A Resource-Efficient Communication Architecture for Chip Multiprocessors on FPGAs

Significant advances in field-programmable gate arrays (FPGAs) have made it viable to explore innovative multiprocessor solutions on a single FPGA chip.For multiprocessors,an efficient communication network that matches the needs of the target application is always critical to the overall performance.Wormhole packet-switching network-on-chip (NoC) solutions are replacing conventional shared buses to deal with scalability and complexity challenges coming along with the increasing number of processing elements (PEs).However,the quest for high performance networks has led to very complex and resource-expensive NoC designs,leaving little room for the real computing force,i.e.,PEs.Moreover,many techniques offer very small performance gains or none at all when network traffic is light while increasing the resource usage of routers.We argue that computation is still the primary task of multiprocessors and sufficient resources should be reserved for PEs.This paper presents our novel design and implementation of a resource-efficient communication network for multiprocessors on FPGAs.We reduce not only the required number of routers for a given number of PEs by introducing a new PE-router topology,but also the resource requirement of each router.Our communication network relies on the NEWS channels to transfer packets in a pipelined fashion following the path determined by the routing network.The implementation results on various Xilinx FPGAs show good performance in the typical range of network load for multiprocessor applications.     

Long-range dependence and on-chip processor traffic

Long-range dependence is a property of stochastic processes that has an important impact on network performance, especially on the buffer usage in routers. We analyze the presence of long-range dependence in on-chip processor traffic and we study the impact of long-range dependence on networks-on-chip. long-range dependence in communication traces of processor ips at the cycle-accurate level. We also study the impact of long-range dependence on a real network-on-chip using the SocLib simulation environment and traffic generators of our own. Our experiments show that long-range dependence is not an ubiquitous property of on-chip processor traffic and that its impact on the network-on-chip is highly correlated with the low level communication protocol used.     

面向温度优化的片上网络任务调度方法

随着片上网络规模的扩大和研究的逐步深入,如何将芯片上众多的任务进行合理的调度成为系统温度优化的关键之一。针对片上网络任务调度问题, 提出一种基于最短曼哈顿距离的任务调度SMDS方案。该策略充分考虑核通信图中通信节点对之间最短曼哈顿路径,通过搜索算法寻找任务调度的目的节点,使用模拟退火算法确定任务调度对。实验结果显示,与传统的分布式任务调度 DTM策略相比,针对6*6、8*8和10*10的拓扑结构,SMDS实验方案在迁移次数方面的平均优化率分别为2208%、21.74%和23.02%。在平均跳数方面的平均优化率分别为24.04%、29.18%和23.04%,实现了系统温度优化。     

Dynamic QoS queuing control mechanism for multimedia differentiated services

An essential goal of communication networks is to provide multimedia services with QoS streaming. A properly designed multimedia QoS system must reserve requested resources according to user QoS requirements and the available network resources. However, the static resource allocation among priority queues in DiffServ networks leads to insufficient resource usage when a burst occurs in one priority queue while other queues starve. This study presents a User-Oriented QoS Streaming System to achieve perceptible satisfaction based on novel streaming and media differentiation policies in DiffServ networks. This study also proposes that the Dynamic QoS Queue Mapping (DQ²M) mechanism dynamically control queue scheduling by adaptively maximizing the utilization of queues and network resources according to the soft states of the DiffServ network. Evaluation results indicate that the proposed DQ²M algorithm can improve the fairness and efficiency of resource utilization for low-priority queues.     

基于OpenStack的高资源利用率Docker调度模型

在现有OpenStack云平台与Docker容器技术的集成方案中,基于容器初始资源请求的调度模型由于未充分考虑容器运行时的实际资源使用情况,导致资源利用率较低。为满足云计算领域的高资源利用率和低成本需求,构建基于OpenStack云平台的Docker调度模型（DSM）,将其与OpenStack的Keystone、Glance以及Neutron组件的API进行交互,获取创建容器所需的镜像、网络等资源,同时调用Docker Engine提供的API部署容器,对容器生命周期进行高效灵活管控。通过融合初始化模块、资源实时感知模块、容器调度模块、资源实时监测模块和容器迁移模块,并在容器调度模块中利用资源可用度评估与优先级决策调度机制为容器选择最优的计算节点,实现OpenStack云平台中资源的高效利用。实验结果表明,与经典Nova-Docker和Yun集成方案采用的调度模型相比,DSM调度模型在CPU和内存利用率上至少提升38.54、30.17个百分点和38.40、28.69个百分点。     

面向Dataflow的异构集群混合式资源调度框架研究

Dataflow模型的使用,使得大数据计算的批处理和流处理融合为一体.但是,现有的针对大数据计算的集群资源调度框架,要么面向流处理,要么面向批处理,不适合批处理与流处理作业共享集群资源的需求.另外,GPU用于大数据分析计算时,由于缺乏有效的CPU-GPU资源解耦方式,降低了资源使用效率.在分析现有的集群资源调度框架的基础上,设计并实现了一种可以感知批处理/流处理应用的混合式资源调度框架HRM.它以共享状态架构为基础,采用乐观封锁协议和悲观封锁协议相结合的方式,确保流处理作业和批处理作业的不同资源要求.在计算节点上,提供CPU-GPU资源的灵活绑定,采用队列堆叠技术,不但满足流处理作业的实时性需求,也减少了反馈延迟并实现了GPU资源的共享.通过模拟大规模作业的调度,结果显示, HRM的调度延迟只有集中式调度框架的75%左右;使用实际负载测试,批处理与流处理共享集群时,使用HRM调度框架, CPU资源利用率提高25%以上;而使用细粒度作业调度方法,不但GPU利用率提高2倍以上,作业的完成时间也能够减少50%左右.     

Adaptive scheduling algorithm for media-optimized traffic management in software defined networks

Multi-policy resource management have been considered as an efficient methodology for delivering ready-to-use media-optimized applications in Software-Defined Networks (SDNs). Prioritized flow scheduling ensures high-speed communication in SDNs under large-scale distribution, heterogeneity of network resources, and exponential distribution of the flows granularity. The effectiveness of priority-based approaches depends usually on the control mechanism of the resource management. In this paper we improve the resource utilization by developing a novel adaptive scheduling strategy. We came with an effecting scheduling strategy to determine what resource to be allocated to a set of flows keeping their priority, increasing the average utilization of resources and, most importantly, establishing a virtual circuit for a specific flow over a network. Our theoretical remarks and extensive simulation results show that the proposed scheduling strategies can achieve the described goals.     

基于代价优化的云工作流调度改进PSO算法

云计算可以通过即付即用的方式向用户工作流提供资源。为了解决资源服务代价异构环境下的云工作流任务调度代价问题,提出一种基于改进粒子群算法的云工作流任务调度算法WSA-IPSO。通过综合考虑任务的执行代价和依赖任务间发生数据传输时的通信代价,算法将总代价优化问题形式化为有向无环图DAG中的任务调度模型,并提出基于改进粒子群算法的优化模型对其进行求解。通过改进传统粒子群算法的粒子速度更新策略和惯性权重更新策略,算法可以以更快的收敛速度得到代价最小化的调度方案。通过仿真实验,与MCT算法及标准粒子群算法进行性能比较。实验结果表明,WSA-IPSO算法在降低总代价、任务分布的负载均衡以及算法收敛性方面比较同类算法均表现出更好的性能。     

On the Performance of a Dual-Objective Optimization Model for Workflow Applications on Grid Platforms

In attempts to exploit a diverse set of resources in grids efficiently, numerous assays in resource management, particularly scheduling, have been made. The primary objective of these efforts is the minimization of application completion time; however, they tend to achieve this objective at the expense of redundant resource usage. This paper investigates the problem of scheduling workflow applications on grids and presents a novel scheduling algorithm for the solution of this problem. Our algorithm performs the scheduling by accounting for both completion time and resource usage—dual objectives. Since the performance of grid resources changes dynamically and the accurate estimation of their performance is very difficult, our algorithm incorporates rescheduling to deal with unforeseen performance fluctuations effectively. The paper provides a comparative evaluation study conducted by using an extensive set of experiments. The study demonstrates that the proposed algorithm delivers promising performance in three respects: completion time, resource utilization, and robustness to resource-performance fluctuations.     

大数据流式计算框架Heron环境下的流分类任务调度策略

新型大数据流式计算框架Apache Heron默认使用轮询调度算法进行任务调度，忽略了拓扑运行时状态以及任务实例间不同通信方式对系统性能的影响。针对这个问题，提出Heron环境下流分类任务调度策略（DSC-Heron），包括流分类算法、流簇分配算法和流分类调度算法。首先通过建立Heron作业模型明确任务实例间不同通信方式的通信开销差异；其次基于流分类模型，根据任务实例间实时数据流大小对数据流进行分类；最后将相互关联的高频数据流整体作为基本调度单元构建任务分配计划，在满足资源约束条件的同时尽可能多地将节点间通信转化为节点内通信以最小化系统通信开销。在包含9个节点的Heron集群环境下分别运行SentenceWordCount、WordCount和FileWordCount拓扑，结果表明DSC-Heron相对于Heron默认调度策略，在系统完成时延、节点间通信开销和系统吞吐量上分别平均优化了8.35%、7.07%和6.83%；在负载均衡性方面，工作节点的CPU占用率和内存占用率标准差分别平均下降了41.44%和41.23%。实验结果表明，DSC-Heron对测试拓扑的运行性能有一定的优化作用，其中对接近真实应用场景的FileWordCount拓扑优化效果最为显著。     

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。     

提高卫星服务寿命的任务规划方法研究

基于组网卫星协同工作规划可提高复杂任务处理的基础上,本文探索降低规划风险,提高卫星服务寿命的卫星任务规划方法.从提高卫星服务寿命角度,建立用于描述资源使用期望的目标规划模型;建立卫星任务规划模型,在任务方案与资源使用中考虑了规划风险要素,并将资源使用目标模型引入到任务规划模型中,以期在规划中同时优化卫星服务寿命要素;基于邻域贪心扩展改善遗传禁忌求解算法,并运用分布并行技术提高工程实用性;利用STK提供基础仿真数据,对模型与算法进行了验证.     

Connection resource management for compiler-generated communication

Traditionally parallel compilers have targeted a standard message passing communication library when generating communication code (e.g. PVM, MPI). The standard message passing model dynamically reserves communication resources for each message. For regular, repeating communication patterns, a static communication resource reservation model can be more efficient. By reserving resources once for many communication exchanges, the communication startup time is better amortized. Plus, with a global view of communication, the static model has a wider choice of routes. While the static resource reservation model can be a more efficient communication target for the compiler, this model reveals the problems of scheduling use of limited communication resources. This paper uses the abstraction of a communication resource to define two resource management problems and presents three algorithms that can be used by the compiler to address these problems. Initial measures of the effectiveness of these algorithms are presented from two programs for an $8 \times 8$ iWarp system. © 1997 by John Wiley & Sons, Ltd.     

基于通信代价的网格资源调度算法

针对基于时间和预算限制的资源调度算法在调度数据密集型应用程序时存在的问题,提出一种新的基于通信代价的网格资源调度算法,综合考虑用户的时问限制和预算要求,根据用户作业的计算量与通信量选择具有一定计算能力,且通信代价较小的资源节点作为目标节点,通过减少此类程序提交到目标资源节点的通信代价,达到减少整个应用程序完成时间的目的。实验结果表明,该算法能够获得较好的性能。     

Mapping streaming applications on multiprocessors with time-division-multiplexed network-on-chip

This paper addresses mapping of streaming applications (such as MPEG) on multiprocessor platforms with time-division-multiplexed network-on-chip. In particular, we solve processor selection, path selection and router configuration problems. Given the complexity of these problems, state of the art approaches in this area largely rely on greedy heuristics, which do not guarantee optimality. Our approach is based on a constraint programming formulation that merges a number of steps, usually tackled in sequence in classic approaches. Thus, our method has the potential of finding optimal solutions with respect to resource usage under throughput constraints. The experimental evaluation presented in here shows that our approach is capable of exploring a range of solutions while giving the designer the opportunity to emphasize the importance of various design metrics.     

基于SaaS的弹性云平台优化调度策略设计

云计算平台利用虚拟化技术使软件应用变得更有效率的同时, 也给资源管理和服务调度带来了挑战。在研究了软件服务（SaaS）与基础设施服务（IaaS）调度的区别基础上, 重点考虑SaaS层的资源调度, 提出基于随机理论的调度模型, 把该层调度描述成一种多目标的优化问题。除了服务质量的要求, 还考虑了弹性这一云服务的重要特性, 并提供了任务调度与弹性服务副本的匹配策略。实验表明本调度机制的设计优化了云平台的整体性能, 达到了较好的负载均衡与资源利用率。     

GCC的流水冲突识别器和并行调度器

由于超长指令字处理器通常都有多级流水线和复杂的资源使用限制,如何准确地描述处理器的流水线模型,快速地判断是否存在资源冲突并不是个简单地任务。文章介绍GCC新引入的正则表达式语法的流水线描述机制。在将GCC移植到笔者所开发的SuperV芯片的过程中,利用该机制对SuperV芯片的流水线结构和资源使用限制进行详尽地描述,启动了GCC的指令级并行调度。通过并行调度,测试程序的性能提高了大约6%—35%。     

基于TDMA的空中自组网MAC层资源调度算法研究

空中移动无线自组织网络是一种拓扑结构快速变化,有自组织性的多跳无中心网络;针对传统时隙分配算法资源利用率低、吞吐量不足、通信距离近等问题,采用引入分配系数的混合时隙分配模式,通过节点业务优先级和流量预测相结合,设计了一种基于TDMA定向分布式资源动态调度算法（M-TDMA）;对比分析了节点数量、传输速率、分配系数以及不同拓扑等多个维度对算法传输时延、吞吐量以及丢包率的影响;最后通过仿真实验对资源调度算法进行验证;仿真结果表明,在20个网络节点时,网络的最大传输时延小于600 ms,网络吞吐量可以达到4.5 Mbps以上, M-TDMA算法通过高效的资源调度,有效降低了网络传输时延并提高了网络吞吐量;     

Novel Neighborhood Search for Multiprocessor Scheduling with Pipelining

This paper presents a neighborhood search algorithm for heterogeneous multiprocessor scheduling in which loop pipelining is used to exploit parallelism between iterations. The method adopts a realistic model for interprocessor communication where resource contention is taken into consideration. The schedule representation scheme is flexible so that communication scheduling can be performed in a generic manner. Base on a general time formulation of the schedule performance, the algorithm improves an initial schedule in an efficient way by successive modification to the task processor mapping and task ordering. Simulation results show that significant improvement over existing methods can be obtained. A parallel software video encoder was implemented based on the scheduling result and real time performance was achieved with pipelining of frame encoding.     

基于随机模型的云平台资源调度策略设计

针对云计算资源管理的实际需求,提出一种基于随机模型的云平台调度策略,设计合理高效的资源调度算法,解决传统代数模型请求丢失率高以及其他随机模型负载均衡指标性能较差的问题,从而在服务性能和执行效率的基础上保证服务器的资源负载,使云平台处于相对稳定的状态。在实验环境中的验证结果表明,该调度策略能够优化虚拟资源的使用效率和服务响应时间,同时能够达到较好的负载均衡并降低运营成本。