Utility2:Appropriate Evaluation Policy
UCP收集所有Managed Instance的数据的机制,是通过启用各个Managed Instances上的Collection Set:Utility information(位于Management->Data Collection->System Data collection Sets),Utility Information 是On Demand 模式。UCP每15分钟调用Utility information一次,Utility information 收集所在Instance 对CPU 和Storage 资源的使用数据,发送到UCP。这样UCP就能获各个Managed Instances的资源利用数据,并存储在sysutility_udw 数据库中,通过dashboard report 显示一个consolidated view给DBA,便于DBA监控多个managed instances的资源利用率。
1,Utility information
2,UCP收集数据的频率
collection Set Upload Frequency :Every 15 minutes,Frequency Value不可更改。
3,Evaluation Policy
Evaluation Policy 有三部分组成:Time Window,Fault Tolerance 和 Collection Frequency.
Collection Requency 是 Every 15 minutes,不可更改。
Time Window 和 Fault Tolerance 显示Volatile Resource Policy Evaluation Tab中,可以配置,默认的设置如图
默认设置:time window 是 1hour, Fault tolerance 是20%,Upload Frequency 是15Minutes。
意思是,在1hour内,UCP会收集4次数据,如果有1次数据超出Overutilized threshold,那么 percent in violation 就是1/4=25%,由于 25%>20%,因此,UCP 评估的结果是Overutilized。默认的设置是不允许任何一次数据超出Overutilized threshold,而不被评估为Overutilized,意思是只要有一次Overutilized,那么就会被mark 为 Overutilized。
如果对资源的Overutilized不是很 sensitive,允许1hour内出现1次Overutilized,如果其他3次没有Overutilized,那么USP就不会Mark为Overutilized。实现这个scenario,可以设置较大的Fault tolerance,也可以设置较大的Overutilized threshold。
Example1,设置 Fault tolerance 是30%
time window 是 1hour, Fault tolerance 是30%,Upload Frequency 是15Minutes。
意思是,在1hour内,UCP会收集到4次数据,如果有1次数据超出Overutilized threshold,那么percent in violation 就是1/4=25%,由于 25%<30%,因此,UCP 评估的结果是没有Overutilized。如果25% 大于Underutilized 的 Fault tolerance,那么UCP会标记Instance处于Well Utilized状态。
Example2,设置更大的Overutilized threshold
思路是使资源利用达不到Overutilized threshold,默认的Overutilized threshold 设置是70%。
如果一个小时内,资源的utility 是 75%,60%,65%,40%,有1次的utility超出70%,那么Instance就被标记为Overutilized;
如果将Overutilized threshold 设置是80%,那么,资源的utility都不会超过80%,那么Instance就不会被标记为Overutilized。
可以通过配置Global Policy 或 Individual Policy来 设置Overutilized threshold 的值。
A: 配置Global Policy
B:配置Individual Policy
Appenix: 引用MSDN:《Reduce Noise in CPU Utilization Policies (SQL Server Utility)》
The data collection interval is 15 minutes. This value is not configurable.
The default upper threshold processor utilization policy is 70%. Options range from 0% to 100%.
The default evaluation period for processor overutilization is 1 hour. Options range from 1 hour to 1 week.
The default percentage of data points in violation before CPU is reported as overutilized is 20%. Options range from 0% to 100%.
For example, based on default values, 4 data points will be collected every hour, and the policy threshold is 20%. So by default, any violation in a 1-hour collection period will be 25% of 4 data points. The default values report any violation of the CPU overutilization policy threshold.
To reduce noise generated by a single violation, consider the following options:
Increase the evaluation period by 1 increment to 6 hours. A single violation in 6 hours would be 1 data point in a sample size of 24. In this case, the policy would tolerate 4 violations of the policy threshold (16.7% of data points) in 6 hours, but would report overutilization for 5 or more violations (>20% of data points) in a 6-hour collection period.
Increase the tolerance for percentage of violations by 1 increment to 30%. A single violation in 1 hour would be 1 data point in a sample size of 4. In this case, the policy would tolerate 1 violation per hour, but would report overutilization for 2 or more violations (>30% of data points) in a 1-hour collection period.
Increase policy thresholds for SQL Server managed instance and data-tier application processor utilization. For more information about how to change the global CPU utilization policies for managed instances of SQL Server or data-tier applications, see Utility Administration (SQL Server Utility). For more information about how to change CPU utilization policies for individual instances of SQL Server, see Managed Instance Details (SQL Server Utility). For more information about how to change CPU utilization policies for individual data-tier applications, see Deployed Data-tier Application Details (SQL Server Utility).
The data collection interval is 15 minutes. This value is not configurable.
The default lower threshold for processor utilization policy is 0%. Options range from 0% to 100%.
The default evaluation period for processor underutilization is 1 week. Options range from 1 day to 1 month.
The default percentage of data points in violation before CPU is reported as underutilized is 90%. Options range from 0% to 100%.
Based on default values, 672 data points are collected every week, but the policy threshold is 0%. So by default, this policy does not generate processor underutilization violations. For more information about how to change the global CPU utilization policies for managed instances of SQL Server or data-tier applications, see Utility Administration (SQL Server Utility). For more information about how to change CPU utilization policies for individual instances of SQL Server, see Managed Instance Details (SQL Server Utility). For more information about how to change CPU utilization policies for individual data-tier applications, see Deployed Data-tier Application Details (SQL Server Utility).
参考文档:
Reduce Noise in CPU Utilization Policies (SQL Server Utility)
Utility2:Appropriate Evaluation Policy的更多相关文章
- SLC-Microsoft:Microsoft Lifecycle Policy
ylbtech-SLC-Microsoft:Microsoft Lifecycle Policy Microsoft Lifecycle Policy The Microsoft Lifecycle ...
- PLT:说说Evaluation strategy
Brief 在学习方法/函数时,我们总会接触到 按值传值 和 引用传值 两个概念.像C#是按值传值,但参数列表添加了ref/out后则是引用传值,但奇怪的事出现了 namespace Foo{ cla ...
- Signal Processing and Pattern Recognition in Vision_15_RANSAC:Performance Evaluation of RANSAC Family——2009
此部分是 计算机视觉中的信号处理与模式识别 与其说是讲述,不如说是一些经典文章的罗列以及自己的简单点评.与前一个版本不同的是,这次把所有的文章按类别归了类,并且增加了很多文献.分类的时候并没有按照传统 ...
- 转载:android audio policy
Audio policy basic:https://www.cnblogs.com/CoderTian/p/5705742.html Set volume flow:https://blog.csd ...
- 基音检测算法的性能:Performance Evaluation of Pitch Detection Algorithms
http://access.feld.cvut.cz/view.php?cisloclanku=2009060001 Vydáno dne 02. 06. 2009 (15123 přečtení) ...
- 【Deep Learning Nanodegree Foundation笔记】第 9 课:Model Evaluation and Validation
In this lesson, you'll learn some of the basics of training models. You'll learn the power of testin ...
- Security Policy:行级安全(Row-Level Security)
行级安全RLS(Row-Level Security)是在数据行级别上控制用户的访问,控制用户只能访问数据库表的特定数据行.断言是逻辑表达式,在SQL Server 2016中,RLS是基于安全断言( ...
- Homomorphic Evaluation of the AES Circuit:解读
之前看过一次,根本看不懂,现在隔这么久,再次阅读,希望有所收获! 论文版本:Homomorphic Evaluation of the AES Circuit(Updated Implementati ...
- Policy Management
策略管理用于管理数据库实例.数据库以及数据库对象的各种属性,Policy Management 位于Management Catalog下, 一,Basic concepts 引用园子里深蓝的博客&l ...
随机推荐
- C++-Qt【1】-退出程序&静态调试
目前还没有发现很好的调试qt代码的方法,权且记录一下: #include "mainwindow.h" #include "ui_mainwindow.h" # ...
- [资料分享]Java35期基础班和就业班
直接上链接:https://pan.baidu.com/s/1jI8p2Xw 密码:gr9x
- switch..case使用
1.多个if...else连在一起使用的时候,可以转为使用更方便的switch结构.switch (XXX) { case "aaa": // ... break; case &q ...
- Torch7学习笔记(四)StochasticGradient
使用随机梯度下降训练神经网络 StochasticGradient是一个比较高层次的类,它接受两个参数,module和criterion,前者是模型结构,后者是损失函数的类型.这个类本身有一些参数: ...
- C++中const的全面总结
C++中的const关键字的用法非常灵活,而使用const将大大改善程序的健壮性,本人根据各方面查到的资料进行总结如下,期望对朋友们有所帮助. Const 是C++中常用的类型修饰符,常类型是指使用类 ...
- java 过滤表情符号
/** * 过滤表情符号 * @create by ldw on 2016-10-25 * @param str * @return str(去掉表情符号的字符串) * @version 1.0 * ...
- ubuntu 16.04 Dockerfile 安装mysql
默认上MariaDB的包并没有在Ubuntu仓库中.要安装MariaDB,我们首先要设置MariaDB仓库. 配置MariaDB 仓库 # apt-get install software-prope ...
- Hadoop_常用存储与压缩格式
HDFS文件格式 file_format: TEXTFILE 默认格式 RCFILE hive 0.6.0 和以后的版本 ORC hive 0.11.0 和以后的版本 PARQUET hive 0.1 ...
- SQL语句操作数据与一些函数使用的丰富数据库
数据库有多重要,其实不用我说,但该怎么运用好数据库下SQL语句与其它的如“函数”等等,那就需要我们大家多多去练习并总结其中的窍门,或许你的总结没那么好,担只要你的练习足够多,就算那不是窍门,那也将是你 ...
- 安装windows server 2012 r2 的那点事儿
windows server 2012 r2 安装无法找到install.wim 错误代码0x80070026,以及制作U启动盘决解ISO文件超过5G大小限制的解决方案 用UltaISO刻录后,sou ...