回归的能源有效网络大数据流汇聚算法研究

为了降低传感器网络数据流汇聚时的能源消耗,提出了一种基于回归的能源有效数据流汇聚算法。首先,将传感器节点分为活跃节点和能源有效节点。然后,以活跃节点为中心点将所有节点进行聚类,并应用回归方法通过活跃节点的数据流对能源有效节点的数据进行预测。接下来,通过节点预测值的累积误差不断修正活跃节点集。最后,应用活跃节点的数据流信息对能源有效节点的数据进行预测。实验表明,本文提出的算法与其它相关算法相比具有更好的预测准确性。     

基于高速双端口的网络扫描系统研究

网络扫描是网络测试工作中一个非常重要的环节,为提高网络扫描速率和增强扫描功能,对双端口网络扫描系统进行了研究;该系统拟采用双端口线速数据流发送与接收相分离的方式,同时结合发送数据流字段变化、接收地址学习、接收规则过滤等技术,使其形成多种高速网络扫描功能;应用分析表明,扫描速率最高可达156万次/秒,且功能具备实用、灵活和多样化等特点。     

Achieving High Data Throughput in Research Networks

After less than a year of operation ,the BaBar experiment at SLAC has collected almost 100 million particle collision events in a database approaching 165TB.Around 20 TB of data has been exported via the Internet to the BaBar regional center at IN2P3 in Lyon,France,and around 40TB of simulated data has been imported from the Lawrence Livermore National Laboratory(LLNL),BaBar Collaborators plan to double data collection each year and export a third of the data to IN2P3.So within a few years the SLAC OC3 (155Mbps) connection will be fully utilized by file transfer to France alone.Upgrades to infrastructure is essential and detailed understanding of performance issues and the requirements for reliable high throughput transfers is critical.In this talk results from active and passive monitoring and direct measurements of throughput will be reviewed.Methods for achieving the ambitious requirements will be discussed.     

一种改进的指针安全分析算法

指针的安全性是C语言程序中的常见问题,它严重影响到系统的安全性.针对指针静态分析中的精度问题,比较分析了目前常见的5种指针分析算法的优缺点,提出了一种基于流不敏感的改进指针静态分析算法.该算法是在Steensgaard算法的基础上,对程序的控制流图进行改进,通过该算法可以得到比常见的数据流不敏感算法更为精确的指针分析结果.在实验室开发的国内首款商用编译器中实现了该算法,实验结果表明该算法可以提高指针分析的精度.     

Quality of Service on Linux for the Atlas TDAQ Event Building Network

Congestion control for packets sent on a network is important for DAQ systems that contain an event builder using switching network technologies.Quality of Service(QoS) is a technique for congestion control.Recent Linux releases provide QoS in the kernel to manage network traffic.We have analyzed the packet-loss and packet distribution for the event builder prototype of the Atlas TDAQ system.We used PC/Linux with Gigabit Ethernet network as the testbed.The result showed that QoS using CBQ and TBF eliminated packet loss on UDP/IP transfer while the UDP/IP transfer in best effort made lots of packet loss.The result also showed that the QoS overhead was small.We concluded that QoS on Linux performed efficiently in TCP/IP and UDP/IP and will have an important role of the Atlas TDAQ system.     

静态数据流计算机的性能分析

本文首先给出了静态数据流计算机的模型和数据流语言的模型,然后提出了把指令执行开销作为参数来计算数据流计算机实际并行度的系统模型,基于该系统模型通过统计可知程序在静态数据流计算机上运行时,实际最大的并行度由MP/(OH+1)决定(MP是程序并行度的期望,OH是指令执行的开销).因此,我们说开销极大地影响了数据流计算机的性能.