一种基于微环的新型4×4非阻塞光路由开关

光互连技术因诸多特性优于电互连而成为片上多核互连最具前景的解决方案。为了提高片上光互连网络架构的性能,采取光器件模块搭建的方法,提出了一种基于微环的新型4×4光路由开关,仅用7个微环构建的拓扑结构便实现了4个双向端口的非阻塞交换,降低了功耗和面积;波导交叉的数量减少到6个,优化了插入损耗。结果表明,该结构相对于经典结构光器件的功耗节省了约8%,光互连层的插入损耗降低了约7%。     

MPI技术在胶州湾水质预报系统的应用研究

为了能够更高效的提高胶州湾水质预报系统的运行速度,在原有基于串行模式的预报系统中引入了消息传递接口MPI (message-passing interface)并行处理技术,并在该系统中分别采用组通信模式和非阻塞通信模式,把整个水质预报系统分解为多个子任务,在基于MPI消息传递模式的集群系统中分别进行运算,通过实验表明了两种通信模式在运算速度上的差异,在预报系统中非阻塞通信模式比组通信模式更加优化.因此,非阻塞通信模式对水质预报模式的改进具有很大的潜力.     

改进型锁无关双端队列的设计与实现

高性能实时系统对系统性能、确定性和容错性有着更高的要求。非阻塞同步在任务同步方面满足要求,实现方法之一就是设计锁无关数据结构。介绍了设计锁无关数据结构算法的关键技术,通过对已有算法不足的分析提出了一种改进型的锁无关双端队列算法,介绍了对该算法的实验分析和实际应用。实验结果表明,该算法提高了访问双端队列的执行速度,并避免了多任务间同步引发的死锁、优先级逆转、低容错性等缺点。     

话音无阻塞的可视电话系统设计与实现

为实现话音无阻塞、视频图像清晰流畅和低延迟的高品质可视通话,提出了一种使用PSTN单独传送话音,使用IP网络单独传送通话者视频并在终端同步还原的可视电话系统及其实现方法,系统由终端和支撑服务器群构成。设计用PSTN的话音拨号信号或来电振铃信号作为视频呼叫的触发信号,终端通过捕获电话号码封装为SIP的Invite消息并发起视频呼叫,与对端建立P2P连接并使用UDP报文传送RTP视频流来实现视频通话。系统测试表明通话话音始终无阻塞;在带宽为256kbit／s、IP网络无拥塞的条件下,实现了视频CIF分辨率,帧频30fps,视频延迟小于60ms;用户使用习惯如电话号码和拨号方式无需任何改变。其综合性能优于现有的PSTN可视电话和IP可视电话。     

非周期演化数据流的无阻塞连接算法

数据流的连接常作为数据流查询操作的支撑算法.以往算法多考虑的是周期性演化的数据流,对于非周期性数据流连接涉及较少.提出一种变换高斯分布下的数据流连接算法.通过采样统计确定当前高斯中心点,并以此为中心划分数据块.提出在变换高斯分布下的确定数据连接块的方法.实验表明本算法与同类算法相比可以在有限的内存下产生更高的连接率,更小的I/O代价.     

The split-switching network (SSN): A high-performance multicast ATM switch

This paper proposes a new high-performance multicast ATM switch architecture. The switch, called the split-switching network (SSN), is based on banyan networks. The SSN achieves multicasting in a way that is non-typical for banyan-based switches: copying and routeing of multicast cells are carried out simultaneously and within the same fabric. Thus, cells are copied only when needed as they traverse the switch towards the appropriate output ports. The SSN consists of successive spliting stages, and buffering is provided in front of each stage. The SSN is non-blocking with complexity of order Nlog2/2N for a switch of size N, and is characterized by distributed and parallel control. The throughput-delay performance of the SSN is shown to be similar to that of a non-blocking output-buffering switch under different mixtures of unicast/multicast traffic. In particular, the SSN achieves a maximum throughput of 100 per cent and the cell delay and delay variation remain small for loads just below the maximum throughput.     

光计算中全排列无阻塞双Omega光互连网络的光学实现方法

本文分析了单个Ｏｍｅｇａ光互连网络的局限性，提出了一种全排列无阻塞双Ｏｍｅｇａ光互连网络的光学实现方法。光回路系统由Ａｒ＋３离子激光器、光束分束器、偏振组合棱镜和液晶空间光开关组成，能够实现输入信号列阵光束的并行传输和全排列无阻塞光互连。     

On deadlock control for a class of generalised Petri net models of flexible manufacturing systems

This paper develops a deadlock prevention policy for a class of Petri nets that can model flexible manufacturing systems with assembly and disassembly operations. Siphons in a plant Petri net model are divided into elementary and dependent siphons according to the linear dependency of their characteristic T-vectors. The proposed approach is to make every siphon satisfy the controlled-siphon property (the cs-property), i.e., at any reachable marking, any siphon is max-marked, so that no deadlock states can be reached. The satisfiability of the cs-property is achieved by explicitly adding a monitor for each elementary siphon. The max-controllability of a dependent siphon is ensured by properly supervising its elementary siphons. More permissive behaviour of the non-blocking supervisor is obtained through the rearrangement of the output arcs of the monitors. Compared with existing policies reported in the literature, the advantage of the present method is that a small number of monitors are added and the iterative computing process is accordingly avoided. Finally, the application of the proposed method to an FMS example is presented.     

A Mechanically Checked Proof of a Multiprocessor Result via a Uniprocessor View

We describe a mechanically checked correctness proof for a system of n processes, each running a simple, non-blocking counter algorithm. We prove that if the system runs longer than 5n steps, the counter is increased. The theorem is formalized in applicative Common Lisp and proved with the ACL2 theorem prover. The value of this paper lies not so much in the trivial algorithm addressed as in the method used to prove it correct. The method allows one to reason accurately about the behavior of a concurrent, multiprocess system by reasoning about the sequential computation carried out by a selected process, against a memory that is changed externally. Indeed, we prove general lemmas that allow shifting between the multiprocess and uniprocess views. We prove a safety property using a multiprocess view, project the property to a uniprocess view, and then prove a global progress property via a local, sequential computation argument. Our uniprocessor view is a formal compositional semantics for a shared memory system.     

Enabling scalable parallel implementations of structured adaptive mesh refinement applications

Parallel implementations of dynamic structured adaptive mesh refinement (SAMR) methods lead to significant runtime management challenges that can limit their scalability on large systems. This paper presents a runtime engine that addresses the scalability of SAMR applications with localized refinements and high SAMR efficiencies on large numbers of processors (upto 1024 processors). The SAMR runtime engine augments hierarchical partitioning with bin-packing based load-balancing to manage the space-time heterogeneity of the SAMR grid hierarchy, and includes a communication substrate that optimizes the use of MPI non-blocking communication primitives. An experimental evaluation on the IBM SP2 supercomputer using the 3-D Richtmyer-Meshkov compressible turbulence kernel demonstrates the effectiveness of the runtime engine in improving SAMR scalability.