min-min算法相比,当任务数量等于1 000时,EGS算法的执行时长分别平均降低至原来的58%、64%和80%,并且能有效提升GPU资源利用率。     

Double auction-inspired meta-scheduling of parallel applications on global grids

Meta-schedulers map jobs to computational resources that are part of a Grid, such as clusters, that in turn have their own local job schedulers. Existing Grid meta-schedulers either target system-centric metrics, such as utilisation and throughput, or prioritise jobs based on utility metrics provided by the users. The system-centric approach gives less importance to users’ individual utility, while the user-centric approach may have adverse effects such as poor system performance and unfair treatment of users. Therefore, this paper proposes a novel meta-scheduler, based on the well-known double auction mechanism that aims to satisfy users’ service requirements as well as ensuring balanced utilisation of resources across a Grid. We have designed valuation metrics that commodify both the complex resource requirements of users and the capabilities of available computational resources. Through simulation using real traces, we compare our scheduling mechanism with other common mechanisms widely used by both existing market-based and traditional meta-schedulers. The results show that our meta-scheduling mechanism not only satisfies up to 15% more user requirements than others, but also improves system utilisation through load balancing.     

A probabilistic and adaptive scheduling algorithm using system-generated predictions for inter-grid resource sharing

Rapid advancement and more readily availability of Grid technologies have encouraged many businesses and researchers to establish Virtual Organizations (VO) and make use of their available desktop resources to solve computing intensive problems. These VOs, however, work as disjointed and independent communities with no resource sharing between them. We, in previous work, have proposed a fully decentralized and reconfigurable Inter-Grid framework for resource sharing among such distributed and autonomous Grid systems (Rao et al. in ICCSA, [2006]). The specific problem that underlies in such a collaborating Grids system is scheduling of resources as there is very little knowledge about availability of the resources due to the distributed and autonomous nature of the underlying Grid entities. In this paper, we propose a probabilistic and adaptive scheduling algorithm using system-generated predictions for Inter-Grid resource sharing keeping collaborating Grid systems autonomous and independent. We first use system-generated job runtime estimates without actually submitting jobs to the target Grid system. Then this job execution estimate is used to predict the job scheduling feasibility on the target system. Furthermore, our proposed algorithm adapted itself to the actual resource behavior and performance. Simulation results are presented to discuss the correctness and accuracy of our proposed algorithm.

计算网格的资源分发和发现机制

1 引言计算网格的资源管理系统是为实现计算网格系统资源共享所应提供的最主要的服务之一。计算网格资源管理系统的基本功能是接受来自计算网格范围内的机器的资源请求,并且把特定的资源分配给资源请求者,并且合理地调度相应的资源,使请求资源的作业得以运行。资源分发、资源发现和资源的调度构成了计算网格资源管理系统的最主要的内容。资源分发和资源发现提供方法,通过该方法,在计算网格内部的机器能够形成一个可用的资源和其状态的一个视图。资源     

Scheduling Task Parallel Applications for Rapid Turnaround on Enterprise Desktop Grids

Desktop Grids are popular platforms for high throughput applications, but due to their inherent resource volatility it is difficult to exploit them for applications that require rapid turnaround. Efficient desktop Grid execution of short-lived applications is an attractive proposition and we claim that it is achievable via intelligent resource selection. We propose three general techniques for resource selection: resource prioritization, resource exclusion, and task duplication. We use these techniques to instantiate several scheduling heuristics. We evaluate these heuristics through trace-driven simulations of four representative desktop Grid configurations. We find that ranking desktop resources according to their clock rates, without taking into account their availability history, is surprisingly effective in practice. Our main result is that a heuristic that uses the appropriate combination of resource prioritization, resource exclusion, and task replication can achieve performance within a factor of 1.7 of optimal in practice.     

Java Bytecode Transformations for Efficient, Portable CPU Accounting

Resource management is essential to build reliable middleware and to host potentially untrusted software components. Resource accounting allows to study and optimize program performance and to charge users for the resource consumption of their deployed components, while resource control can limit the resource consumption of components in order to prevent denial-of-service attacks. In the approach presented here, program transformations enable resource management in Java-based environments, even though the underlying runtime system may not expose information concerning the resource consumption of applications. We present a fully portable program transformation scheme to enhance standard Java runtime systems with mechanisms for CPU management. We implemented several optimizations in order to reduce the overhead of our CPU accounting scheme. Detailed performance measurements quantify this overhead and show the impact of various optimizations.     

Reducing Communication Overhead in Multi-GPU Hybrid Solver for 2D Laplace’s Equation

The possibility of porting algorithms to graphics processing units (GPUs) raises significant interest among researchers. The natural next step is to employ multiple GPUs, but communication overhead may limit further performance improvement. In this paper, we investigate techniques reducing overhead on hybrid CPU–GPU platforms, including careful data layout and usage of GPU memory spaces, and use of non-blocking communication. In addition, we propose an accurate automatic load balancing technique for heterogeneous environments. We validate our approach on a hybrid Jacobi solver for 2D Laplace’s Equation. Experiments carried out using various graphics hardware and types of connectivity have confirmed that the proposed data layout allows our fastest CUDA kernels to reach the analytical limit for memory bandwidth (up to 106 GB/s on NVidia GTX 480), and that the non-blocking communication significantly reduces overhead, allowing for almost linear speed-up, even when communication is carried out over relatively slow networks.     

Grid Economics: A Selective Discussion of Two Research Problems

The last 5 years have seen considerable discussion of various types of Grids—compute Grids, storage Grids, and data Grids. Using the checklist given in Foster (, 2002) to define a Grid, two important problems that arise in the context of resource sharing in Grid computing environments are discussed. First, the well documented problem in compute Grid environments that arises from the inability of consumers to accurately estimate their resource requirements is presented. This results in incorrect scheduling of requests for Grid resources and social welfare loss. To address this problem, two research proposals are briefly described. The first approach argues for the design of decision support tools to help users with resource estimation while the second approach studies the design of resource allocation mechanisms that can work with stochastic specifications of resource requirements. This is in contrast to the traditional point estimates of resource required by extant mechanisms. Next, resource provisioning and pricing problems that arise in data storage and retrieval Grids are described. These Grids differ fundamentally from compute Grids but share some economic characteristics with P2P file sharing networks. Drawing on this connection, pricing mechanisms and resource provisioning research is briefly discussed.     

Autonomic Clouds on the Grid

Computational clouds constructed on top of existing Grid infrastructure have the capability to provide different entities with customized execution environments and private scheduling overlays. By designing these clouds to be autonomically self-provisioned and adaptable to changing user demands, user-transparent resource flexibility can be achieved without substantially affecting average job sojourn time. In addition, the overlay environment and physical Grid sites represent disjoint administrative and policy domains, permitting cloud systems to be deployed non-disruptively on an existing production Grid. Private overlay clouds administered by, and dedicated to the exclusive use of, individual Virtual Organizations are termed Virtual Organization Clusters. A prototype autonomic cloud adaptation mechanism for Virtual Organization Clusters demonstrates the feasibility of overlay scheduling in dynamically changing environments. Commodity Grid resources are autonomically leased in response to changing private scheduler loads, resulting in the creation of virtual private compute nodes. These nodes join a decentralized private overlay network system called IPOP (IP Over P2P), enabling the scheduling and execution of end user jobs in the private environment. Negligible overhead results from the addition of the overlay, although the use of virtualization technologies at the compute nodes adds modest service time overhead (under 10%) to computationally-bound Grid jobs. By leasing additional Grid resources, a substantial decrease (over 90%) in average job queuing time occurs, offsetting the service time overhead.     

QoS and preemption aware scheduling in federated and virtualized Grid computing environments

Resource provisioning is one of the challenges in federated Grid environments. In these environments each Grid serves requests from external users along with local users. Recently, this resource provisioning is performed in the form of Virtual Machines (VMs). The problem arises when there are insufficient resources for local users to be served. The problem gets complicated further when external requests have different QoS requirements. Serving local users could be solved by preempting VMs from external users which impose overheads on the system. Therefore, the question is how the number of VM preemptions in a Grid can be minimized. Additionally, how we can decrease the likelihood of preemption for requests with more QoS requirements. We propose a scheduling policy in InterGrid, as a federated Grid, which reduces the number of VM preemptions and dispatches external requests in a way that fewer requests with QoS constraints get affected by preemption. Extensive simulation results indicate that the number of VM preemptions is decreased at least by 60%, particularly, for requests with more QoS requirements.     

Concurrent warp execution: improving performance of GPU-likely SIMD architecture by increasing resource utilization

Hardware parallelism should be exploited to improve the performance of computing systems. Single instruction multiple data (SIMD) architecture has been widely used to maximize the throughput of computing systems by exploiting hardware parallelism. Unfortunately, branch divergence due to branch instructions causes underutilization of computational resources, resulting in performance degradation of SIMD architecture. Graphics processing unit (GPU) is a representative parallel architecture based on SIMD architecture. In recent computing systems, GPUs can process general-purpose applications as well as graphics applications with the help of convenient APIs. However, contrary to graphics applications, general-purpose applications include many branch instructions, resulting in serious performance degradation of GPU due to branch divergence. In this paper, we propose concurrent warp execution (CWE) technique to reduce the performance degradation of GPU in executing general-purpose applications by increasing resource utilization. The proposed CWE enables selecting co-warps to activate more threads in the warp, leading to concurrent execution of combined warps. According to our simulation results, the proposed architecture provides a significant performance improvement (5.85 % over PDOM, 91 % over DWF) with little hardware overhead.     

Kubernetes异构资源细粒度调度策略的设计与实现

在异构资源环境中高效利用计算资源是提升任务效率和集群利用率的关键。Kuberentes作为容器编排领域的首选方案,在异构资源调度场景下调度器缺少GPU细粒度信息无法满足用户自定义需求,并且CPU/GPU节点混合部署下调度器无法感知异构资源从而导致资源竞争。综合考虑异构资源在节点上的分布及其硬件状态,提出一种基于Kubernetes的CPU/GPU异构资源细粒度调度策略。利用设备插件机制收集每个节点上GPU的详细信息,并将GPU资源指标提交给调度算法。在原有CPU和内存过滤算法的基础上,增加自定义GPU信息的过滤,从而筛选出符合用户细粒度需求的节点。针对CPU/GPU节点混合部署的情况,改进调度器的打分算法,动态感知应用类型,对CPU和GPU应用分别采用负载均衡算法和最小最合适算法,保证异构资源调度策略对不同类型应用的正确调度,并且在CPU资源不足的情况下充分利用GPU节点的碎片资源。通过对GPU细粒度调度和CPU/GPU节点混合部署情况下的调度效果进行实验验证,结果表明该策略能够有效进行GPU调度并且避免资源竞争。     

Prediction of Resource Availability in Fine-Grained Cycle Sharing Systems Empirical Evaluation

Fine-Grained Cycle Sharing (FGCS) systems aim at utilizing the large amount of computational resources available on the Internet. In FGCS, host computers allow guest jobs to utilize the CPU cycles if the jobs do not significantly impact the local users. Such resources are generally provided voluntarily and their availability fluctuates highly. Guest jobs may fail unexpectedly, as resources become unavailable. To improve this situation, we consider methods to predict resource availability. This paper presents empirical studies on resource availability in FGCS systems and a prediction method. From studies on resource contention among guest jobs and local users, we derive a multi-state availability model. The model enables us to detect resource unavailability in a non-intrusive way. We analyzed the traces collected from a production FGCS system for 3 months. The results suggest the feasibility of predicting resource availability, and motivate our method of applying semi-Markov Process models for the prediction. We describe the prediction framework and its implementation in a production FGCS system, named iShare. Through the experiments on an iShare testbed, we demonstrate that the prediction achieves an accuracy of 86% on average and outperforms linear time series models, while the computational cost is negligible. Our experimental results also show that the prediction is robust in the presence of irregular resource availability. We tested the effectiveness of the prediction in a proactive scheduler. Initial results show that applying availability prediction to job scheduling reduces the number of jobs failed due to resource unavailability. This work was supported, in part, by the National Science Foundation under Grants No. 0103582-EIA, 0429535-CCF, and 0650016-CNS. We thank Ruben Torres for his help with the reference prediction algorithms used in our experiments.     

Enhanced GridSim architecture with load balancing

Grid is a network of computational resources that may potentially span many continents. Maximization of the resource utilization hinges on the implementation of an efficient load balancing scheme, which provides (i) minimization of idle time, (ii) minimization of overloading, and (iii) minimization of control overhead. In this paper, we propose a dynamic and distributed load balancing scheme for grid networks. The distributed nature of the proposed scheme not only reduces the communication overhead of grid resources but also cuts down the idle time of the resources during the process of load balancing. We apply the proposed load balancing approach on Enhanced GridSim in order to gauge the effectiveness in terms of communication overhead and response time reduction. We show that significant savings are delivered by the proposed technique compared to other approaches such as centralized load balancing and no load balancing.     

Role-Based Access Control in a Data Grid Using the Storage Resource Broker and Shibboleth

In this paper, we propose a role-based access control (RBAC) system for data resources in the Storage Resource Broker (SRB). The SRB is a Data Grid management system, which can integrate heterogeneous data resources of virtual organizations (VOs). The SRB stores the access control information of individual users in the Metadata Catalog (MCAT) database. However, because of the specific MCAT schema structure, this information can only be used by the SRB applications. If VOs also have many non-SRB applications, each with its own storage format for user access control information, it creates a scalability problem with regard to administration. To solve this problem, we developed a RBAC system with Shibboleth, which is an attribute authorization service currently being used in many Grid environments. Thus, the administration overhead is reduced because the role privileges of individual users are now managed by Shibboleth, not by MCAT or applications. In addition, access control policies need to be specified and managed across multiple VOs. For the specification of access control policies, we used the Core and Hierarchical RBAC profile of the eXtensible Access Control Markup Language (XACML); and for distributed administration of those policies, we used the Object, Metadata and Artifacts Registry (OMAR). OMAR is based on the e-business eXtensible Markup Language (ebXML) registry specifications developed to achieve interoperable registries and repositories. Our RBAC system provides scalable and fine-grain access control and allows privacy protection. Performance analysis shows that our system adds only a small overhead to the existing security infrastructure of the SRB.     

Resource provisioning based on preempting virtual machines in distributed systems

Resource provisioning is one of the main challenges in large‐scale distributed systems such as federated Grids. Recently, many resource management systems in these environments have started to use the lease abstraction and virtual machines (VMs) for resource provisioning. In the large‐scale distributed systems, resource providers serve requests from external users along with their own local users. The problem arises when there is not sufficient resources for local users, who have higher priority than external ones, and need resources urgently. This problem could be solved by preempting VM‐based leases from external users and allocating them to the local ones. However, preempting VM‐based leases entails side effects in terms of overhead time as well as increasing makespan of external requests. In this paper, we model the overhead of preempting VMs. Then, to reduce the impact of these side effects, we propose and compare several policies that determine the proper set of lease(s) for preemption. We evaluate the proposed policies through simulation as well as real experimentation in the context of InterGrid under different working conditions. Evaluation results demonstrate that the proposed preemption policies serve up to 72% more local requests without increasing the rejection ratio of external requests. Copyright © 2013 John Wiley & Sons, Ltd.     

网格环境下费用约束的科学工作流可靠调度算法

网格基础设施是目前科学工作流应用规划、部署和执行的主要支撑环境.然而由于网格资源的自治、动态及异构性,如何在保障用户QoS约束下有效调度科学工作流是一个研究热点.针对费用约束下的科学工作流调度问题,为了提高其执行的可靠性,本文使用随机服务模型描述资源节点的动态服务能力并考虑本地任务负载对资源执行性能的影响,给出一种资源可靠性的评估方法,在此基础上提出一种费用约束下的科学工作流可靠调度算法RSASW.仿真实验结果表明RSASW算法相对于GAIN3,GreedyTime-CD及PFAS算法,对工作流的执行具有很好的可靠性保障.     

Role-based access control for a Grid system using OGSA-DAI and Shibboleth

In this paper, we propose a new role-based access control (RBAC) system for Grid data resources in the Open Grid Services Architecture Data Access and Integration (OGSA-DAI). OGSA-DAI is a widely used framework for integrating data resources in Grids. However, OGSA-DAI’s identity-based access control causes substantial administration overhead for the resource providers in virtual organizations (VOs) because of the direct mapping between individual Grid users and the privileges on the resources. To solve this problem, we used the Shibboleth, an attribute authorization service, to support RBAC within the OGSA-DAI. In addition, access control policies need to be specified and managed across multiple VOs. For the specification of access control policies, we used the Core and Hierarchical RBAC profile of the eXtensible Access Control Markup Language (XACML); and for distributed administration of those policies and the user-role assignments, we used the Object, Metadata and Artifacts Registry (OMAR). OMAR is based on the e-business eXtensible Markup Language (ebXML) registry specifications developed to achieve interoperable registries and repositories. Our RBAC system provides scalable and fine-grain access control and allows privacy protection. It also supports dynamic delegation of rights and user-role assignments, and reduces the administration overheads for the resource providers because they need to maintain only the mapping information from VO roles to local database roles. Moreover, unnecessary mapping and connections can be avoided by denying invalid requests at the VO level. Performance analysis shows that our RBAC system adds only a small overhead to the existing security infrastructure of OGSA-DAI.