Jianwen的共享空间: 2008年5月

Jianwen 的个人资料Jianwen的共享空间照片日志列表更多

工具

帮助

<< 第一页 < 上一页下一页 > 最后一页 >>

经典悖论漫谈

泽熙

古今中外有不少著名的悖论，它们震撼了逻辑和数学的基础，激发了人们求知和精密的思考，吸引了古往今来许多思想家和爱好者的注意力。解决悖论难题需要创造性的思考，悖论的解决又往往可以给人带来全新的观念。
本文将根据悖论形成的原因，粗略地把它归纳为六种类型，分上、中、下三个部份。这是第一部份：
由概念自指引发的悖论和引进无限带来的悖论
(一)由自指引发的悖论
以下诸例都存在着一个概念自指或自相关的问题：如果从肯定命题入手，就会得到它的否定命题；如果从否定命题入手，就会得到它的肯定命题。

1-1谎言者悖论
公元前六世纪，哲学家克利特人艾皮米尼地斯(Epimenides)：“所有克利特人都说谎，他们中间的一个诗人这么说。”这就是这个著名悖论的来源。
《圣经》里曾经提到：“有克利特人中的一个本地中先知说：‘克利特人常说谎话，乃是恶兽，又馋又懒’”(《提多书》第一章)。可见这个悖论很出名，但是保罗对于它的逻辑解答并没有兴趣。
人们会问：艾皮米尼地斯有没有说谎？这个悖论最简单的形式是：“我在说谎”。
1-2“我在说谎”
如果他在说谎，那么“我在说谎”就是一个谎，因此他说的是实话；但是如果这是实话，他又在说谎。矛盾不可避免。它的一个翻版：“这句话是错的。”
1-3“这句话是错的”
这类悖论的一个标准形式是：如果事件A发生，则推导出非A，非A发生则推导出A，这是一个自相矛盾的无限逻辑循环。拓扑学中的单面体是一个形像的表达。
问题并不简单
哲学家罗素曾经认真地思考过这个悖论，并试图找到解决的办法。他在《我的哲学的发展》第七章《数学原理》里说道：“自亚里士多德以来，无论哪一个学派的逻辑学家，从他们所公认的前提中似乎都可以推出一些矛盾来。这表明有些东西是有毛病的，但是指不出纠正的方法是什么。在1903年的春季，其中一种矛盾的发现把我正在享受的那种逻辑蜜月打断了。”
他说：谎言者悖论最简单地勾画出了他发现的那个矛盾：“那个说谎的人说：‘不论我说什么都是假的’。事实上，这就是他所说的一句话，但是这句话是指他所说的话的总体。只是把这句话包括在那个总体之中的时候才产生一个悖论。”(同上)
罗素试图用命题分层的办法来解决：“第一级命题我们可以说就是不涉及命题总体的那些命题；第二级命题就是涉及第一级命题的总体的那些命题；其余仿此，以至无穷。”但是这一方法并没有取得成效。“1903年和1904年这一整个时期，我差不多完全是致力于这一件事，但是毫不成功。”(同上)
《数学原理》尝试整个纯粹的数学是在纯逻辑的前提下推导出来的，并且使用逻辑术语说明概念，回避自然语言的歧意。但是他在书的序言里称这是：“发表一本包含那么许多未曾解决的争论的书。”可见，从数学基础的逻辑上彻底地解决这个悖论并不容易。
接下来他指出，在一切逻辑的悖论里都有一种“反身的自指”，就是说，“它包含讲那个总体的某种东西，而这种东西又是总体中的一份子。”这一观点比较容易理解，如果这个悖论是克利特以为的什么人说的，悖论就会自动消除。但是在集合论里，问题并不这么简单。
1-4理发师悖论
在萨维尔村，理发师挂出一块招牌：“我只给村里所有那些不给自己理发的人理发。”有人问他：“你给不给自己理发？”理发师顿时无言以对。
这是一个矛盾推理：如果理发师不给自己理发，他就属于招牌上的那一类人。有言在先，他应该给自己理发。
反之，如果这个理发师给他自己理发，根据招牌所言，他只给村中不给自己理发的人理发，他不能给自己理发。
因此，无论这个理发师怎么回答，都不能排除内在的矛盾。这个悖论是罗素在一九○二年提出来的，所以又叫“罗素悖论”。这是集合论悖论的通俗的、有故事情节的表述。显然，这里也存在着一个不可排除的“自指”问题。
1-5集合论悖论
“R是所有不包含自身的集合的集合。”
人们同样会问：“R包含不包含R自身？”如果不包含，由R的定义，R应属于R。如果R包含自身的话，R又不属于R。
继罗素的集合论悖论发现了数学基础有问题以后，1931年歌德尔(KurtGodel，1906-1978，捷克人)提出了一个“不完全定理”，打破了十九世纪末数学家“所有的数学体系都可以由逻辑推导出来”的理想。这个定理指出：任何公设系统都不是完备的，其中必然存在着既不能被肯定也不能被否定的命题。例如，欧氏几何中的“平行线公理”，对它的否定产生了几种非欧几何；罗素悖论也表明集合论公理体系不完备。
1-6书目悖论
一个图书馆编纂了一本书名词典，它列出这个图书馆里所有不列出自己书名的书。那么它列不列出自己的书名？这个悖论与理发师悖论基本一致。
1-7苏格拉底悖论
有“西方孔子”之称的雅典人苏格拉底(Socrates，公元前470-前399)是古希腊的大哲学家，曾经与普洛特哥拉斯、哥吉斯等著名诡辩家相对。他建立“定义”以对付诡辩派混淆的修辞，从而勘落了百家的杂说。但是他的道德观念不为希腊人所容，竟在七十岁的时候被当作诡辩杂说的代表。在普洛特哥拉斯被驱逐、书被焚十二年以后，苏格拉底也被处以死刑，但是他的学说得到了柏拉图和亚里斯多德的继承。
苏格拉底有一句名言：“我只知道一件事，那就是什么都不知道。”这是一个悖论，我们无法从这句话中推论出苏格拉底是否对这件事本身也不知道。古代中国也有一个类似的例子，那就是“言尽悖”。
1-7“言尽悖”
这是《庄子·齐物论》里庄子说的。后期墨家反驳道：如果“言尽悖”，庄子的这个言难道就不悖吗？再看看我们常说的：“世界上没有绝对的真理”。我们不知道这句话本身是不是“绝对的真理”。
1-8“荒谬的真实”
有字典给悖论下定义，说它是“荒谬的真实”，而这种矛盾修饰本身也是一种“压缩的悖论”。悖论(paradox)来自希腊语“para+dokein”，意思是“多想一想”。
这些例子都说明，在逻辑上它们都无法摆脱概念自指所带来的恶性循环。有没有进一步的解决办法？在下面一节的最后一部份还将继续探讨。
(二)引进无限带来的悖论
《墨子·经说下》中有一句话：“南方有穷，则可尽；无穷，则不可尽。”如果在有限中引进无限，就可能引起悖论。
2-1阿基里斯悖论
稍晚于毕达哥拉斯的古希腊数学家芝诺(ZenoofElea)，曾经提出过一些著名的悖论，对以后数学、物理概念产生了重要影响，阿基里斯悖论是其中的一个。
阿基里斯(Achilles)是希腊神话中善跑的英雄。芝诺讲：阿基里斯在赛跑中不可能追上起步稍微领先于他的乌龟，因为当他要到达乌龟出发的那一点，乌龟又向前爬动了。阿基里斯和乌龟的距离可以无限地缩小，但永远追不上乌龟。
有人用物理语言描述这个问题说，在阿基里斯悖论中使用了两种不同的时间度量。一般度量方法是：假设阿基里斯与乌龟在开始时的距离为S，速度分别为V1和V2。当时间T=S/(V1-V2)时，阿基里斯就赶上了乌龟。
但是芝诺的测量方法不同：阿基里斯将逐次到达乌龟在前一次的出发点，这个时间为T'。对于任何T'，可能无限缩短，但阿基里斯永远在乌龟的后面。关键是这个T'无法度量T=S/(V1-V2)以后的时间。
2-2二分法悖论
这也是芝诺提出的一个悖论：当一个物体行进一段距离到达D，它必须首先到达距离D的二分之一，然后是四分之一、八分之一、十六分之一、以至可以无穷地划分下去。因此，这个物体永远也到达不了D。这些结论在实践中不存在，但是在逻辑上无可挑剔。

芝诺甚至认为：“不可能有从一地到另一地的运动，因为如果有这样的运动，就会有‘完善的无限’，而这是不可能的。”如果阿基里斯事实上在T时追上了乌龟，那么，“这是一种不合逻辑的现象，因而决不是真理，而仅仅是一种欺骗”。这就是说感官是不可靠的，没有逻辑可靠。
他认为：“穷尽无限是绝对不可能的”。根据这个运动理论，芝诺还提出了一个类似的运动佯谬：“飞矢不动”。
2-3“飞矢不动”
在芝诺看来，由于飞箭在其飞行的每个瞬间都有一个瞬时的位置，它在这个位置上和不动没有什么区别。那么，无限个静止位置的总和就等于运动了吗？或者无限重复的静止就是运动？中国古代也有类似的说法。
2-4“飞鸟之景，未尝动也”
这是中国名家惠施的命题，与“飞矢不动”同工异曲。这就是不可抗拒的推理和不可回避的实事相冲突。
德国哲学家尼采在《希腊悲剧时代的哲学》里有一章《可疑的悖论》，称芝诺的悖论为“否定感官的悖论”。尽管阿基里斯在赛跑中追上起步领先的乌龟完全合乎事实，但为什么“不合逻辑”？因为芝诺运用了“无限”这个概念，这是一种逻辑上的假设，而现实世界里是不可能有无限者存在的，这就出现了假设与现实的矛盾。
尼采的结论

尼采说道：在这两个悖论里，“无限”被利用来作为化解现实的硝酸。如果无限是决不可能成为完善的，静止决不可能变为运动，那么，真相是箭完全没有飞动，它完全没有移位，没有脱离静止状态，时间并没有流逝。
换句话讲，在这个所谓的、终究只是冒牌的现实中，既没有时间、空间，也没有运动。最后，连箭本身也是一个虚象，因为它来自多样性，来自由感官唤起的非一的幻象。下面是尼采的分析：
假定箭拥有一种存在，那么，它就是不动的、非时间的、非造而有的、固定的、永恒的。这是一个荒谬的观念！
假定运动是真正的实在，那么，就不存在静止。因而，箭没有位置、没有空间。又是一个荒谬的观点！
假定时间是实在的，那么，它就不可能被无限地分割。箭飞行所需要的时间必定由一个有限数目的瞬间组成，其中每个瞬间都必定是一个原子。仍然是一个荒谬的观念！
尼采得出这样的结论：我们的一切观念，只要其经验所与的、汲自这个直观世界的内容被当作“永恒真理”，就会陷入矛盾。如果有绝对运动，就不会有空间；如果有绝对空间，就不会有运动；如果有绝对存在，就不会有多样性；如果有绝对的多样性，就不会有统一性。
极限理论的诞生
事实上，这两个悖论中提到的这个“动与不动”的对立统一，今天都已经得到了完美的解决，这就是极限理论的诞生。牛顿在运动学研究时，初创微积分，但由于没有巩固的理论基础，出现了历史上的“第二次数学危机”。十九世纪初，法国科学家以柯西为首建立了极限理论，后来又经过德国数学家维尔斯特拉斯进一步的严格化，使极限理论成为微积分的坚定基础，运动问题也得到了合理的解释。
可以想见，在微积分和极限理论发明或被接受以前，人们很难解释这一运动佯谬。感官不同于思维，当希腊人用概念来判决现实的时候，如果逻辑与现实发生矛盾，芝诺指责感官为“欺骗”。当思维找不到合理解释的时候，直观的形式、象征或比喻都无济于事。尼采的分析虽然详细、精辟，但他无法把它们综合起来。
2-5“一尺之捶，日取其半，万世不竭”
这是《庄子·天下》中惠施的一句名言。二千多年前中国古人同样运用了无限的概念。战国名家宋国人惠施(约公元前370-前310)曾任梁国的宰相，论辩奇才，是庄子的朋友，和公孙龙并列为名家的代表人物。他的著作多已亡佚，只能从其他诸家的论述中看到他的言行片段。
惠施的学说强调万物的共相，因而事物之间的差异只是一种相对的概念，现存与惠施有关的奇怪命题，例如，“山与泽平”、“卵有毛”、“鸡三足”、“犬可以为牛”、“火不热”、“矩不方”、“白狗黑”、“孤驹未尝有母”等，都可以说是悖论，但是大部份没有留下具体的争辩过程。惠施的悖论在西方也很有影响。
毛泽东的偏爱
毛泽东从辩证法的角度基本接受惠施无限可分的观点。一九六四年八月十八日，他同哲学工作者谈话时说：“列宁讲过，凡事可分。举原子为例，不但原子可分，电子也可分。”又说：“电子本身到现在还没有分裂，总有一天能分裂的。‘一尺之捶，日取其半，万世不竭’，这是个真理。不信，就试试看。如果有竭就没有科学了。”
有人注意到，毛泽东十分偏爱这句话，如五十年代中期对家钱三强，一九六四年八月同周培源、于光远，一九七三年、一九七四年接见杨振宁、李政道，等等，都提到这句话。
2-6“1厘米线段内的点与太平洋面上的点一样多”
多少哲学家、数学家都唯恐陷入悖论而退避三舍。二十三岁获博士学位的德国数学家康托尔(1845-1918)六年以后向无穷宣战。他成功地证明了：一条直线上的点能够和一个平面上的点一一对应，也能和空间中的点一一对应。由于无限，1厘米长的线段内的点，与太平洋面上的点，以及整个地球内部的点都“一样多”。
然而，康托尔的“无穷集合”与传统的数学观念发生冲突，遭到谩骂。直到一八九七年第一次国际数学家会议，他的成果才得到承认，几乎全部数学都以集合论为基础。罗素称赞他的工作“可能是这个时代所能夸耀的最巨大的工作。”
集合论的矛盾
同时，集合论中也出现了一些自相矛盾的现象，尤其是罗素的理发师悖论，以极为简明的形式震撼了数学的基础，这就是“第三次数学危机”。此后，数学家们进行了不懈地探讨。
例如，一九九六年英国剑桥大学出版社出版了亨迪卡的《数学原理的重新考察》，这本书以罗素的《数学原理》(1903)为蓝本的，试图完善逻辑和数学基础。它主要阐述了亨迪卡和桑朵新创的IF(Independence-FriendlyFirst-OrderLogic)逻辑及其可能产生的影响。它挑战了许多公认的观念，如公理集合论作为数学理论的适当框架，对说谎者悖论也作了进一步的探讨。它是否将引起一场逻辑和数学基础的革命？我们还将拭目以待。
(摘自美国“新语丝”网站)

5月31日 11:03 | 添加评论 | 固定链接 | 写入日志 | philosophy

关闭WinXP2数据保护DEP的方法

1、右键我的电脑→属性→高级→启动和故障恢复→设置→编辑，调出记事本
2、找到“/noexecute=optin”的位置
3、把等号后“optin”五个字母改为“alwaysoff”
4、点文件→保存，点击“确定”关闭启动和故障恢复，再点击“确定”关闭系统属性对话框。重启后即可生效。

5月31日 11:00 | 添加评论 | 固定链接 | 写入日志 | 操作系统

注册表常用技巧

注册表常用技巧

禁止气球状的弹出信息
找到HKEY_CURRENT_USER-Software-Microsoft-Windows-CurrentVersion-ExplorerAdvanced子键，把十六进制值EnableBalloonTips设置成0。

显示隐藏文件
在HKEY_CURRENT_USER-Software-Microsoft-Windows-Current Version-Explorer-Advanced子键下，有几个值影响隐藏文件和文件夹的显示。要让XP显示出隐藏的文件和文件夹，把Hidden设置成1。要显示出文件的扩展名，把HideFileExt设置成0。要显示出受保护的操作系统文件，把ShowSuperHidden设置成1。注意：这部分设置也可以通过“文件夹选项”对话框的“查看”选项卡修改。

winXP copy-move-send 增加右键菜单
HKEY_CLASSES_ROOT\AllFilesystemObjects\shellex\ContextMenuHandlers\Move To:
{C2FBB631-2971-11D1-A18C-00C04FD75D13}

HKEY_CLASSES_ROOT\AllFilesystemObjects\shellex\ContextMenuHandlers\Copy To:
{C2FBB630-2971-11d1-A18C-00C04FD75D13}

HKEY_CLASSES_ROOT\AllFilesystemObjects\shellex\ContextMenuHandlers\Send To:
{7BA4C740-9E81-11CF-99D3-00AA004AE837}

修改Windows XP安装源文件路径
展开HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion子项，双击右边窗格中的SourcePath值项，将其值改成新的Windows XP安装源文件路径即可。

改变收藏夹的目录指向
展开HKEY_CURRENT_USER\Software\Microsoft\Windows\CurrentVersion\Explorer\User Shell Folders子项，然后在右边窗口双击Fovarites，将其值改变你想指向的目录，如：F:\My documents\收藏夹。其实这个子项下的内容还可以让你共享Cookie和桌面文件夹等。另外可以把以个人用户方式登录的遨游浏览器上的收藏夹导出到IE收藏夹。

重新安装IE
HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Active Setup\Installed Components\{89820200-ECBD-11cf-8B85-00AA005B4383} 将值项IsInstalled的值改为0，这样系统就会认为IE还未安装，重启后便会自动下载安装文件并安装。

清除任务栏中的多余工具栏分隔符/保存工具栏
有时，明明已锁定了任务栏，可重起系统后会多出一两个多余的工具栏分隔符，而选“关闭工具栏”关闭后，重起系统就又出来了，可通过修改注册表来修复此问题。打开注册表编辑器：

1. 展开HKEY_CURRENT_USER\Software\Microsoft\Windows\CurrentVersion\Explorer\TrayNotify

删除右边空格的IconStreams和PastIconsStream，然后打开任务管理器，结束Explorer.exe进程，然后再重建explorer进程。

2. 展开HKEY_CURRENT_USER\Software\Microsoft\Windows\Current

Version\Explorer\StuckRets2，删除该子键下所有键值。

3.（解除锁定桌面项）展开HKEY_CURRENT_USER\Software\Microsoft\Windows\Current

Version\Policies\Explorer选中No Save Setting，将键值从“1” 改为“0” ；将NoToolbarCustomize的值设为1。

解锁任务管理器
[HKEY_CURRENT_USER\Software\Miocrosoft\Windows\Current Version\Policies\System]
"Disable TaskMgr"=REG_DWORD:0x00000000(0)

5月31日 10:24 | 添加评论 | 固定链接 | 写入日志 | 操作系统

删除ATI右键

regsvr32 atiacmxx.dll /u

[转]逃课(搞笑)

成龙__霸王洗发水"
逃课篇:
其实刚开始让我逃课的时候我是拒绝的，因为你不能让我逃我就逃，至少你应该让我试一下，
我不想你拍的时候加了很多特技，说我逃课有多厉害，老师老是不点我名，也不找我谈话等等...出来之
后同学一定会骂我，根本没有这么厉害的学生，证明上面那个是假的....
后来我知道他们逃课是有计划有预算的，那么试了一个月来以后，觉得还不错...我自己现在还每天
都逃课，还介绍给我的舍友一块儿逃，来来来，大家一起来逃课......
那拍之前我就跟导演讲，拍的时候就拍我正常逃课，不要加任何的特技上去，没有，逃课就是逃
课...我要给同学看到，我逃课是这样子，你们逃课也是这样子(完)

5月27日 10:40 | 添加评论 | 固定链接 | 写入日志 | 趣事

优化系统速度改变工作线程数

Technorati 标签: Critical worker threads,Delayed worker threads,工作线程

< Day Day Up >

System Worker Threads

During system initialization, Windows creates several threads in the System process, called system worker threads, that exist solely to perform work on behalf of other threads. In many cases, threads executing at DPC/dispatch level need to execute functions that can be performed only at a lower IRQL. For example, a DPC routine, which executes in an arbitrary thread context (because DPC execution can usurp any thread in the system) at DPC/dispatch level IRQL, might need to access paged pool or wait for a dispatcher object used to synchronize execution with an application thread. Because a DPC routine can't lower the IRQL, it must pass such processing to a thread that executes at an IRQL below DPC/dispatch level.

Some device drivers and executive components create their own threads dedicated to processing work at passive level; however, most use system worker threads instead, which avoids the unnecessary scheduling and memory overhead associated with having additional threads in the system. A device driver or an executive component requests a system worker thread's services by calling the executive functions ExQueueWorkItem or IoQueueWorkItem. These functions place a work item on a queue dispatcher object where the threads look for work. (Queue dispatcher objects are described in more detail in the section "I/O Completion Ports" in Chapter 9.) Work items include a pointer to a routine and a parameter that the thread passes to the routine when it processes the work item. The routine is implemented by the device driver or executive component that requires passive-level execution.

For example, a DPC routine that must wait for a dispatcher object can initialize a work item that points to the routine in the driver that waits for the dispatcher object, and perhaps points to a pointer to the object. At some stage, a system worker thread will remove the work item from its queue and execute the driver's routine. When the driver's routine finishes, the system worker thread checks to see whether there are more work items to process. If there aren't any more, the system worker thread blocks until a work item is placed on the queue. The DPC routine might or might not have finished executing when the system worker thread processes its work item. (On a uniprocessor system, a DPC routine always finishes executing before its work item is processed because thread scheduling doesn't take place when the IRQL is at DPC/dispatch level.)

There are three types of system worker threads:

Delayed worker threads execute at priority 12, process work items that aren't considered time-critical, and can have their stack paged out to a paging file while they wait for work items.
execute at priority 13, process time-critical work items, and on Windows Server systems, have their stacks present in physical memory at all times.
A single hypercritical worker thread executes at priority 15 and also keeps its stack in memory. The process manager uses the hypercritical work item to execute the thread "reaper" function that frees terminated threads.

The number of delayed and critical worker threads created by the executive's ExpWorkerInitialization function, which is called early in the boot process, depends on the amount of memory present on the system and whether the system is a server. Table 3-11 shows the initial number of threads created on different system configurations. You can specify that ExpInitializeWorker create up to 16 additional delayed and 16 additional critical worker threads with the AdditionalDelayedWorkerThreads and AdditionalCriticalWorkerThreads values under the registry key HKLM\SYSTEM\CurrentControlSet\Control\Session Manager\Executive.

Table 3-11. Initial Number of System Worker Threads

Windows 2000

Windows 2000 Server

Windows XP and Windows Server 2003

Delayed

Critical

Hypercritical

The executive tries to match the number of critical worker threads with changing workloads as the system executes. Once every second, the executive function ExpWorkerThreadBalanceManager determines whether it should create a new critical worker thread. The critical worker threads that are created by ExpWorkerThreadBalanceManager are called dynamic worker threads, and all the following conditions must be satisfied before such a thread is created:

Work items exist in the critical work queue.
The number of inactive critical worker threads (ones that are either blocked waiting for work items or that have blocked on dispatcher objects while executing a work routine) must be less than the number of processors on the system.
There are fewer than 16 dynamic worker threads.

Dynamic worker threads exit after 10 minutes of inactivity. Thus, when the workload dictates, the executive can create up to 16 dynamic worker threads.

EXPERIMENT: Listing System Worker Threads

You can use the !exqueue kernel debugger command to see a listing of system worker threads classified by their type:

kd> !exqueue

Dumping  ExWorkerQueue: 8046A5C0



**** Critical WorkQueue(  current  =0  maximum=  1)

THREAD 818a2d40  Cid 8.c    Teb: 00000000  Win32Thread: 00000000 WAIT

THREAD 818a2ac0  Cid 8.10   Teb: 00000000  Win32Thread: 00000000 WAIT

THREAD 818a2840  Cid 8.14   Teb: 00000000  Win32Thread: 00000000 WAIT

THREAD 818a25c0  Cid 8.18   Teb: 00000000  Win32Thread: 00000000 WAIT

THREAD 818a2340  Cid 8.1c   Teb: 00000000  Win32Thread: 00000000 WAIT



**** Delayed WorkQueue( current  =0  maximum  =1)

THREAD 818a20c0  Cid 8.20   Teb: 00000000  Win32Thread: 00000000 WAIT

THREAD 818a1020  Cid 8.24   Teb: 00000000  Win32Thread: 00000000 WAIT

THREAD 818a1da0  Cid 8.28   Teb: 00000000  Win32Thread: 00000000 WAIT



**** HyperCritical  WorkQueue( current=  0maximum  =1  )

THREAD 818a1b20  Cid 8.2c   Teb: 00000000  Win32Thread: 00000000 WAIT

< Day Day Up >

5月27日 8:26 | 添加评论 | 固定链接 | 写入日志 | 操作系统

内存优化修改注册表

Windows IT Pro

- Advertise

Windows IT Pro Logo

search for on this site the network power search help

Performance Registry Secrets
View the book table of contents

Author:
Tim Daniels

Published:
December 1997

Publisher:
29th Street Press

Abstract
In this chapter, you will find specific ways to improve the performance of your general Windows NT system, your network, your file systems, and your printers. Mark Russinovich contributed to this chapter, selecting some of his favorite performance-oriented registry modifications to help you squeeze every drop of power from your Windows NT machines.

GENERAL SYSTEM PERFORMANCE
The tuning parameters in this section tune the general performance of NT, rather than a specific aspect of NT such as file systems or memory. By our definition, these keys tune NT itself, not specific device drivers or user-level applications.
VI-1 Managing the number of threads can help you optimize performance. When NT boots, it creates a pool of worker threads that NT’s kernel, Executive Subsystems, and device drivers use. When one of these components places a work item, such as the Cache Manager’s lazy cache flusher or the Memory Manager’s idle memory zero-filler, in a queue, a thread is assigned to process it. If too many threads are in the pool, they needlessly consume system resources. However, if not enough threads are in the pool, work items are not serviced in a timely manner. Adjust the following two registry keys to change the number of threads.

Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Control\Session Manager\Executive

Value Name:
AdditionalCriticalWorkerThreads

Data Type:
REG_DWORD

Value:
<any number>

VI-2
Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Control\Session Manager\Executive

Value Name:
AdditionalDelayedWorkerThreads

Data Type:
REG_DWORD

Value:
<any number>

The number you specify in the value increases the number of threads created for the specified work queue. NT has two thread queues. Threads in the Critical Work Queue execute in the low real-time priority range, so they have precedence over most system activity. Threads in the Delayed Work Queue have a slightly lower priority that results in higher latency — the threads that service it compete with other processing for CPU time. The number of threads created for each queue is based on two criteria: whether the system is a server or workstation and how much physical memory the system has. You can’t tell directly whether more threads in a particular situation give you better performance, but you can experiment with benchmark workloads to see if they provide a benefit.
VI-3 To make your foreground windows more responsive than your background windows, in Windows 3.51 you could boost priorities. In NT 4.0 Workstation, instead of boosting priorities, you set quantums to boost foreground windows (Server treats foreground and background windows alike). Change this registry entry to set the boost value.

Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Control

Value Name:
PriorityControl

Data Type:
REG_DWORD

Value:
0

You can also set the boost value with a slider in the Performance tab of the System Control Panel applet. Valid values are shown below.

0
No quantum boost for foreground windows

1
Foreground windows have quantums twice as long as other windows

2
Foreground windows have quantums three times as long as other windows

VI-4 Here’s a tricky registry entry.

Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Control\Session Manager\Executive

Value Name:
PriorityQuantumMatrix

Data Type:

Value:

Although you can’t use this value to enhance performance, its name implies that you can. This value actually encodes expiration data about NT Beta and Release Candidates.
MEMORY PERFORMANCE
When you think of memory performance, fast SRAMs and multilevel caching come to mind. Although these hardware components play a critical role in the latency and throughput of a computer’s memory subsystem, tweaking the Registry obviously cannot upgrade your machine’s hardware. So in this section, improving memory performance means improving the way NT uses physical and virtual memory. You know you’re experiencing memory problems that you can fix with registry entries, rather than additional hardware, when you get “not enough memory to complete task…” messages in the system Event Log or on the display monitor.
VI-5 The file system cache grows and shrinks as the applications’ memory demands change. However, at system startup, the cache’s minimum, maximum, and “ideal” sizes are calibrated based on the Registry value you set below.

Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Control\Session Manager\
Memory Management

Value Name:
LargeSystemCache

Data Type:
REG_DWORD

Value:
0

This value controls the size of the system’s file system cache. A value of 0 indicates a small cache, which is the default setting for Workstation, while 1 indicates a large cache, which is the default setting for Server. Servers usually perform some file sharing and therefore require a larger cache.
VI-6 NT categorizes memory as either paged or nonpaged. Paged memory holds data that can temporarily be moved from physical memory to the system’s paging file if space is needed for different data in physical memory. Nonpaged memory stores data that must be present in physical memory at all times. NT allocates a portion of physical memory for nonpaged memory when the system starts up, and this allocation can’t grow. If drivers or NT exhaust this storage, the system may become unstable or operations may start to fail. Modifying these two registry entries lets you override the defaults set for the paged and nonpaged pools.

Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Control\Session Manager\
Memory Management

Value Name:
NonPagedPoolSize

Data Type:
REG_DWORD

Value:
<number in bytes >

This value is the number of bytes of physical memory you want to allocate for nonpaged memory. To monitor your memory, use Performance Monitor to watch the percentage of committed memory that’s used. If more than 80 percent of memory is used consistently, you should increase this value. Likewise, if the system’s nonpaged memory usage is low, reducing the amount allocated for it increases system performance.

VI-7
Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Control\Session Manager\
Memory Management

Value Name:
PagedPoolSize

Data Type:
REG_DWORD

Value:
<number in bytes >

This value is the number of bytes in the paged memory pool. It overrides the system’s default calculation, which is roughly equal to the physical memory on Workstation and a minimum of 50 MB on Server. Paged pool is different from virtual memory size — the paged pool is space reserved in the system’s virtual memory map for NT and device drivers to allocate pageable data.
You can monitor the amount of paged pool the system uses with the Performance Monitor, which also shows the maximum allocation possible. If the amount of memory used is consistently more than 80 percent, you should use this setting to override the default.
NETWORKING PERFORMANCE
The registry gives you lots of control over Windows NT’s various networking components. Because your system may not run every networking service, you should identify which components are active so that you can target them when evaluating networking performance with tools like Performance Monitor.
General Networking
VI-8 This value controls the order in which your network redirectors are used. If you have more than one network redirector, such as Lanmanager Workstation or Client Services for NetWare, loaded on your system, you should consider changing this entry.

Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Control\NetworkProvider

Value Name:
Order

Data Type:
REG_SZ

Value:
<redirector name,redirector name,redirector name>

In this value, you set the order in which the WNet API uses your network providers. When the WNet API is called, it processes requests by sending them to each network provider in the order listed. If it gets back a response that the provider doesn’t process that type of request, the API calls the next provider listed. Thus, if most of your system’s activity relates to accessing NetWare shares, you should move the NetWare provider name to the start of the list. This small modification can improve performance significantly.
Redirector
The Windows NT Redirector manages remote volume connections. When you access a volume across the network, Redirector takes the file system request and invokes the appropriate protocol to send the request to a file system server component on the other machine.
VI-9 The entry below switches the behavior of Redirector between write-through and write-behind. By default, Redirector delays sending the server write requests to a volume, which lets other operations proceed or complete. When you change this value, Redirector sends write operations immediately and waits for a response before proceeding with other operations.

Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Services\Rdr\Parameters

Value Name:
UseWriteBehind

Data Type:
REG_DWORD

Value:
0

To disable write-behind, change this value to 1; the default is 0. You may need to use this highly synchronous mode in specialized environments or when tracking down networking problems with a network monitoring tool; otherwise, you should use the write-behind value because write-through degrades performance.
VI-10 The value in the registry entry below is a hybrid between write-through and write-behind modes.

Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Services\Rdr\Parameters

Value Name:
UseAsyncWriteBehind

Data Type:
REG_DWORD

Value:
1

The default value of 1 enables asynchronous write-behind, which means that Redirector immediately sends write requests to the server but doesn’t wait for a response before processing other requests. Changing the value to 0 disables asynchronous write-behind; you should disable it only when the network requires special coherency guarantees or when you’re debugging.
Lanmanager Workstation
Lanmanager Workstation is Microsoft’s implementation of a Redirector. It speaks SMB (Server Message Block).

V-11
Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Services\LanmanWorkstation\ Parameters

Value Name:
MaxCmds

Data Type:
REG_DWORD

Value:
15

This value specifies the number of threads that the Workstation service creates during initialization to handle requests; the default value is 15. If Workstation has more requests than it has threads to service those requests, it holds up some requests until a thread finishes servicing a request. Therefore, if multiple users or applications are simultaneously accessing remote shares, increasing this value can improve network throughput.

VI-12
Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Services\LanmanWorkstation\ Parameters

Value Name:
SessTimeout

Data Type:
RED_DWORD

Value:
45

This value sets the number of seconds after a request is queued before Lanmanager indicates that the request has failed (if it hasn’t been processed). The default value is 45. Applications that access network shares may report not having enough memory to process a request — too many outstanding operations are on the volume and some requests are timing out. Raising this value lets more operations be serviced asynchronously and may cause the messages to go away.

VI-13
Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Services\LanmanWorkstation\ Parameters

Value Name:
CacheFileTimeout

Data Type:
REG_DWORD

Value:
10

This value specifies the time after which Lanmanager Workstation purges its locally cached data after closing a file. Many applications open files, perform an operation, close them, and repeat this sequence to perform more operations. If you have a system with a dedicated application, you can adjust this value to improve performance. Increase the time if applications close and reopen the same files at intervals longer than 10 seconds, and reduce the time if applications access few files and do not open and close them regularly.

VI-14
Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Services\LanmanWorkstation\ Parameters

Value Name:
DormantFileLimit

Data Type:
REG_DWORD

Value:
45

This value sets the number of files that Lanmanger leaves open in the file system cache after an application closes it. By default, Lanmanager Server allows only a certain number of open files from any one Lanmanager Redirector connection. Having a file open on the connection can improve performance if the application opens it again. However, keeping files open can also overload the server’s limit, and in those cases you should reduce this value.
VI-15 Normally, Lanmanager Workstation reads and stores data in the local file cache before an application requests it. These read-aheads can cause network degradation when applications rarely read files sequentially on shared volumes or when the network’s throughput is low. To better manage the read-ahead function, change this registry entry.

Hive:
HKEY_LOCAL_MACHINE

Key:
System\CurrentControlSet\Services\LanmanWorkstation\ Parameters

Value Name:
ReadAheadThroughput

Data Type:
REG_DWORD

Value:
-1

This value sets the minimum network throughput, in kilobytes per second, before Lanmanager Workstation enables the read-ahead function. The default value is -1.
Lanmanager Server
Lanmanger Server is Microsoft’s implementation of an SMB (Server Message Block) file server. It is a focal point of performance tuning for any file-sharing server.
VI-16 If you frequently get “server paged” or “server nonpaged” errors in Performance Monitor, you are seeing Lanmanager Server running out of memory that it’s allocated for itself. You may want to raise these values to give Lanmanager Server more memory; however, giving these memory resources to Lanmanager keeps other system drivers and applications from using them. To give Lanmanager more paged and nonpaged memory, change the registry entries below.

Hive:
HKEY_LOCAL_MACHINE