异构计算系统中弹性节能调度策略研究

目前,节能已成为异构计算系统中减少电量开销、提高系统可靠性和保护环境的重要研究内容.传统的节能调度策略侧重于研究如何节能而忽略了用户对任务完成时间的期望,使得任务执行效果受到较大影响.特别是当系统负载较重时,由于电压调节缺乏自适应性,导致在某些情况下(如应急服务)的任务执行效果不可容忍.文中提出一种弹性节能调度策略(Elastic Energy-Aware Scheduling,EEAS),用于动态调度异构计算系统中非周期、独立任务.EEAS策略根据系统负载情况在系统节能与用户期望之间进行权衡,即当系统负载较重时,EEAS优先考虑用户期望,通过动态调整计算节点局部队列中等待任务的执行电压提高任务完成率;当系统负载较轻时,EEAS在尽量满足用户期望的基础上最大限度地降低任务执行电压以实现节能.文中通过大量的模拟实验比较了EEAS、GEA、HVEA和LVEA的性能.实验结果表明,EEAS的调度质量优于其他策略,可有效提高系统弹性.     

Energy-Aware Scheduling on Multicore Heterogeneous Grid Computing Systems

We address a multicriteria non-preemptive energy-aware scheduling problem for computational Grid systems. This work introduces a new formulation of the scheduling problem for multicore heterogeneous computational Grid systems in which the minimization of the energy consumption, along with the makespan metric, is considered. We adopt a two-level model, in which a meta-broker agent (level 1) receives all user tasks and schedules them on the available resources, belonging to different local providers (level 2). The computing capacity and energy consumption of resources are taken from real multi-core processors from the main current vendors. Twenty novel list scheduling methods for the problem are proposed, and a comparative analysis of all of them over a large set of problem instances is presented. Additionally, a scalability study is performed in order to analyze the contribution of the best new bi-objective list scheduling heuristics when the problem dimension grows. We conclude after the experimental analysis that accurate trade-off schedules are computed by using the new proposed methods.     

新型分布式计算系统中的异构任务调度框架

随着大数据和机器学习的火热发展,面向机器学习的分布式大数据计算引擎随之兴起.这些系统既可以支持批量的分布式学习,也可以支持流式的增量学习和验证,具有低延迟、高性能的特点.然而,当前的一些主流系统采用了随机的任务调度策略,忽略了节点的性能差异,因此容易导致负载不均和性能下降.同时,对于某些任务,如果资源要求不满足,则会导...     

一种新的异构实时分布式系统的容错调度算法

一般来说,异构分布式实时系统中任务的周期并不完全相同且任务的时限不等于它们的周期,同时系统中还有一些无容错需求的任务.因此现有的任务调度算法一般不能满足这些要求.针对这类系统,在结合基版本/副版本技术和EDF算法的基础上,给出了一种新的容错调度算法.该算法由两部分组成：任务分配调度算法和单处理器调度算法.对于单处理器调度算法,本文采用了EDF算法;在此基础上,给出一种启发式静态任务分配算法.分析了系统的可调度性,给出了任务可调度条件和基版本/副版本时限的设置方法.仿真结果表明,这种算法是有效的.     

一种适于异构环境的任务调度算法

针对异构环境独立任务调度问题提出两个调度原则,并基于Min-min算法提出优先级最小最早完成时间算法(Priority min-min,PMM).该算法将任务在各处理机上执行时间的标准误差作为任务的优先级.选取最早完成时间较小的k个任务,优先调度其中优先级最高的一个.在实验基础上分析了参数$k$对PMM算法性能的影响. PMM算法克服了min-min算法单纯追求局部最优的局限性,更适合于异构环境.实验数据表明PMM算法能有效地降低调度跨度,其性能比min-min算法有明显提高.     

异构环境下独立任务调度算法的研究

本文基于Min-min算法和Sufferage算法提出了基于任务调度损失的最小最早完成时间算法(Sufferage Min-min,SMM).该算法将任务调度损失引入Min-min算法,选取最早完成时间较小的k个任务,再优先对其中任务调度损失最大的一个进行调度.SMM算法克服了Min-min算法单纯追求局部最优而缺少全局意识的缺点.测试表明,SMM算法可以做到调度跨度低与平均等待时间小的统一,在综合性能上较Min-min算法有所提高.     

异构机群系统上带返回信息的可分负载多轮调度算法

针对处理机具有不同的计算速度、通信能力的异构机群计算环境,以及实际应用中许多问题的求解在处理完任务后向中心处理机节点返回处理结果信息的情形,通过允许计算和通信操作重叠执行,采取FIFO调度策略和多次并行分配计算任务的方法,提出一种带返回结果信息的调度轮数可变的可分负载多轮调度算法.实验结果表明,该算法对于处理具有返回结果信息的应用的调度性能优于UMR可分负载多轮调度算法,并且可以获得近似最优的调度轮数.     

An Integrated Technique for Task Matching and Scheduling onto Distributed Heterogeneous Computing Systems

This paper presents a problem-space genetic algorithm (PSGA)-based technique for efficient matching and scheduling of an application program that can be represented by a directed acyclic graph, onto a mixed-machine distributed heterogeneous computing (DHC) system. PSGA is an evolutionary technique that combines the search capability of genetic algorithms with a known fast problem-specific heuristic to provide the best-possible solution to a problem in an efficient manner as compared to other probabilistic techniques. The goal of the algorithm is to reduce the overall completion time through proper task matching, task scheduling, and inter-machine data transfer scheduling in an integrated fashion. The algorithm is based on a new evolutionary technique that embeds a known problem-specific fast heuristic into genetic algorithms (GAs). The algorithm is robust in the sense that it explores a large and complex solution space in smaller CPU time and uses less memory space as compared to traditional GAs. Consequently, the proposed technique schedules an application program with a comparable schedule length in a very short CPU time, as compared to GA-based heuristics. The paper includes a performance comparison showing the viability and effectiveness of the proposed technique through comparison with existing GA-based techniques.     

Linear Programming-Based Affinity Scheduling of Independent Tasks on Heterogeneous Computing Systems

Resource management systems (RMS) are an important component in heterogeneous computing (HC) systems. One of the jobs of an RMS is the mapping of arriving tasks onto the machines of the HC system. Many different mapping heuristics have been proposed in recent years. However, most of these heuristics suffer from several limitations. One of these limitations is the performance degradation that results from using outdated global information about the status of all machines in the HC system. This paper proposes several heuristics which address this limitation by only requiring partial information in making the mapping decisions. These heuristics utilize the solution to a linear programming (LP) problem which maximizes the system capacity. Simulation results show that our heuristics perform very competitively while requiring dramatically less information.     

Method and Algorithms for Adaptive Multiagent Resource Scheduling in Heterogeneous Distributed Computing Environments

Automation and Remote Control - This article deals with the creation of new methods and algorithms for resource scheduling in heterogeneous distributed computing systems using the example of cloud...     

一种实时异构嵌入式系统的任务调度算法

异构分布式系统已被广泛应用在实时嵌入式系统中,而调度算法是在进行嵌入式系统综合时,确保系统实现性能目标的一个关键问题,这是一个NP-完全问题.现有的算法主要是启发式算法,性能还有待提高.提出了一个异构分布式系统的动态BLevel优先(dynamic BLevel first,简称DBLF)算法,算法选择就绪任务中动态BLevel值最大的任务进行调度,用插入法为任务分配处理器,遵循以下3个插入原则:满足任务先后顺序关系;任务的最早完成时间(earliest-finish-time,简称EFT)最小;在EFT相等时,优先分配到利用率较低的处理器上.与现有算法比较可以看出,DBLF算法可以有效降低调度长度.     

Fault-Tolerant Dynamic Rescheduling for Heterogeneous Computing Systems

As the scale and complexity of heterogeneous computing systems grow, failures occur frequently and have an adverse effect on solving large-scale applications. Hence, fault-tolerant scheduling is an imperative step for large-scale computing systems. The existing fault-tolerant scheduling algorithms belong to static scheduling, and they allocate multiple copies of each task to several processors no matter whether processor failures affect the execution of tasks. Such active replication strategies not only waste resource but also sacrifice the makespan. What is more, they cannot guarantee the successful execution of applications. In this paper, we propose a fault-tolerant dynamic rescheduling algorithm named FTDR, which can overcome above drawbacks. FTDR keeps listening to the processor failure, and reschedules the suspended tasks once failures occur. Because FTDR reschedules the tasks that are suspended because of failures, it can tolerate an arbitrary number of failures. Randomly generated DAGs are tested in our experiments. Experimental results show that the proposed algorithm achieves good performance in terms of makespan and resource consumption compared with its direct competitors.     

基于状态转换的信息服务系统可生存性评估方法

可生存性评估是使用各种方法对目标系统进行定性及定量的生存性分析。本文提出了一个基于状态转换的信息服务系统可生存性评估方法,该方法包括一个系统生存性模型和一个生存性函数的定义。最后给出了使用这种评估方法对目标系统进行评估的实例。     

一种异构计算环境下基于复制的调度算法

提出了一种异构环境下的基于复制的调度算法（TDNH）,并与同为异构环境下的HEFT算法进行了比较,结果证明TDNH算法减小了时间跨度。最后通过实验证明了TDNH算法能够得到比较好的结果。     

OpenCL: A Parallel Programming Standard for Heterogeneous Computing Systems

We provide an overview of the key architectural features of recent microprocessor designs and describe the programming model and abstractions provided by OpenCL, a new parallel programming standard targeting these architectures.     

异构分布式实时仿真系统的容错调度算法

异构分布式实时仿真系统是一类特殊的实时系统,基于改进的SP(spare processor)容错模型(checkpoint-based spare processor,简称CSP)对其容错问题进行了研究.首先,根据仿真系统的特点提出了两个命题,这是后续工作的基础;而后,基于Markov链对仿真任务的最坏反应时间进行了分析,并提出了仿真任务的可调度性分析规则;最后,基于CSP容错模型和上述可调度分析规则提出了异构分布式实时仿真系统的容错调度算法CSP-RTFT.算法的仿真结果表明:该算法较之基于SP模型的算法SP-RTFT可获得更好的稳定性、更高的任务接收率;缺点是资源利用率比PB模型下的算法要低.     

异构计算环境下一种新型的多DAG任务调度算法

随着计算机技术的发展与用户需求的不断提升,多有向无环图(DAG)共享一组异构计算资源的问题受到广泛的关注。但由于实际任务的复杂多变,多个DAG之间存在一定的差异,导致多DAG调度策略存在公平性问题。为此,提出一种改进的启发式公平调度算法IFairness。在选择待调度DAG阶段采用一种新的评判指标DAG完成度,代替原Fairness算法中的剩余Makespan作为DAG选择依据,在计算每个DAG的滞后程度阶段,采用"向后看"一步的原则,解决某些DAG在初期得不到调度的问题。仿真结果表明,与原Fairness算法相比,IFairness算法不公平程度降低了7.28%,资源利用率提升了11.97%,有效提高了调度算法的公平性及资源利用率。     

A Load Balanced Task Scheduling Heuristic for Large-Scale Computing Systems

Optimal task allocation in Large-Scale Computing Systems (LSCSs) that endeavors to balance the load across limited computing resources is considered an NP-hard problem. MinMin algorithm is one of the most widely used heuristic for scheduling tasks on limited computing resources. The MinMin minimizes makespan compared to other algorithms, such as Heterogeneous Earliest Finish Time (HEFT), duplication based algorithms, and clustering algorithms. However, MinMin results in unbalanced utilization of resources especially when majority of tasks have lower computational requirements. In this work we consider a computational model where each machine has certain bounded capacity to execute a predefined number of tasks simultaneously. Based on aforementioned model, a task scheduling heuristic Extended High to Low Load (ExH2LL) is proposed that attempts to balance the workload across the available computing resources while improving the resource utilization and reducing the makespan. ExH2LL dynamically identifies task-to-machine assignment considering the existing load on all machines. We compare ExH2LL with MinMin, H2LL, Improved MinMin Task Scheduling (IMMTS), Load Balanced MaxMin (LBM), and M-Level Suffrage-Based Scheduling Algorithm (MSSA). Simulation results show that ExH2LL outperforms the compared heuristics with respect to makespan and resource utilization. Moreover, we formally model and verify the working of ExH2LL using High Level Petri Nets, Satisfiability Modulo Theories Library, and Z3 Solver.