一种策略灵活的多版本文件系统的设计与实现

设计并实现了一种新型的多版本文件系统——HFS。HFS把启发式规则和用户规则结合起来,提供灵活的版本保留策略,为必要的文件保留必要的版本,节省了存储空间。同时,还通过写时复制和目录版本等技术,进一步降低对存储资源的消耗。分析表明,该系统可以在保证数据可靠性的同时,有效地提高存储资源的利用率。HFS的实现基于Ext3,对HFS和Ext3的文件读写性能的测试和比较的结果表明,HFS的顺序读文件的性能与Ext3相当,顺序写文件的性能和随机读写的性能比Ext3降低了2％-5％,版本机制对文件系统性能的影响很小。     

一种基于捎带回收的瓦记录Raid5写顺序化方法

瓦记录技术(SMR)采用部分叠加相邻磁道的方式增大磁盘碟片的面密度,其磁道相互叠加,使得瓦记录磁盘无法进行原地更新,是因为直接更新某一磁道上的数据可能会覆盖掉相邻磁道上的有效数据,导致瓦记录磁盘的非顺序写需要引入额外的写放大开销。对于由瓦记录磁盘组成的Raid5来说,在非顺序写场景下,最终落盘的IO也是非顺序的,因此单盘性能下降,会极大地影响瓦记录Raid5的非顺序写性能。针对瓦记录Raid5非顺序写性能低的问题,提出了一种基于捎带回收的瓦记录Raid5。该方法改进了传统Raid5的小写过程,在小写读的时候捎带读取附近的可被回收的数据或校验,通过将落盘写IO日志化,实现了落盘写IO的顺序化,避免了瓦记录磁盘的写放大开销。系统维护了地址映射表等元数据,在重构时只需对应用访问过的条带进行读写和计算,这减少了重构时需要读写的数据量。测试显示,在某些非顺序写场景下,写带宽超过了以Bcache作为缓存的MD Raid5。从计算结果看,重构时的读写开销相比普通磁盘组建的MD Raid5也有明显减小。     

基于多版本并发控制的移动事务模型的研究

移动数据库系统(MDS)的事务管理必须运作于诸如低带宽,电量限制,弱连接,和移动性等很多有限的资源下。由于这些限制,传统的并行控制机制不能够维持活动事务的可用性。所以改进的事务调度和并行控制机制必须能够充分的发挥MDS的潜能。阐述关于多版本事务处理方法和基于结合时间戳的多版本两段加锁的无死锁并行控制机制。目标模型通过改进的并行度,减少的事务等待时间和更小撤消率提供更高的吞吐量。     

双光子多层荧光存储中的变焦选层技术

为了在双光子多层荧光数据存储中实现对信息层的跟踪聚焦,基于目前的光盘存储系统和双元透镜控制方案,提出了一种变焦选层技术。两个可独立控制的音圈电机负责调节两物镜的相对位置。利用聚焦模块提取出的聚焦误差信号,读写系统控制两个电机协调动作改变焦点位置,使得读写焦点在对选定层读写过程中跟踪盘片抖动。选层操作可通过控制一个音圈电机完成。因此,读写系统能实现简单盘片结构条件下的多层荧光数据存储,同时聚焦模块可使系统与现有CD/DVD盘片兼容。利用二维精密平移台和三洋光头(SF-HD60S)中的音圈电机和物镜进行了双光子三维信息读写实验,在材料中存储了三层数据,点间距4μm,层间距20μm。     

DHT系统的安全性优化方法研究

对分布式哈希表(DHT)系统的安全脆弱性问题进行了研究,提出了多种安全性优化策略,并给出了一个原型系统。进行了真实网络实验,实验数据表明,现有DHT网络易受索引毒害和路由污染攻击,产生的错误查询结果甚至会引发更大规模的网络安全事件。通过改进一个个DHT系统的节点ID生成机制、路由表更新机制和搜索路径选择机制,从系统运行的各个阶段提升其安全场,抵御攻击者共谋。基于上述方法设计的原型系统在保证平均查询跳数增加不到1跳的情况下,在共谋攻击节点占比60%的网络中,将系统查询成功率保持在65%以上,其方法适用于各种分布式哈希表结构,具有重要的实际应用前景。     

NUMA结构的高效实时稳定的垃圾回收算法

针对非一致性内存访问架构(NUMA)在垃圾回收(GC)过程中存在大量的远程内存读写导致GC性能降低的问题,对GC过程的各个阶段进行分析与研究,提出了一种基于NUMA结构的高效实时稳定的GC算法。该算法首先基于NUMA结构改进传统分代GC机制的堆空间布局,然后通过控制GC过程中扫描活跃对象阶段的初始根对象选取、动态负载均衡阶段截取任务队列的选取以及复制活跃对象阶段对象复制位置的选取,大大减少GC过程中的远程访问次数。这种改进的GC机制对所有NUMA结构具有通用性。以Godson-3处理器的NUMA平台为例进行的实验结果显示,优化的GC机制极大地缩短了GC的时间,而且提高了应用程序的性能以及稳定性。在SPECjvm2008测试中,GC时间平均缩短了14.6%(GC总时间缩短4.1%~41.58%),应用程序的性能平均提升了4.68%(最高提升17.8%),应用程序的性能稳定性提升了76.2%。     

基于MOIS-树的快速最近邻查询算法

以提高最近邻查询和k最近邻查询效率为目标,给出了空间对象间序关系的定义,以此为基础提出了一种空间数据索引结构--MOIS(multi-order index structure)-树,规定MOIS-树中的中间节点的所有孩子节点按其几何位置满足某种序的关系,从而使得在中间节点中进行最近邻和k最近邻查询时可以进行快速定位....     

基于级联稀疏查询机制的轻量化火灾检测算法

针对现有火灾检测算法仍存在的模型复杂、检测速度慢、误检率高等问题,提出一种基于级联稀疏查询机制的轻量化火灾检测网络LFNet。首先,建立了轻量化的图像特征提取模块ECDNet,其通过在YOLOv5s主干网络中嵌入轻量化注意力模块ECA (efficient channel attention),用于解决火灾检测中火焰与烟雾的多尺度难点;其次,利用深层特征提取模块FPN+PAN,对不同层级的特征图进行深度处理和多尺度融合;最后,利用嵌入轻量化的级联稀疏查询模块CSQ (cascade sparse query)提升对早期火灾中的小火焰与薄烟雾的检测准确率。实验表明,本文方法在mAP和Precision等客观指标上的综合表现达到最优,同时在实现较高检测精度时的参数量也较低,能够满足实际场景的火灾检测要求。     

基于GIS的绿潮漂移预测数值模拟系统的建设与应用分析

实现了海洋数值模式计算与地理信息系统(GIS)空间分析及产品制作的集成.针对数值模式计算时间长的问题,从空间数据组织角度研究了绿潮数值模式的并行计算,将数据与计算模式按空间范围进行分解并建立其对应的空间索引,实现了分区计算过程自动化,大幅缩短了计算时间.采用GIS空间表达技术,研究了绿潮数值模拟结果在GIS中的可视化,将绿潮预测结果转换为以散点表示的绿潮覆盖点与以边框线表示的绿潮分布范围,并以矢量元素实时绘制,实现了绿潮漂移过程的空间动态表达.采用GIS专题地图制作技术,结合绿潮业务化特点,研究了绿潮预测产品的业务化制作,从长时间序列数据查询与抽取、多源预报数据综合、标准化报告等角度,实现了绿潮专题图与快报等产品的标准化输出.     

用Fluent与MongoDB构建高效海量日志采集系统

由于传统方法不能满足海量日志存储的需求,对于海量日志的分析传统方法也出现了瓶颈问题,所以本文通过不同类型存储介质对海量日志读写性能实验,并与日志采集软件相结合提出了解决存储海量日志的一种方案:用Fluent(日志采集软件)与MongoDB来构建日志采集系统。实验表明:这种方式与传统存储方式相比,突破单机环境下日志存取的瓶颈,大幅度提高海量日志的存取速度,简化了日志分析与统计操作。在MongoDB集群环境下,高并发的读写满足了海量日志存取的需求,并且这种集群具有可扩展性和高可用性。     

A WWW-based integrated product development platform for sheet metal parts intelligent concurrent design and manufacturing

This paper presents a WWW (World Wide Web) based integrated product development platform for intelligent concurrent design and manufacturing of sheet metal parts. In order to achieve this platform, several major modules are discussed. These modules mainly include the structure of the WWW-based integrated product development platform, an information integration framework, a RTCAPP (real time computer aided process planning) module, a customer interface module, and a design/manufacturing knowledge-based module for supporting product design and manufacturing. This paper gives the structure of the information integration framework for concurrent design and manufacturing of sheet metal parts. An information integration framework, called 'step structure information framework' is proposed. The principles, called 'zero thickness and zero bend radius' are put forward, which can be used to abstract the geometric entities of sheet metal parts in order to facilitate product design and modelling. Finally, case studies are given.     

The value and limitations of using process models to describe the manufacturing organization

Process models offer a systematic, well-denned way of representing the structure of a firm's manufacturing operations. They record the activities that are performed in order to achieve a well-defined purpose of some kind (especially a commercial one), together with the activities' inter-dependencies. In notations such as that of IDEF_o these models have a hierarchical decomposition, in which activities are successively decomposed into more detailed activities, connected by a pattern of constraints of various kinds. Process models can be used to substantiate a number of claims about the satisfactoriness of a firm's operating structures: claims that its activities can be carried out with a greater degree of concurrency, for instance, or that there are redundant or duplicated activities, or spans of control that do not match readily-identified processes. This paper describes a number of observations that were made about the practice of process modelling in an engineering firm. It discusses, in particular, the limited expressive power of process modelling notations and the problems encountered when a normative approach is taken to the analysis of a model-when the analysis refers to some notion of an ideal factory (such as a cellular organization).     

Performance-driven object-oriented program re-modularisation

The aim is to achieve the highest possible speedup when distributing a program across a cluster of computational nodes. The speedup may be achieved by concurrent execution of the distributed modules. In order to maximise the concurrency, a two-stage approach is proposed. In the first stage, a novel statement-reordering algorithm reorders the statements within a given program, to maximise the distance between each call instruction and the instructions, which are data-dependent on any value affected by the call. In the second stage, a clustering algorithm is applied to look for a modular structure of the program which results in the highest concurrency in its execution. The clustering algorithm applies a novel performance evaluation function which is derived from the program code, automatically.     

图案协同设计中的数据处理技术

图案的协同设计是计算机支持的协同工作的一个典型的应用,对数据的并发访问控制是其中心问题。本文基本作者开发的一个基于Ｉｎｔｅｒｎｅｔ的图案协同设计原型系统ＣｏＤｅｓｉｇｎ提出了在协同过程中,将用户对数据的操作行为表述为数据库事务操作的思想,并形式化地研究了表达操作行为的事务模型,较好地解决了协同数据处理问题。     

Model predictive control optimisation using the metaheuristic optimisation for blood pressure control

Given the importance of high blood pressure, it is important to control and maintain a constant blood pressure level in the normal state. The main aim of this article is to design a model predictive controller with a genetic algorithm (GA) for the regulation of arterial blood pressure. The present study is an applied cross‐sectional study. In order to do this research, studies related to designing mathematical models for blood pressure regulation and mechanical models for heart muscle and pressure sensors are investigated. Then, a model predictive controller with GA is designed for blood pressure control. All control and design operations are performed in the MATLAB software. According to the viscoelasticity of blood, transducer, and injection set, we can assume the mechanical model as Mass, Spring, and Damper. Initially, the patient''s blood pressure is lower than normal, and after controlling, the patient''s blood pressure returned to normal. By using a GA‐based model predictive control (MPC), mathematical validation, and mechanical model, the patient''s blood pressure can be adjusted and maintained. The simulation result shows that the GA‐based MPC offers acceptable response and speed of operation and the proposed controller can achieve good tracking and disturbance rejection.     

High coercivity sputter-deposited maghemite thin-film disk

The fabrication and read-write characteristics of a high-coercivity sputter-deposited maghemite (γ-Fe2O3) thin-film disk medium is discussed. By employing a low sputtering gas pressure, severe internal strain is introduced into the films. This internal strain markedly increases coercivity. Furthermore, the films are composed of fine crystallites (300-400 Å in diameter), which result in extremely low media noise. A high coercivity (1060 Oe) γ- (Fe0.954Co0.02Ti0.02Cu0.015)2O3thin-film disk medium with a thickness of 0.095 μm exhibited superior read-write characteristics (e.g., a 2493 FRPM recording density D50and a 42 dB signal-to-noise ratio). These results show that sputter-deposited maghemite films have a promising potential for use as high-density disk media.     

Models and logics for true concurrency

A distributed computer system consists of different processes oragents that function largely autonomously and coordinate their actions by communicating with each other. In such a situation, actions may be performed by different agents of the system locally, in a concurrent manner. In this paper, we first discuss formal models of distributed systems in which concurrency is specifiedexplicitly, in contrast to more traditional approaches where concurrency is representedimplicitly as a nondeterministic choice between all possible sequentializations of concurrent actions. This naturally leads to models based on partially-ordered sets of actions rather than sequences of actions and is often called thetrue concurrency approach. The models we focus on are distributed transition systems, elementary net systems and event structures. In the second half of the paper, we develop a family of logics to specify and reason about the behavioural properties of the models we have described. The logics we define are extensions of temporal logic with new modalities to directly describe concurrency. This paper is essentially a survey of work done by the authors during the last few years on modelling distributed systems with true concurrency and using logic to reason about these models. The emphasis is on motivating definitions through examples and on presenting major results, without going into too many formal details. We provide pointers to the literature where these details can be found.     

Reversible single-molecule switching in an ordered monolayer molecular dipole array

Making electronic devices using a single molecule has been the ultimate goal of molecular electronics. For binary data storage in particular, the challenge has been the ability to switch a single molecule in between bistable states in a simple and repeatable manner. The reversible switching of single molecules of chloroaluminum phthalocyanine (ClAlPc) dipolar molecules within a close-packed monolayer is demonstrated. By pulsing an scanning tunneling microscopy tip, read-write operations of single-molecular binary bits at ~40 Tb/cm(2) (~250 Tb/in(2)) are demonstrated.     

Nanomaterial-based ionic polymer metal composite insect scale flapping wing actuators

Small size actuators (8 mm × 1 mm), IPMNC (RuO₂/Nafion) and IPMNC (LbL/CNC) are studied for flapping at the frequency of insects and compared to Platinum IPMC-Pt. Flapping wing actuators based on IPMNC (RuO₂/Nafion) are modeled with the size of three dragonfly species. To achieve maximum actuation performance with Sympetrum Frequens scale actuator with optimized Young's modulus, the effect of variation of thickness of electrode and Nafion region of Sympetrum Frequens scale actuator is studied. A trade-off in the electrode thickness and Young's modulus for dragonfly size IPMNC-RuO₂/Nafion actuator is essential to achieve the desirable flapping performance.     

制备印迹分子电化学传感器检测食品接触材料中DDM

目的 将分子印迹技术与电化学相结合，实现食品接触材料中4,4''–二氨基二苯甲烷(DDM)的快速检测。方法 采用分子印迹技术，以羧基化碳纳米管(OH–MWCNT)为增敏材料，分别以DDM和吡咯(PPY)作为模板分子和功能单体，在玻璃碳电极(GCE)表面电沉积制备DDM印迹分子薄膜(MIP)，对印迹电极的检测能力使用扫描电镜(SEM)和电化学分析法进行表征。结果 电沉积的DDM电化学传感器具有优良的检测性能和可重复性，其线性范围为10~50 μmol/L，检出限为116 ng/L。结论 该方法具备操作简单、精度高、速度快等优点，能够实现对食品接触材料中DDM的痕量检测。