GIS课程实验论文

基于GIS的城市地震次生火灾危险性分析系统【摘要】众所周知，地震往往会造成严重的灾难，同时也会引发一系列的次生灾害。

从以往的地震来看，它引发的次生灾害，最危险的当属次生火灾。

历史的经验与教训告诉了我们，地震次生火灾有时候造成的破坏比地震本身更巨大，严重威胁着人类的生存。

为了尽量避免地震火灾引发的危害，减少经济损失与人员伤亡，应该针对性地做好地震次生火灾的预防和扑救工作，因此有必要在城市的各个消防区都建立完善的火灾信息系统。

此外，还应该建立相应的危险性分析预估模型、扑救路线的实时搜索模型以及灾后损失的评估模型等，尽量做好事前的防备工作、事后的补救工作。

当前较实用、先进的地震次生火灾危险性分析系统主要是基于地理信息系统（gis）而建立的。

本文通过对城市地震次生火灾危险性分析系统简介、地震次生火灾危险性的分析与模型预估以及城市地震次生火灾危险性分析系统三个方面进行阐述分析，给出了基于gis的地震次生火灾危险性分析系统的模型与方法。

【关键词】gis 地震次生火灾;危险性;分析系统gis，即地理信息系统，它是一门介于空间科学、信息科学与地球科学之间的新技术学科和交叉学科。

它把地学中的空间数据处理同计算机技术结合起来，通过系统地建立、操作以及分析模型，产生一些对区域规划、资源环境、灾害防治、管理决策等方面有用的信息。

近几年，gis已经广泛应用于环境的保护、自然灾害的模拟与预测、自然资源的管理以及相关的灾害应急反应等防灾工程领域中。

关于地震次生灾害研究，大致可以分为两个类别：第一类是采用回归统计的方法进行研究，通过回归统计分析，给出次生火灾发生率同房屋倒塌率的关系式；第二类则用非确定性的概率模型的方法，给出在一定超越概率的条件下次生火灾发生次数的预测值。

从逻辑上来看，采用第二类方法研究不确定性的地震次生火灾是否发生要更为合理些。

1.城市地震次生火灾危险性分析系统简介1.1基本构成地震次生火灾危险性分析系统的构成框架如图1所示，它的基本构成包括：数据的输入、数据的管理与存储、图形的编辑、信息检索和查询、模型的分析以及结果输出等。

浅谈GIS及其在土地管理中的应用摘要：土地管理是一项复杂而系统的工程，gis具有强大的信息管理能力。

将gis应用于土地管理将极大地提高社会效益和经济效益。

笔者对gis进行了介绍，并根据工作实际对gis的应用、存在的问题及发展的趋势进行了探讨。

关键词：数字地球；gis；土地管理；应用；存在的问题；发展的趋势1、地理信息系统（gis）联合国有关文件曾明确指出“地理空间信息是实现可持续发展的利器”，而且“进入信息化社会后，人们每天接触到的信息中的80％以上与地理空间信息有关”。

1998年江泽民总书记在接见两院院士代表的讲话中指出：“当今世界，以信息技术为主要标志的科技进步日新月异，高科技成果向现实生产力的转化越来越快，初见端倪的知识经济预示人类的经济社会生活将发生新的巨大变化”。

顺应经济全球化，全球信息化的发展趋势，为使国土资源工作更好地适应我国可持续发展的需要，国土资源部将”加强信息系统建设，实现信息服务社会化”列为五大目标任务之一，并在新一轮国土资源大调查计划中设立“数字国土”工程，旨在通过信息基础设施建设，加速实现国土资源调查评价、规划、管理、保护和合理利用的现代化，为全社会提供方便快捷的国土资源信息服务，充分发挥国土资源在国家社会经济发展中的基础性、公益性和战略性作用。

数字地球在这方面可以发挥更大的作用。

什么是“数字地球”呢?所谓“数字地球”，可以理解为对真实地球及其相关现象统一的数字化重现和认识。

其核心思想是用数字化的手段来处理整个地球的自然和社会活动等方面的问题，最大限度地利用资源，并使普通百姓能够通过一定方式方便地获得他们所想了解的有关地球的信息，其特点是嵌入海量地理数据，实现多分辨率、三维对地球的描述，即“虚拟地球”。

数字地球的核心是地球空间信息科学，地球空间信息科学的技术体系中最基础和基本的技术核心是3s技术及其集成。

所谓3s是全球定位系统(gps)、地理信息系统(gis)和遥感(rs)的统称。

科技论文写作测绘科学与技术学院地理信息科学13011310050125丁文博2014年11月29日城市地理信息系统原理与应用--在城市规划中的应用摘要随着生活水平的提高和城市化的发展，人们对地理信息的需求激增，信息量急剧膨胀，社会管理日益复杂，对城市管理手段的要求越来越高。

面对有限的空间资源，如何使之产生最大的效益，是城市规划管理面临的共同课题。

在对生存空间不断挑战并设法超越的过程中，城市地理信息系统(UGIS)业已成为强有力的武器。

电子地图、公共场合的多媒体导游导购系统、汽车自动导航装置、卫星定位仪等渐渐多起来。

UGIS知识对每一个现代城市居民都必不可少，而对于政府部门、众多企业和商业机构，则是关系到竞争成败的关键因素。

城市地理信息系统带动的产业正在急剧膨胀，深入市政工程、资源管理、房地产业、交通运输、邮电话信、公安消防、医疗保健、城市应急、企业决策、金融保险、水利电力、环保旅游，以及科研教育的各个方面。

近年来，随着城市化进程的加快，在城市规划和管理领域，传统的规划和管理手段已不能适应城市飞速发展的需要。

城市的发展，必然带来诸如水资源匮乏、用地紧张、交通拥堵、能源不足、环境污染等一系列棘手的城市问题，这给城市建设管理和规划提出了更新、更高的要求。

而信息技术的应用使城市建设管理和规划走上了自动化、定量化、科学化和信息共享的道路，因为信息技术对城市规划和管理的影响最突出、最直接的是空间数据基础设施的建设。

信息技术作为城市信息处理和分析的工具，逐步成为现代城市规划和管理的技术手段。

而城市地理信息系统是超越城市问题传统解决方法的先进手段并作为现代城市必不可少的基础设施，渗透到社会活动的每一细节。

引言城市地理信息系统的英文名称为Urban Geographic Information System，简写为“UGIS”，它是地理信息系统的一个分支，是一种运用计算机硬件、软件和网络技术，实现对城市各种空间、非空间数据的输入、存贮、查询、检索、处理、分析、显示、更新和提供应用。

论文范文：基于GIS和RS技术概述崂山区雨岛效应1 引言近年来，随着改革日趋深入，城市工业化和社会化发展迅速，高楼大厦林立，城市面积扩展，城区规模明显扩大，出现了郊区降水量小于城区的现象，而且这种现象愈演愈烈，这与城市“雨岛效应”存在很大关系。

目前，我国正处在城镇化高峰期，城市建设速度很快，城市人口数量及财富量不断累积，城市规模扩展迅速，雨岛效应导致的洪涝灾害，对生产生活带来诸多不便。

如何才能更好的应对雨岛效应所导致的城市洪涝灾害显得尤为重要，为此需要对雨岛效应有一个更好的了解。

目前国内国外有关城市雨岛效应的监测方法以及研究方法比较少，大多是对导致雨岛效应的因素进行论述以及对雨岛效应产生的危害进行分析，监测一个地区是否存在雨岛效应无论是从技术上还是方法上都没有明确地研究思路。

本文在对崂山区的基本水文气候进行研究的基础上，结合崂山区气温、降水、风速等因素之间的演变趋势，选取适合的雨量观测站，并研究土地覆被及降水的年际变化，城区郊区汛期的降雨变化，验证雨岛效应存在的同时，探究城市化等因素对雨岛效应的影响，在明确雨岛效应的基础上，更好应对雨岛效应。

1.1 研究背景和意义城市是现代社会政治、经济、文化的核心。

改革开放以来，随着城市化社会化经济的迅猛发展，城镇化进程的加快，人口、经济、社会都有极大的发展，大中型城市越来越多。

而随着城市化进程加快，就地表因素而言，地表水泥等覆盖面积增多，绿地等天然植被覆盖面积减少，导致地面反射的太阳辐射能增多；就地表以上而言，城市建筑群无论在数量上还是质量上都有所增加，阻碍了风所带动的热量流动，减少热量的水平扩散。

一系列人为建设的影响，使得城市地区整体或局部温度高于周围地区。

尤其是到了炎热夏天，建筑物反射太阳辐射以及自身热辐射，空调氟利昂及汽车尾气等的排放，会产生大量热量，这些热量的超长排放，超出城市调节能力，并随着热量的上升，在城市上空形成热气流，随着热量堆积，最终导致降水形成，这种效应被称之为“雨岛效应”。

基于组件技术的分布式GIS研究综述摘要：由于空间数据量巨大，网络带宽有限，海量数据的存储与管理已经成为制约gis发展的技术因素，结合计算机科学技术的发展，分布式gis应运而生。

文章介绍了分布式gis的系统架构，基于组件技术的分布式gis的几项关键技术，分布式gis的性能评价以及存在的问题。

关键词：系统架构；分布式web gis；组件技术；com；corba；java中图分类号：g644 文献标识码：a 文章编号：1002-7661（2011）12-021-01一、系统架构该系统的应用模型采用金字塔模型，整个系统按等级不同可以分为县级结点、省级结点、国家级结点，与目前我国现行的行政划分一致，有利于数据的管理与应用。

具体架构的应用模型如下：在上图中处于最底层的为末结点，所有数据都有末结点存放，当处于高级的结点接到数据查询指令时，譬如查询全国的医疗状况，这时系统将指令进行分解成若干个医疗状况查询指令分发给下级结点，最终由下级结点来执行这些指令。

当指令执行完毕，再将执行结果反馈给上级结点，由上级结点对数据进行综合整理，反馈给用户。

二、分布式gis的关键技术目前流行的分布式组件技术模型主要有以下四种：microsoft公司的分布式构件对象模型(com／dcom)、对象管理组织(omg)的公共对象请求代理体系结构(corba)、sun公司的java企业级beans(ejb)／j2ee和现在比较流行的．net环境下的web services组件技术。

下面首先对这四种技术进行介绍，然后进行比较分析。

1、基于dcom和com + 技术的分布式web gisdcom和com + 技术是微软公司在com组件技术基础上推出的一个高级com的运行环境,是windows 2000操作系统的一部分。

dcom 和com + 技术同样也为分布式gis提供了一个分布式基础服务环境,如事务服务、同步服务、安全服务、负载均衡服务等。

基于dcom和com + 技术的分布式web gis系统最大的技术优势在于提供了对象池技术和负载均衡技术。

摘要近年来，地理信息系统（GIS）是储存和处理与地理空间分布有关信息的集合。

在各行各业得到越来越广泛的应用，GIS以其混合数据结构和独特的地理空间分析功能独树一帜在税务系统中也开始广泛的应用，不仅仅表现为提高管理的效率，而且增加了管理的功能。

通过对组件式GIS技术的分析，还比较了WebGIS 和VC++等技术，我们决定选用VC++基于进行GIS的开发企业分布地理信息系统。

本系统借助计算机完成企业的分布的电子地图，首先注册才能获取应用的权限，实现了鹰眼功能，一个地区地图的放大缩小漫游，箭头编号标柱点选框选圆选选择符号图层控制按企业编号，企业名字查找及企业的添加，删除等功能，当然后几项功能只有管理员才可以有使用权限。

当选择选择方式进行选择之后，可以显示出被选中企业的名字，然后你可以点击企业的名字在地图上就会把这个企业的位置显放大显示在中心位置，况且不断的闪烁，还可以显示企业的详细信息关键词：VC++;地理信息系统;电子地图ABSTRACTRecently, geographical information system (GIS) is the combination of savior and operation and information regarding distribution of geographical space. Applied to more and more wide fields of industry, GIS has developed a school of one´s own with its mixed structure of data and the unique annalistic of geographical space, making it start to be employed into tax system. All the above can be approved by improving efficiency of management as well as increasing capability of management. Through annalistic GIS technique of subassembly and comparison to WebGIS and VC++, we decide to adopt VC++ based on geographical information system of enterprise distribution that was projected for second times.By dint of computer, the system complete the electrical map of enterprise distribution which be utilized only after the first registration. It realizes such features as eye of hawk zoom out and zoom in of a city’s map, roam, arrowhead, marks of serial number, dot option, column option, round option, optical symbol, drawing control, giving number according enterprises, seeking a name of enterprise, adding to and delete enterprise, etc. certainly the latter several features only be eligible to the administrators. When choose the optical type and select from them, the selected name of enterprise would be visible, and then click the name of enterprise. As result of this, the name of enterprise will be amplifier and show in the center of a map, flickering continuously, show the detailed information of a company.Owing to the limit of time and condition, I am afraid that the unfinished features that I should have planned to complete in a form of C/S need to realize in the later extending.Key words：VC++;GIS;electrical map目录1 实现图形系统的文档和视图 (1)1.1 组织矢量图形系统的图形元素类 (1)1.2 组织矢量图形系的文档 (3)1.2.1 组织面向对象的文档管理机制 (3)1.2.2 利用MFC摸板创建管理图形元素对象指针的对象 (3)1.2.3 实现矢量图形系统的文档 (3)1.2.4 实现文档的管理功能 (3)1.3 实现矢量图形系统的视图 (5)1.3.1 建立坐标系 (5)1.3.2 实现各类图形元素的绘制功能 (6)1.3.3 实现视图 (6)1.4 各类图形元素几何属性的计算 (7)2 鼠标交互绘图 (7)2.1 用鼠标绘图要解决的主要问题 (7)2.1.1 捕获鼠标操作消息 (7)2.1.2 捕捉所有的鼠标输入 (7)2.1.3 在屏幕上拖动图形 (7)2.1.4 保存图形数据到文档 (7)2.1.5 将图形以实际的形态重画 (7)2.2 交互绘制各类图形元素 (8)3矢量图形系统的操作功能 (9)3.1 增加图形操作菜单 (9)3.2 图形重画 (9)3.3 图形放大和摆动 (11)3.4 重画上屏和重画首屏 (11)3.5 显示全图 (13)3.5.1 各类图形元素的边界矩形计算 (13)3.5.2 实现显示全图功能 (13)3.6 提高矢量图形系统重画速度的基本方法 (14)3.6.1 提高图形重画速度的方法 (14)3.6.2 提高图形元素的绘制速度 (16)4 图形的选中、移动、旋转、删除 (16)4.1 鼠标点选图形元素 (16)4.2 图形移动 (20)4.3 图形旋转 (21)4.3.1 点与点的旋转操作 (21)4.3.2 各类图形元素的旋转操作 (22)4.3.3 实现旋转操作功能 (22)4.4 图形元素的删除 (22)5 数据库应用程序开发技术 (24)5.1 创建基于ODBC的数据库应用程序 (24)5.2 CRecordSet类功能分析 (24)5.3 CRecordView视图类分析 (26)6 实现数据库浏览功能 (27)6.1 创建一个数据库浏览视图 (27)6.2 创建一个CRecordSet派生类对象 (28)6.3 实现数据库浏览试图 (28)7 实现数据库编辑功能 (28)7.1 建立并初始化存储记录指针的变量 (28)7.2 建立数据库编辑功能操作菜单 (29)7.3 建立编辑数据记录的对话框类 (29)7.4 增加和修改数据记录 (29)7.5 删除记录 (29)8 通用数据库过滤功能 (30)8.1 过滤操作的实现方式 (30)8.2 创建组织过滤条件的对话框类 (30)8.3 组织过滤条件编辑器的各种功能 (30)8.4 在数据库浏览视图中实现过滤功能 (31)8.5 增加过滤条件编辑器的功能 (31)参考文献 (33)1 实现图形系统的文档和视图1.1 组织矢量图形系统的图形元素类城市的信息化为城市GIS发展带来了机遇。

图２—１ＧＩＳ技术发展里程为解决集成式ＧＩＳ与模块化ＧＩＳ的缺点，ＧＩＳ和计算机领域的专家们提出了核心式ＧＩＳ（ＣｏｒｅＯｌＳ）的概念。

核心式ＧＩＳ被设计为操作系统的基本扩展。

Ｗｉｎｄｏｗｓ系列操作系统上的核心式ＧＩＳ提供了一系列动态连接库∞ＬＬ），开发ＧＩＳ应用系统时可以采用现有的高级编程语言，通过应用程序接ｔ］（ＡＰＩ）访问内核所提供的ＧＩＳ功能。

除了一些基本的动态连接库以外，实现各种功能的动态连接库可以被拆卸和重组，它提供了动态连接库一级的更底层的组件化方式，给用户提供更大的灵活性。

对数据库管理要求较多的用户甚至可以选择ＭＩＳ开发工具来构造ＧＩＳ应用，为ＧＩＳ与ＭＩＳ的无缝集成提供了全新的解决思路。

但是，由于核心式ＧＩＳ提供的组件过于底层，给应用开发者带来一定难度，也不适应可视化程序设计的潮流。

随着计算机软件技术的发展，ＧＩＳ组件化发展到了一个全新的阶段，出现了组件式ＧＩＳ（ＣｏｍｐｏｎｅｎｔｓＧＩＳ，缩写为ＣｏｍＧＩＳ）。

组件式ＧＩＳ基于标准的组件式平台，各个组件之间不仅可以进行自由、灵活的重组，而且具有可视化的界面和使用方便的标准接口。

组件式平台主要有Ｍｉｃｒｏｓｏｆｔ的ＣＯＭ（ＣｏｍｐｏｎｅｎｔＯｂｊｅｃｔＭｏｄｅｌ，组件对象模塑．Ａ）／ＤＣＯＭ（ＤｉｓｔｄｂｕｔｅｄＣｏｍｐｏｎｅｎｔＯｂｊｅｃｔＭｏｄｅｌ，分布式组件对象模型）和ＯＭＧ的ＣＯＲＢＡ（ＣｏｍｍｏｎＯｂｊｅｃｔＲｅｑｕｅｓｔＢｒｏｋｅｒＡｒｃｈｉｔｅｃｔｕｒｅ，公共对象请求代理体系结构），目前Ｍｉｃｒｏｓｏｆｔ的ＣＯＭ／ＤＣＯＭ占市场领导地位。

基于ＣＯＭ／ＤＣＯＭ，Ｍｉｃｒｏｓｏｆｔ推出了ＡｃｔｉｖｅＸ技术，ＡｃｔｉｖｅＸ控件是当今可视化程序设计中应用最为广泛的标准组件。

新一代的组件式ＧＩＳ也大都是ＡｃｔｉｖｅＸ控件或者其前身ＯＬＥ控件。

组件式ＧＩｓ代表着当今ＧＩＳ发展的潮第１０页信息工程大学硕士学位论文第五章应用案例研究本文选取黄河段硫化物污染、丹江口多湖库ＢＯＤ污染、郑州市地下水污染分别作为河流、湖泊（水库）和地下水污染的案例进行水质污染模拟研究。

GIS应用实验报告1. 引言地理信息系统（GIS）是一种用于收集、存储、分析和管理地理和空间数据的技术。

在本实验中，我们将探索GIS的应用，并学习如何利用GIS进行地理数据分析和空间可视化。

2. 实验目标本次实验的主要目标是了解GIS的基本概念和原理，并学会使用GIS软件进行地理数据的处理和分析。

具体的实验目标包括： 1. 学习如何导入地理数据到GIS 软件中； 2. 掌握地理数据的符号化和渲染技巧； 3. 学会使用GIS软件进行空间查询和空间分析； 4. 学习如何制作地图和空间可视化。

3. 实验步骤步骤1：准备工作在开始实验之前，我们需要安装一个GIS软件。

这里我们选择使用QGIS，一个功能强大且免费的开源GIS软件。

在安装完成后，我们可以打开软件并开始实验。

步骤2：导入地理数据首先，我们需要导入一些地理数据到GIS软件中，以便后续的分析和可视化。

可以从官方的地理数据库或者其他来源下载一些地理数据集，如道路网络、土地利用数据等。

将这些数据导入到GIS软件中，并确保数据被正确地显示在地图上。

步骤3：符号化和渲染地理数据符号化是将地理数据表示为图形符号或颜色的过程，以便更好地理解和分析。

我们可以根据数据的属性值对地理要素进行符号化，并选择合适的颜色和图标来表示不同的属性。

通过符号化，我们可以直观地看到不同地理要素的分布和特征。

步骤4：空间查询和空间分析使用GIS软件，我们可以进行各种空间查询和空间分析。

例如，我们可以选择特定的地理区域，并查询该区域内的特定属性值。

我们还可以进行空间分析，如缓冲区分析、叠加分析等，以获得更深入的地理信息。

步骤5：制作地图和空间可视化制作地图是GIS的一个重要应用。

我们可以根据需要选择不同的地图样式和要素，将地理数据以地图的形式展示出来。

在制作地图时，我们可以调整地图的比例尺、添加图例和注释等，以使地图更加清晰和易于理解。

4. 实验结果和讨论在完成实验步骤后，我们可以得到一些地理数据的分析结果和地图可视化。

gis的实验心得gis的实验心得(3篇)当我们受到启发，对生活有了新的感悟时，就十分有必须要写一篇心得体会，这样可以不断更新自己的想法。

到底应如何写心得体会呢？以下是小编为大家整理的gis的实验心得(3篇)，欢迎大家借鉴与参考，希望对大家有所帮助。

gis的实验心得(3篇)1随着GIS实习的结束我们也学了不少东西。

本次实习为期一周，要求我们通过运用软件ArcView制作广东省人口密度和产业结构专题图，了解和掌握利用GIS软件进行专题图制作的方法和技巧，巩固《GIS软件》课程的所学内容，提高综合应用能力和创造能力。

主要任务为制作广州市十区(荔湾、越秀、海珠、天河、白云、黄埔、番禺、花都、南沙和萝岗)两市(增城和从化)的生产总值和财政收入专题图。

刚开始时困难很多，主要是不知怎么开始，经过我们组员的讨论和请教，大家都知道一点点了。

随着时间的过去，同学们都在摸索之中，任务也一点点的在进展中，最后大家都很努力地完成了。

在实习中我们都巩固了课本学到的知识，也熟练了运用GIS软件进行专题制图的方法和技巧。

另外，也体体会到了合作完成任务的喜悦。

这既包括有个人的操作技能实习也有组员合作的成果。

实习让我更加的加深和巩固运用GIS软件进行专题制图的方法和技巧。

我们通过上网搜索资料后在用GIS软件进行修改。

编辑点主题，编辑线主题，编辑多边形主题，文字标注等。

图形数据的输入利用鼠标进行，也可以借助数字化仪进行。

前者通常被称为屏幕数字化。

所创建的主题分为点主题、线主题和面主题3种。

主题被创建以后，ARCVIEW会为主题自动生成主题属性表。

主题属性表是图形数据与表格数据(属性数据)相联系的纽带。

ARCVIEW会为主题自动生成的主题属性表最初仅包含少量的缺省字段，以后可以通过编辑该属性表来添加字段，从而实现属性数据的连接。

ArcView的地图编辑，主要是通过图例编辑器来实现的，它可以制作能向地图用户表达数据重要信息的可视化地图。

Demand-Based Tree: Replica Update Propagationfor Weak Consistency in the Grid DatabaseRuixuan Ge1, Yong-Il Jang1, Ho-Seok Kim1, Jae-Dong Lee2, Hae-Young Bae1 1Dept. of Computer Science & Information Engineering, INHA Univ.2Dept. of Computer Science, Dan Kook Universitygeruixuan@dblab.inha.ac.kr, yijang@dblab.inha.ac.kr, bluesnow@dblab.inha.ac.krletsdoit@, hybae@inha.ac.krAbstractSince some replicas will be in greater demand than others, not all replicas can be treated in the same way. To satisfy requests as more as possible first in a shorter period of time, a fast consistency algorithm has been presented, whereas the poor performance is showed in multiple regions of high demand for forming the island of locally consistent replicas. Although a leader election method is proposed, much additional cost for periodic leader election, information storage, and message passing is required. Also, false leader can be created. In this paper, a tree-based algorithm for replica update propagation is proposed. Leader replicas with high demand, which are considered as the roots of tree, are interconnected by pointers. All the other replicas are sorted by geographical distribution and considered as nodes of the trees. Once an update occurs at any replica, the update should be propagated to the leader replicas first. Every node propagates the update to all its children in the tree. The demand-based tree is flexible for the dynamic model in which the demand conditions do change with time. And the replica propagation is optimized by cost reduction for fixed propagation schedule.1. IntroductionGrid is an important new flat roof for net computing. Database system is required for managing the large amounts of data on the grid. The Grid Database is the outcome of inter-combination of Grid and Database technique. It becomes one of the most important resources to provide data management service in the Grid environment.Data Replication is a significant aspect in Grid Database [6]. All the data can have many same replicas stored in other nodes in Grid Database environment. Replication can reduce the access delay and the bandwidth consumption. Accordingly, the speed can be increased and the efficiency can be improved. Replica consistency is one of the key issues in replica application. Currently, there are two groups of updating algorithms for replica consistency. One is strong consistency, the other is weak consistency.Strong consistency must ensure that all the replicas are always in a consistent state. But it is very impractical for Grid Database system. In contrast, weak consistency, which provides less reply time and higher availability, is more doable. When an update occurs on one replica, it can be propagated to other nodes by some order. This approach tolerates impermanent inconsistency among replicas. But, after a certain period of time, the replicas should be in a consistent state assuredly. Accordingly, to achieve weak consistency, an efficient method should be provided for update propagation [1, 10].Several update propagation methods have been presented to implement replica weak consistency. One is demand based method [7]. A demand-based algorithm for fast update propagation of replicas has been brought forward. But the stability of this algorithm is poor even with leader election. And the additional overhead for sending message is considerable.In this paper, a tree-based algorithm for replica update propagation is proposed. Leader replicas with high demand are considered as the roots of trees which are interconnected. All the other replicas are sorted by geographical distribution and considered as nodes of the trees. Once an update occurs at any replica, it should be transmitted to the leader replicas first. Each node that receives the update propagates it to its children in the tree. Our proposed method optimizes the update propagation for weak consistency by cost reduction for fixed propagation schedule. In this paper,it can be seen that our algorithm performs well not only in static model but also in dynamic model by performance evaluation.The rest of this paper is organized as follows. Section 2 gives an overview of previous work on replica update propagation. Section 3 presents our proposed demand-based trees model and algorithms. In section 4, we evaluate our method and describe the results. And the last section is the brief conclusions of this paper.2. Related WorkThis chapter generalizes several existing researches about replica consistency and stresses describing the fast consistency algorithm.2.1. Replica Consistency Service in Grid EnvironmentGrid system provides supports for global business, government, research, science and corporation as a basic establishment of data and computation resources management. Replica consistency service deals with keeping replicas up to date in the grid system.A Grid Consistency Service (GCS) is proposed to synchronize replication update and maintain consistency in Data Grid [5]. And the data consistency is partitioned to several levels to adapt different requirements. In some levels, not all the replicas are always up to date. Various required locks over grid can be used to achieve different replica consistency.Andrea Domenici generalized the emphases in replica consistency and presented how to design the replica consistency service in [3]. The component of the consistency service is described briefly. Then, he proposed a relaxed data consistency with a replica consistency service called CONStanza in Data Grid [4]. This method updates replicas asynchronously by way of changing the frequency of checking database modifications. It can satisfy database replication as well as file replication. It can keep the remote replicas lazy consistency.2.2. The Fast Consistency AlgorithmUpdate propagation is an important part for replica consistency service. To satisfy more requests from clients with up-to-date data, the demand based approach should be taken into account.Elias proposes a “Fast Consistency” algorithm which prioritizes the replicas by the demand in the distributed system [7]. This algorithm is that all the servers select the neighbor with the most demand to start a consistency session. If the neighbor has another neighbor with greater demand, the process will be repeated. This process continues until all the replicas are updated. This algorithm guarantees that most requests can be satisfied in few sessions efficiently.Then, Elias analyzes behavior of this algorithm in a collection of replicas with multiple zones with high demand [8]. From this research, we can see that the fast consistency algorithm performs very well in a collection of replicas with only one high demand zone. However, in multiple zones of high demand, the performance becomes poor. The low demand replicas can slow down the update propagation as barriers.In [9], Elias proposes a leader election algorithm to avoid these barriers with low demand and generalizes this algorithm to Grid system. By this way, it can be considered that all the zones with high demand combine to form one high demand zone. But this algorithm may choose false leader. And an additional cost that the table of the IDs of leader nodes needs to be dynamically reconstructed periodically is considerable.Furthermore, in the dynamic model, before any update propagation process performs in a replica, the chart with its neighbors’ data must to be updated. That means, the replica need ask all its neighbors before every update process by sending messages. This problem also causes much additional overhead.3. Replica Update Propagation Using Demand-Based TreeIn this chapter, we introduce a demand-based tree structure for update propagation. The propagation algorithm and dynamic algorithm are presented also.3.1. Demand-Based Tree StructureIn the grid database system, we choose some replicas with high demand value as leader nodes. Every leader node can be considered as the root of a tree. Other replicas are sorted by finding the nearest leader node. Each set of these replicas aligns by the sequence of descent demand value. Based on this sequence, these replicas can compose a tree. The leader nodes are interconnected by a bidirectional pointer by the sequence of descent demand value.We consider a node as a candidate leader node when its demand is more than all its neighbors. Then, we define the average demand of a node d/n as a threshold, where d is the total demand of the network and n is the amount of replica nodes in the network. The candidates whose demand value is more than the threshold are chosen as leader nodes, others areAll the other replicas are sorted by finding theEach set of nodes including a leader node queues by the sequence of descent demand value. So, every node has a serial number as 0, 1, 2, 3…by the sequence. Obviously, the serial number of leader node is 0. Figure3 shows an example for the distribution ofThe first set of nodes queue as the sequence for example in Figure4. Other nodes queue as the same.a a 1a 2a 3a 4a 5a 6a 7a 8a 9a 10a 1101234567891011Figure 4. A set of nodes queue by the sequence of descentdemand valueEach set of nodes can compose a tree in whichevery node whose serial number is i . Every node hasthe children whose serial number is , where c is the upperlimit amount of children of a node in the trees. Andevery replica knows the information of its leader node,parent and children. The data structure is shown inFigure5 (a).c i c i c i c +×+×+×,,2,1"How to decide the upper limit amount of children ofa node in the trees? It lies on the amount of replicanodes in the network and the amount of leader nodes. The amount of nodes in every tree may be different. So, the height of every tree may be different. We define the minimum integer value h , which can satisfy:n c l hj j ≥⋅∑=0as the average height of the trees. Here, l is the amount of leader nodes. Based on the propagation algorithm, because we need to satisfy all the requests to clients in less time,the value h c + needs to be minimal. If there are several pairs of value <c, h> satisfied this condition, we choose the pair in which c is minimal for load balancing. Then, c is an approximate optimal value. In our example, the value pair <2, 4> and <3, 3> have the minimal value for . We choose the pair h c +<2, 4> as the solution. Figure5 (b) shows the data for node b1.b b b3b4(b)Figure 5. (a) Data structure for tree-based structure.(b) The data for node b1The leader nodes are interconnected by a bidirectional pointer by the sequence of descent demand value. So, the leader node also knows the information of its neighbor leader nodes in the structure. The data structure and the data for node a as an example are shown in Figure6. The tree-based structure and the example are shown in Figure7.Forward Pointer Backward Pointer b Null(a) (b)Figure 6. (a) The data structure about neighbors of a leader(b) The data about neighbors of node a………………Figure 7. The tree-based structureWhen an update occurs at a node, it will send a message to its parent initially. It can ensure that its parent won’t propagate the update to itself. Then, it will transmit the update to its leader node first. The leader node, who has received the update, transmits the update to its neighbor leader node with greater demand by the forward pointer first. Next, it transmits the update to another leader node by the backward pointer. Afterward, it transmits the update to its children by the sequence of descent demand value. Every leader node transmits the update to other nodes by this sequence. Every non-leader nodes also transmits the update to its children by the sequence of descent demand value. Figure8 shows the update propagation in our example when the update occurs in node b3.Figure 8. The example of the update propagationIn fact, the demand conditions do change with time. Accordingly, the tree structure need do change with time too. When the demand of a node increases, it should compare with its parent. If its demand is greater than its parent’s, their position needs to be exchanged. Repeat this step, until its demand is less than its parent’s. In our example, we assume some nodes with their corresponding demand as shown in Table 1. If the demand of node b7 increases to 24, it should compare with node b3 first. So, their position needs to be changed. Secondly, it will compare with node b1. Their position should be changed also. Thirdly, it will compare with node b. Its demand value is less than the demand of node b. Therefore, the result is shown as Figure9 (a).Table 1. The demand of replica nodesb b 1b 2b 3b 4b 5b 6b 7b 8b 9b 10b 112523222018151312111086b 12b 13b 14b 155421a 20Replica DemandWhen the demand of a node decreases, it should compare with its children. If its demand is less than any of its children’s, it will exchange the position with its child with most demand. Repeat this step, until its demand is greater than its children’s. In our example, if the demand of node b 2 falls from 22 to 14, it will compare with his child b 5 that has most demand among its children. So, it should change the position with b 5. Next, it will compare with node b 11. Its demand is greater than the demand of b 11. Consequently, the result is shown as Figure9 (b).If any leader node finds that its demand value is more than anterior one or less than the posterior one, the position of these two trees need to be changed. In our example, we assume the demand of node a increases from 20 to 27, which is more than b and less than c , the tree including node aneeds change the position with the tree including node b . The result is shown in Figure9 (c).When the demand of a node does change, its information about its parent should be modified. In our example, if the demand of b 4 increases to 19, the data for node b 1 is shown as Figure9 (d).(c)b b b 4b 3(d)Figure 9. The result of the dynamic algorithm used in ourexample3.2. Propagation AlgorithmFigure10 shows the update propagation algorithm after updating a replica.Input:current_node: Current node IDupdate_data: The update need to be propagated Variables:FP: Forward Pointer of current node BP: Backward Pointer of current node child(i ): the children of current node Begin01: if current_node is a leader 02: if FP is not null03: propagate update_data to FP 04: endif05: if BP is not null06: propagate update_data to BP 07: endif08: for i ← 1 to c && child(i ) is not null 09: propagate update_data to child(i ) 10: else11: send a message to parent12: propagate update to parent13: for i ← 1 to c && child(i ) is not null 14: propagate update_data to child(i )15: endifendFigure 10. The update propagation algorithmIf the current node is a leader, and its forward pointer is not null, the update should be propagated to the previous neighbor leader node (Line01~04) (See Figure11 (a)). And if its backward pointer is not null, the update should be propagated to the following neighbor leader node (Line05~07) (See Figure11 (b)). Then, it transmits the update to its children one by one (Line08~09) (See Figure11 (c)). If the current node is not a leader node, it should send a message to its parent initially (Line10~11) (See Figure11 (d)). Then, it propagates the update to its leader first (See Figure11 (e)). Finally, it transmits the update to its children one by one (Line12~15) (See Figure11 (f)).……(a)……(b)……(c)Figure 11. Illustration of the propagation algorithm 3.3. Dynamic AlgorithmFigure12 describes the dynamic algorithm when the demand of a leader replica changes.Input:current_node: Current node IDVariables:FP: Forward Pointer of current nodeBP: Backward Pointer of current nodecurrent_demand: The current demand of current node previous_demand: The previous demand of current node first_child: The first child of current node that may exchange the position with current nodeexchange_child: The child of current node that may exchange the position with current nodec_parent: The parent of current nodeBegin01: if current_node is a leader02: if current_demand > previous_demand03: while FP is not null and crrent_demand is more than the demand of FP04: FP ← ExchangeLeader (FP, current_node) 05: else06: exchange_child ← first_child07 while current_demand is less than the demandof exchange_child 08:exchange_child←Exchange(current_node , exchange_child)09: if exchange_child ≠ first_child10: while BP is not null and the demand offirst_child is less than the demand of BP11: BP ← ExchangeLeader (first_child, BP) 12: else13: while BP is not null and current_demand isless than the demand of BP14: BP ← ExchangeLeader (current_node,BP)15: endif 16: endif 17: endif endFigure 12. The dynamic algorithm for leader nodeIf the demand of a leader node increases and is greater than the previous leader node, the position of the two leader nodes with the trees should be exchanged. This step should repeat until the demand of current node is less than the previous one (Line01~04) (See Figure13 (a)). ExchangeLeader function is used for exchanging the position of two neighbor leaders. If the demand of a leader node decreases and is less than any of its children, it should exchange the position with its first child. This step should repeat until the demand of current node is more than all its children (Line05~08) (See Figure13 (b)). Exchange function can exchange the position of two replicas with different height. Then, compare the demand of the leader after previous steps with the following leader node. If its demand is less than the following one, their position should be exchanged. This step should repeat until the demand of the leader node is more than the following one (Line09~17) (See Figure13 (c)).(b)Figure 13. Illustration of the dynamic algorithm for leadernodeFigure14 is the presentation of the dynamic algorithm when the demand of a non-leader replica changes.Input:current_node: Current node ID Variables:current_demand: The current demand of current node previous_demand: The previous demand of current node c_parent: The parent of current nodec_child: The first child of current nodeBegin01: if current_node is not a leader02: if current_demand > previous_demand03: while current_demand is greater than the demand of the demand of c_parent04: c_parent ←Exchange (c_parent, current_ node)05: else06: while current_demand is less than the demand of the demand of c_child07: c_child ←Exchange (current_nodet, c_child)08: endif09: endifendFigure 14. The dynamic algorithm for non-leader nodeIf the demand of a non-leader node increases, it should exchange the position with its parent (Line01~04). Contrarily, it should exchange the position with its first child (Line05~09). These steps also should be executed circularly until the node is on the proper position.4. Performance EvaluationThe evaluation environment and the comparison between the fast consistency and the demand-based tree are presented in this chapter.4.1. Evaluation EnvironmentThe Network Simulator (NS-2), which is an object oriented simulator, is used to provide the simulation of the network [11]. In NS-2, the whole simulation is driven by the discrete event.BRITE [2] which is a universal topology generation tool is used to generate the topologies in our simulations. This tool generates the topologies randomly and it is inclusive, flexible, extensible and efficient.The fast consistency algorithm and the demand-based tree algorithm are compared in the dynamic model, which is more similar to the real network. The simulations are implemented with the replicas with the number from 10 to 210. Table 2 shows the evaluation environment for testing the proposed method.Table 2. Evaluation Environment (ut: unit time)Evaluation Element Data RangeSimulation Time 5000 (ut)Number of Nodes 10~210Number of Clients 10~400Transmission Time 2~4 (ut)Update Processing Time 6~8 (ut)4.2. Evaluation ResultsIn the evaluation test, the demand of replicas changes randomly.Figure15 shows the comparison of time for complete consistency between the fast consistency with leader and the demand-based tree method. We can see that the network can achieve the consistency state with less time by using the proposed method than fast consistency method.Figure 15. Time for Complete ConsistencyThe reason is that the proposed method only exchange messages when the demand of any replica changes. The message passing between two nodes will not affect the update propagation process. Whereas fast consistency need to send messages before each update propagation carries out for updating the chart with the data of its neighbors. So, fast consistency method needs exchanging much more additional messages for communication between replicas than demand-based tree method, even with leader election. For this reason, both the time and the overhead of message passing in fast consistency method are more than demand-based tree method.The comparison of the number of overhead messages is shown in Figure16.Figure 16. Number of Overhead MessagesExcept the foregoing words about the good performance of our method, it is proved that the demand-based tree approach will not choose false leader from the evaluation. But fast consistency with leader can choose false leaders. The reason is that the votes for the election method can reach to the replicas with low demand whose neighbors have less demand. 5. ConclusionIn Grid Database, the data can be updated by user at any time. Therefore, replica consistency is an important issue obviously. The demand based approach for replica update propagation is one of the methods to maintain replica weak consistency. And it can satisfy more requests more in time.In this paper, the demand-based tree for replica update propagation method is proposed. The tree construction is composed based on the demand of replicas. Update on one replica can be propagated to its parent, children and the leader by this construction. Compared with fast consistency, the proposed method can achieve the consistency state with less time than the fast consistency method. It will not select false leader. It requires just little message passing. Furthermore, periodic leader election is unnecessary. The dynamic algorithm can ensure that the set of leader replicas keeps up to date. Consequently, compared with fast consistency, the overhead is reduced considerably compared with fast consistency in our method. In a word, the proposed method showed increased performance than fast consistency.This method is suitable for Grid Database. It can also be used for various applications with large number of nodes, such as distribution system, and so on. References[1] A. Adya, “Weak Consistency: A Generalized Theoryand Optimistic Implementations for DistributedTransactions,” PhD thesis Massachusetts Institute ofTechnology, Department of Electrical Engineering andComputer Science, March 1999[2] A. Medina, A. Lakhina, I. Matta, and J. Byers, “BRITE:Universal Topology Generation from a User’sPerspective,” Technical Report BUCS-TR2001 -003,Boston University, 2001[3]Andrea Domenici, Flavia Donno, Gianni Pucciani,Heinz Stockinger, Kurt Stockinger, “Replicaconsistency in a Data Grid,”/~kurts/research/acat2003_replication.pdf, July 2004[4]Andrea Domenici, Flavia Donno, Gianni Pucciani,Heinz Stockinger, “Relaxed Data Consistency withCONStanza,” Proceedings of the Sixth IEEEInternational Symposium on Cluster Computing and theGrid (CCGRID’06), 2006[5] D. Du¨ llmann, W. Hoschek, J. Jaen-Martinez, A.Samar, H.Stockinger, K. Stockinger, “Models forreplica synchronization and consistency in a data grid,”in: Tenth IEEE Symposium on High Performance andDistributed Computing (HPDC-10), San Francisco, CA,August 7–9, 2001[6]Houda Lamehamedi, Boleslaw Szymanski, ZujunShentu, Ewa Deelman, “Data Replication Strategies inGrid Environments,” Proc. Of the Fifth InternationalConference on Algorithms and Architectures forParallel Processing (ICA3PP'02), 2002[7]Jesús Acosta Elias, Leandro Navarro Moldes, “ADemand Based Algorithm for Rapid Updating ofReplicas,” IEEE Workshop on Resource Sharing inMassively Distributed Systems (RESH’02), July 2002[8]Jesús Acosta Elias, Leandro Navarro Moldes,“Behaviour of the fast consistency algorithm in the setof replicas with multiple zones with high demand,”Symposium in Informatics and Telecommunications,SIT2002[9]Jesús Acosta Elias, Leandro Navarro Moldes,“Generalization of The Fast Consistency Algorithm Toa Grid With Multiple High Demand Zones,”Computational Science, ICCS, 2003[10]K. Petersen, M. J. Spreitzer, D. B. Terry, M. M.Theimer, and Demers, “Flexible Update Propagationfor Weakly Consistent Replication,” Proceedings of the16th ACM Symposium on Operating SystemsPrinciples (SOSP-16), Saint Malo, France, October 5-8,1997, pages 288-301[11]he Network Simulator – ns-2,/nsnam/ns/。