Revisiting the Parallel Strategy for DOACROSS Loops

Journal of Computer Science and Technology - DOACROSS loops are significant parts in many important scientific and engineering applications, which are generally exploited pipeline/wave-front...     

CAWR: Buffer Replacement with Channel‐Aware Write Reordering Mechanism for SSDs

A typical solid‐state drive contains several independent channels that can be operated in parallel. To exploit this channel‐level parallelism, a variety of works proposed to split consecutive write sequences into small segments and schedule them to different channels. This scheme exploits the parallelism but breaks the spatial locality of write traffic; thus, it is able to significantly degrade the efficiency of garbage collection. This paper proposes a channel‐aware write reordering (CAWR) mechanism to schedule write requests to different channels more intelligently. The novel mechanism encapsulates correlated pages into a cluster beforehand. All pages belonging to a cluster are scheduled to the same channels to exploit spatial locality, while different clusters are scheduled to different channels to exploit the parallelism. As CAWR covers both garbage collection and I/O performance, it outperforms existing schemes significantly. Trace‐driven simulation results demonstrate that the CAWR mechanism reduces the average response time by 26% on average and decreases the valid page copies by 10% on average, while achieving a similar hit ratio to that of existing mechanisms.     

基于Spark SQL的分布式全文检索框架的设计与实现

随着信息化的深入,大数据在各个领域产生了巨大的价值,海量数据的存储和快速分析成为新的挑战。传统的关系型数据库由于性能、扩展性的不足以及价格昂贵等方面的缺点,难以满足大数据的存储和分析需求。Spark SQL是基于大数据处理框架Spark的数据分析工具,目前已支持TPC-DS基准,成为大数据背景下传统数据仓库的替代解决方案。全文检索作为一种文本搜索的有效方式,能够与一般的查询操作结合使用,提供更加丰富的查询和分析操作。目前,Spark SQL仅支持简单的查询操作,不支持全文检索。为了满足传统业务迁移和现有业务的使用需求,提出了分布式全文检索框架,涵盖了SQL文法、SQL翻译转换框架、全文检索并行化、检索优化4个模块,并在Spark SQL上进行了实现。实验结果表明相比于传统的数据库,在两种检索优化策略下,该框架的索引构建时间、查询时间分别减少到传统数据库的0.6%/0.5%和1%/10%,索引存储量减少为传统数据库的55.0%。     

A stencil of the finite-difference method for the 2D convection diffusion equation and its new iterative scheme

The paper gives the numerical stencil for the two-dimensional convection diffusion equation and the technique of elimination, and builds up the new iterative scheme to solve the implicit difference equation. The scheme's convergence and its higher rate of convergence than the Jacobi iteration are proved. And the numerical example indicates that the new scheme has the same parallelism and a higher rate of convergence than the Jacobi iteration.     

光电综合标校系统光轴平行度标校方法

针对光电综合标校系统的功能组成及标校原理,提出了标校靶标的设计及标校方法,并利用感光相纸对激光光斑进行聚焦采集,实现了对电视/红外/激光三者光轴一致性的标校,检测结果验证了该标校靶标及标校方法的可行性和有效性。     

一种连续变焦距电视的变倍补偿机构设计

为了实现电视的焦距连续变化,设计了一种连续变焦距电视的变倍补偿机构。提出了一种提高变倍、补偿机构运动精度的方法,实测两机构直线运动精度均为15″。通过靶场试验证明连续变焦距电视的变倍补偿机构的运动精度满足电视捕获、跟踪目标的精度要求,两直线滑轨平行度调整机构设计、加工简单,可以在其它直线运动机构中得到推广应用。     

Sorting-Based Calculation of Zeroes and Extrema of Functions as Applied to Search and Recognition. I

A stable address sorting is used for approximate calculation of zeroes and extrema of functions of one variable and two variables. In particular, the paper gives a new method for localization and approximate calculation of all the roots of a polynomial with arbitrary power. The method includes poorly separated roots, and its distinctive features are the computational stability and the maximum parallelism.     

Accelerating sequential programs on commodity multi-core processors

A recently proposed pipelined multithreading (PMT) technique exhibits wide applicability in parallelizing general sequential programs on multi-core processors. However, significant inter-core communication overhead limits PMT performance and prevents its commercial utilization. A simple and effective clustered pipelined multithreading (CPMT) approach is presented to accelerate sequential programs on commodity multi-core processors. This CPMT technique adopts a clustered communication mechanism that can yield very low average communication overhead by eliminating false sharing as well as reducing communication operation and transit delays in the software-only approach. A single-producer/single-consumer concurrent lock-free clusteredQueue algorithm based on a two-level queue structure is also proposed. The accuracy of CPMT is theoretically demonstrated. The performances of the algorithm and CPMT are evaluated on a commodity AMD Phenom four-core processor. The number of enqueue and dequeue times of the algorithm are 20.8 and 23 cycles given an appropriate parameter, respectively. The speedup of CPMT ranges from 13.1% to 119.8% for typical loops extracted from the SPEC CPU 2000 benchmark suite.     

Kokkos: Enabling manycore performance portability through polymorphic memory access patterns

The manycore revolution can be characterized by increasing thread counts, decreasing memory per thread, and diversity of continually evolving manycore architectures. High performance computing (HPC) applications and libraries must exploit increasingly finer levels of parallelism within their codes to sustain scalability on these devices. A major obstacle to performance portability is the diverse and conflicting set of constraints on memory access patterns across devices. Contemporary portable programming models address manycore parallelism (e.g., OpenMP, OpenACC, OpenCL) but fail to address memory access patterns. The Kokkos C++ library enables applications and domain libraries to achieve performance portability on diverse manycore architectures by unifying abstractions for both fine-grain data parallelism and memory access patterns. In this paper we describe Kokkos’ abstractions, summarize its application programmer interface (API), present performance results for unit-test kernels and mini-applications, and outline an incremental strategy for migrating legacy C++ codes to Kokkos. The Kokkos library is under active research and development to incorporate capabilities from new generations of manycore architectures, and to address a growing list of applications and domain libraries.     

一种支持细粒度并行的SDN虚拟化编程框架

软件定义网络(software defined network,简称SDN)通过集中式的控制器提高了网络的可编程性,成为近年来网络领域非常热门的话题。以Openflow网络为代表的软件定义网络将逻辑控制与数据转发相隔离,为网络虚拟化技术提供了良好的平台。集中式的抽象与控制使得SDN虚拟化框架的处理效率成为主要瓶颈。现有的SDN虚拟化框架由于缺乏对细粒度并行的支持,为编程人员充分利用多核/众核资源、控制更大规模的网络带来了极大的挑战。为了提高SDN虚拟化框架的处理效率,提出一种新的SDN虚拟化编程框架,通过新颖的API和运行时,在框架内部支持细粒度的并行处理。该框架通过对网络中流和网络资源进行抽象,使开发人员可以直接通过划分流空间来定义不同的虚拟网络,利用无锁的编程方式对共享的网络资源和流进行操作。实验结果表明,该框架在逻辑控制的执行效率方面具有良好的可扩展性,可以创建出更大规模的虚拟网络,并对其进行更为复杂的控制。