Tài liệu Module 8: Managing Storage and Optimization pptx

Demonstration: Designing Storage for the Sales Cube In this demonstration, you will learn how to create a storage design by using the Storage Design Wizard!. How can MOLAP be faster to p

Contents

Overview 1

Analysis Server Aggregations 17

Lab A: Designing Storage for Sales 23

Optimization

products, people, characters, and/or data mentioned herein are fictitious and are in no way intended

to represent any real individual, company, product, or event, unless otherwise noted Complying with all applicable copyright laws is the responsibility of the user No part of this document may

be reproduced or transmitted in any form or by any means, electronic or mechanical, for any purpose, without the express written permission of Microsoft Corporation If, however, your only means of access is electronic, permission to print one copy is hereby granted

Microsoft may have patents, patent applications, trademarks, copyrights, or other intellectual property rights covering subject matter in this document Except as expressly provided in any written license agreement from Microsoft, the furnishing of this document does not give you any license to these patents, trademarks, copyrights, or other intellectual property

 2000 Microsoft Corporation All rights reserved

Microsoft, BackOffice, MS-DOS, Windows, Windows NT, <plus other appropriate product

names or titles Replace this example list with list of trademarks provided by copy editor Microsoft is listed first, followed by all other Microsoft trademarks in alphabetical order > are either registered trademarks or trademarks of Microsoft Corporation in the U.S.A and/or other countries

<This is where mention of specific, contractually obligated to, third party trademarks, which are added by the Copy Editor>

The names of companies, products, people, characters, and/or data mentioned herein are fictitious and are in no way intended to represent any real individual, company, product, or event, unless otherwise noted

Other product and company names mentioned herein may be the trademarks of their respective owners

Instructor Notes

This module provides students with a comprehensive understanding of Microsoft® SQL Server™ Analysis Services storage options and optimization techniques for online analytical processing (OLAP) cubes The characteristics

of the three storage modes—multidimensional OLAP (MOLAP), relational OLAP (ROLAP), and hybrid OLAP (HOLAP)—are reviewed in detail followed by an overview of aggregations The module then takes students through the Storage Design Wizard with discussion of specific aggregation options and further discussion of the contents of aggregations and design guidelines The module concludes with a review of usage-based optimization and general optimization tuning techniques

There are two labs in the module In lab A, students create a storage design and process a cube by using the Storage Design Wizard In lab B, students learn the interfaces and mechanics of usage-based optimization

After completing this module, students will be able to:

! Explain the advantages and disadvantages of the three data storage modes

! Use the Storage Design Wizard to set storage design

! Describe how aggregations work and design aggregations for cubes

! Describe the concepts and mechanics of usage-based optimization

! Override aggregation settings per dimension

Materials and Preparation

This section lists the required materials and preparation tasks that you need to teach this module

Required Materials

To teach this module, you need the following materials:

! Microsoft PowerPoint® file 2074A_08.ppt

Preparation Tasks

To prepare for this module, you should:

! Read all the student materials

! Read all the instructor notes and margin notes

! Practice the lecture presentation and demonstration

! Complete the labs

! Review the Trainer Preparation presentation for this module on the Trainer Materials compact disc

! Review any relevant white papers that are located on the Trainer Materials compact disc

Presentation:

70 Minutes

Labs:

20 Minutes

Demonstration: Designing Storage for the Sales Cube

In this demonstration, you will learn how to create a storage design by using the Storage Design Wizard

The following demonstration procedures contain information that does not fit in the margin notes or is not appropriate for student notes

! To restore a new database and define a data source

1 In Analysis Manager, right-click the server, click Restore Database, click the Look in list, find and click the file

C:\Moc\2074A\Labfiles\L08\Module 08.CAB, click Open, and then click Restore

2 Click Close, and then double-click Module 08 to expand the database

3 Below Module 08, double-click Data Sources, right-click the Module 08 data source, and then click Edit

4 Click the Connection tab of the Data Link Properties dialog box, and then verify that localhost is selected in step 1

5 In step 2, verify that Use Windows NT Integrated security is selected

6 In step 3, verify that Module 08 is selected

7 Click Test Connection and verify that the test succeeded Click OK twice

! To specify storage type

1 In the Module 08 database, right-click the Sales cube and click Design Storage

2 Click Next to skip the welcome page

3 From the Select the type of data storage step, click MOLAP, and then click Next

! To design aggregations

1 In the Set aggregation options step, click Performance gain reaches from

the Aggregation options pane

2 Type 20 in the percent box for Performance gain reaches to reflect a

20-percent aggregation target

For the Sales cube, the default value of 50 is unnecessarily high

3 Click Start to initiate the graphical simulation of Performance vs Size on the Set aggregation options step, and then click Next

Demonstration:

10 Minutes

! To process the cube

1 In the Finish the Storage Design Wizard step and with the Process now option clicked, click Finish

Regardless of the processing option you choose, Analysis Manager stores the definition of the aggregations in the OLAP repository Storing the definition of the aggregations is different from physically creating them, however The Storage Design Wizard designs aggregations but does not create them The Analysis Server does not create aggregations until you process the cube Processing the cube automatically creates any

aggregations that have been designed

2 Close the Process dialog box when processing is complete

! To examine the metadata

1 In Analysis Manager, click the Sales cube in the Analysis Manager tree pane and then click Metadata in the right details pane

2 Scroll down and notice the process and storage mode statistics

Other Activities

Difficult Questions

Below are difficult questions that students may ask you during the delivery of this module and answers to the questions These materials delve into subjects that are within the scope of the module but are not specifically addressed in the content of the student notes

1 If ROLAP is slow to process and query, why would an organization use this option?

ROLAP would be adopted if the organization needs a real-time OLAP solution—that is, data is always updated with the current fact table values

In this scenario, an organization defines its cube as ROLAP with zero aggregations All detail and aggregate data are calculated as users query the cube While queries are slow, in some situations perfectly updated data is more important than fast query times

2 Do Analysis Services MOLAP cubes have the “data explosion” problem common to OLAP solutions?

MOLAP database engines in competing products often create cubes that grow exponentially from source files to fully calculated cubes For example, a five-megabyte (MB) source file has been known to grow into

a five-gigabyte (GB) cube after processing

The data explosion problem when using MOLAP in Analysis Services does not exist to the extent experienced with other OLAP products In many cases, the MOLAP cube may be smaller than the data source The following are the principal reasons for the MOLAP storage efficiency:

• Analysis Services MOLAP cubes are completely dense in their data storage—that is, no null values are stored

• The Analysis Services query engine is highly optimized, calculating commonly accessed aggregations as the cube is queried so that fewer aggregations need to be precalculated and stored

• The Analysis Services data compression algorithms are highly efficient

Some multidimensional products presumably solve the data explosion problem by not using an OLAP engine, instead accessing data directly from a relational database

Such products classify themselves as ROLAP solutions because they access relational databases directly and give users a multidimensional view of the data, but do not create cubes that consume large amounts of storage Such ROLAP solutions typically suffer in query performance compared to MOLAP solutions

3 MOLAP cubes duplicate the detail data already stored in relational tables How can MOLAP storage be more efficient than ROLAP cube storage? How can MOLAP be faster to process than ROLAP if Analysis Services brings all the detail data to the Analysis Server?

Multidimensional structures and data are extremely compressed and optimized compared to the two-dimensional tables in relational databases In addition, when defining ROLAP cubes, indexes are automatically created in the relational database management system (RDBMS)

Even though MOLAP cubes carry over the detail data, they can still be smaller than their ROLAP cube counterparts The exception is a ROLAP cube that has few or no aggregations

From a processing standpoint, it may be faster to create a multidimensional structure than to create, insert, and update data in relational tables It also may take a long time to build the indexes that are automatically created in ROLAP cubes Again, the exception is when the ROLAP cube has few or no aggregations defined

4 If processing time and disk space are not constraints, should aggregations be set to 100 percent for Performance gains reaches?

If Performance gains reaches is set to 100 percent, all a cube’s possible aggregations will not necessarily be computed The setting simply targets that query performance will be potentially increased by 100 percent

As a cube defines more aggregations, query performance improvements reach a point of diminishing returns Some cubes may slow in their query performance if the aggregation percentage is set too high

5 How does one estimate the size of a cube based on fact table size?

You can estimate the data storage for MOLAP data on disk in bytes, assuming zero aggregations, by using the following formula:

(((2 * total number of levels) + (4 * number of measures)) * number of records) / 3

6 Analysis Services has intelligent algorithms for determining the most optimized aggregation design Why would you choose to override the dimension level aggregations?

In most cases, you use the Storage Design Wizard and the Usage-Based Optimization Wizard to define aggregations However, there may be exceptional situations in which you might want to exercise control, overriding wizard algorithms

For example, you might not want your cube to contain aggregations for the lowest level of the Product dimension, because users will not be accessing data at that level Therefore, you have the ability to turn off aggregations for this level

Module Strategy

Use the following strategy to present this module:

! Analysis Server Cube Storage Deliver an overview discussion of general server storage issues and then talk through the characteristics of each of the three storage modes—

MOLAP, ROLAP, and HOLAP This discussion leads into a basic introduction to the concept of aggregations

! The Storage Design Wizard The materials in this section can be delivered as lecture by using the slides

or integrated with the demonstration Design Storage for the Sales Cube Because the demonstration essentially duplicates the materials in the student notes, it is recommended to integrate lecture and demonstration

The following table is a mapping of lecture topics to demonstration steps

Choose Storage Option To specify storage type Set Aggregation Options To design aggregations

Estimated Storage Reaches To design aggregations Performance Gain Reaches To design aggregations

the cube There will be substantial interest and questions from students about the storage size and performance implications of choosing from each of the three storage methods Do not rush through these materials or limit discussion and questions Review the questions and answers in the previous Difficult Questions section

Be prepared to discuss elements of aggregation again, including the specific

functioning of the three aggregation options—Estimated storage reaches, Performance gain reaches, and Until I click stop These three choices

represent different conceptual approaches and specific underlying algorithms for implementing aggregations

! Analysis Server Aggregations The subject of aggregation is explored in more detail, including review of aggregation tables, general characteristics of aggregations, and details about ROLAP aggregations Because students must thoroughly understand the concepts and wizard implementation of aggregations, the subject is approached repeatedly in this module at increasing levels of detail and sophistication

! Usage-Based Optimization This section presents an important feature set of Analysis Services—usage-based optimization Your lecture, following the materials in the student notes, takes students through the simple Usage-Based Optimization Wizard

Be prepared to answer detailed questions about how each of the five query options work by themselves and in conjunction with each other

The section is followed by lab B, Implementing Usage-Based Optimization, which can be conducted as a hands-on exercise with students following your demonstration The lab allows students to perform their own usage-based optimizations

! Optimization Tuning You complete the module with a discussion of specific optimization tuning methods, including how to override dimension and level settings No exercises or labs are included in this section However, you should switch to Analysis Manager to show the settings and to briefly explain their functions

Overview

! Analysis Server Cube Storage

! The Storage Design Wizard

! Analysis Server Aggregations

! Usage-Based Optimization

! Optimization Tuning

Query performance—how long it takes a user to access requested information—is a primary determining factor for online analytical processing (OLAP) cube storage design An optimal design produces fast queries for users while maintaining reasonable cube processing times Designing the storage mode and aggregations for a cube is one of the most crucial steps in cube development

After completing this module, you will be able to:

! Explain the advantages and disadvantages of the three data storage modes

! Use the Storage Design Wizard to set storage design

! Describe how aggregations work and design aggregations for cubes

! Describe the concepts and mechanics of usage-based optimization

! Override aggregation settings per dimension

In this module, you will learn

about aggregation design

and storage modes, which

are the key factors in

enabling fast query

response times

`

Microsoft® SQL Server™ 2000 Analysis Services supports three storage options:

! Multidimensional OLAP (MOLAP)

! Relational OLAP (ROLAP)

! Hybrid OLAP (HOLAP) Cube developers design the storage for cubes The storage design of a cube is transparent to clients—users do not realize that different cubes have different storage designs

The following list contains descriptions of some key characteristics of cube storage modes:

! The storage mode is transparent to clients Users and client applications see only cubes For users, the only indication of the storage mode is their observations of query performance

! The storage mode can be changed after the initial storage decision is made Once you specify storage and put a cube into production, you can change to

a different storage type later After you change the mode, you must reprocess the cube and then Analysis Server reloads the data and creates new aggregations

! Each partition of a cube can have a different storage mode A cube can consist of multiple partitions One cube might have both a MOLAP partition and a ROLAP partition

Topic Objective

To introduce Analysis

Services storage options

Lead-in

Analysis Services supports

three storage options:

MOLAP, ROLAP, and

HOLAP

Delivery Tips

Avoid going into too much

detail on the three storage

modes here Wait until the

following slides

Point out that in the

preceding illustration, Aggs

stands for aggregations

! Analysis Server does not allocate storage for missing values For example, if

no bikinis are sold in Antarctica, no space is allocated to that missing value

Because missing values take up no storage, cubes are 100 percent dense—

that is, all storage is efficiently used This characteristic of Analysis Server helps avoid the data explosion problems of other OLAP products

MOLAP Storage Mode

! Details and Aggregations Stored in Multidimensional Format

! Fastest Storage Option for Queries

! Often the Most Efficient in Terms of Disk Storage, Due

to Compression

The following list contains characteristics of MOLAP storage:

! Detailed data and aggregations are stored in a multidimensional format on the Analysis Server

• Because the detail data from the fact table is brought into Analysis Server for storage, data is duplicated

• The level of detail imported into a cube is based on the grain of the cube’s dimensions For example, if the fact table contains daily data records, but the grain of the cube time dimension is month, the cube will contain data at a month level The fact table daily records are combined

at cube processing time

• After a MOLAP cube is processed, all data necessary for querying is located on the Analysis Server The source relational database management system (RDBMS) is not accessed other than at processing time

! MOLAP cubes have the fastest query performance for users

! MOLAP is a very economical mode in terms of disk storage, due to efficient data compression algorithms

In MOLAP cubes, detailed

data and aggregations are

stored in a multidimensional

format on the Analysis

Server

Delivery Tips

Tell students that you

design cubes as MOLAP the

vast majority of the time

because of the fast query

times, processing times, and

efficient storage of MOLAP

cubes

Point out that Aggs stands

for aggregations in the

preceding illustration

ROLAP Storage Mode

! Details and Aggregations Stored in RDBMS

! Slowest Query Performance

! Most Often the Slowest to Process

! Analysis Server Can Create Indexed Views

! Useful for Large Data Sources

! Provides Real-Time OLAP Solution

The following are characteristics of ROLAP storage:

! Detailed data and aggregations are stored in relational tables in the source database

• RDBMS indexes are automatically created in the data source to improve cube performance

• All queries, other than those satisfied by the client or server caches, must access the source RDBMS tables

! Under most circumstances, ROLAP cubes are much slower in query performance than MOLAP cube equivalents

! ROLAP cubes are usually the slowest to process, unless the ROLAP cubes contain few aggregations

! When assigning the ROLAP storage mode to cubes that have data sources defined in SQL Server 2000 databases, the Analysis Server attempts to create indexed views instead of tables, assuming certain criteria are met in the data source

! You use the ROLAP storage mode when the data source is too large to be stored and processed effectively in MOLAP or HOLAP

! You use the ROLAP storage mode when you require a real-time OLAP solution

Aggregation tables must be

stored in the same RDBMS

as the data source of a

cube

Delivery Tip

Point out that Aggs stands

for aggregations in the

preceding illustration

Creating Real-Time Cubes

Some cubes require immediate refreshing of data when changes occur in the data source However, by using standard cubes, you are forced to reprocess cubes when data changes in the underlying database To overcome the delay of data updates, you have the ability to create real-time OLAP cubes in Analysis Services

You create a real-time cube by performing the following steps:

! Define the cube by using the ROLAP storage mode

! Select the Enable real-time updates check box in the Select the type of data storage page in the Storage Design Wizard

Real-Time Cube Behavior

The following behavior occurs in real-time cubes:

! The Analysis Server polls the database to determine if changes have been made to the data source

! The Analysis Server flushes the server cache after it detects any database changes to ensure that clients do not query outdated data

! Cube data automatically refreshes when fact table data changes

HOLAP Storage Mode

! Details Maintained in RDBMS

! Aggregations Created in Multidimensional Format

! Good Option where Disk Consumption Is a Concern

! Good Compromise if Details Are Accessed Infrequently

The following are characteristics of HOLAP storage:

! Detailed data is maintained in the RDBMS

! Aggregations are created in the multidimensional cube format and are stored

on the Analysis Server

! Because detailed data is not duplicated, HOLAP is a reasonable storage compromise where disk consumption is a concern

! In a situation when users do not frequently access the details stored in the RDBMS and the cube contains a high degree of aggregation, HOLAP is a good option for cube storage

Most cubes use MOLAP as the cube storage mode However, you can define a cube with a HOLAP design to use less cube storage than if the cube used the MOLAP storage design The following are effects of using the HOLAP design

Point out that Aggs stands

for aggregations in the

preceding illustration

Cube Aggregations

! Full Aggregation Not Necessary

! Effects on Cube Size and Processing Time

$ Cube size and processing times increase as aggregations are added to a cube

! Tools for Implementing Aggregations

$ Storage Design Wizard

$ Usage-Based Optimization Wizard

$ Dimension and level aggregation properties

Aggregations are precalculated summaries of detailed data that enable Analysis Server to answer queries quickly Cubes contain aggregations designed with the Storage Design Wizard or with the Usage-Based Optimization Wizard

Precalculated aggregations are fundamental to OLAP cubes, making user queries significantly faster than calculating aggregations at query time

Accessing aggregations is transparent to users and client applications The Analysis Server accesses aggregations automatically

A cube can contain multiple partitions Aggregations can be designed like a cube’s storage mode, on a partition-by-partition basis For more information about partitions, see module 10, “Managing Partitions,” in course

2074A, Designing and Implementing OLAP Solutions with Microsoft SQL

Server 2000

Full Aggregation Not Necessary

It is not necessary to fully aggregate a cube in Analysis Services Analysis Server utilizes a variety of algorithms to optimize data access, thereby eliminating the need for total cube aggregation

If an aggregation does not exist to satisfy a query containing summarized data, Analysis Server does not need to query the lowest level of data Instead, the server uses an intermediate aggregation, if one exists, to satisfy the query

detailed data that enable

Analysis Server to answer

through the bullets on the

slide Point out that there

are more sections covering

aggregations later in the

module

Note

Here is an example of how Analysis Server uses an intermediate aggregation to satisfy a query:

! Assume a hierarchy for the Time dimension consisting of the levels Year, Quarter, and Month Aggregations exist for Quarter but not for Year

! When a user queries the Year level, the server does not access Month level

data to calculate the yearly totals Instead, the server derives the totals from the quarter totals that are already aggregated

Effects on Cube Size and Processing Time

Cube size increases as aggregations are added to a cube In addition, processing times increase because pre-aggregations are calculated at process time

Long processing times tend to be more detrimental to an OLAP application’s success than large cube sizes Disk space is inexpensive compared

to time lost waiting for a cube to process

Size of the cube depends on several factors:

! The number of aggregations

! The number of dimensions

! The number of levels

! The number of measures

! The number of members

! The data distribution of the cube When designing aggregations, the goal is to maximize query performance while maintaining reasonable cube sizes and processing times

Tools for Implementing Aggregations

In the following sections, you will learn about three tools available to you for implementing aggregations:

! Storage Design Wizard

! Usage-Based Optimization Wizard

! Dimension and Level aggregation properties

Note

# The Storage Design Wizard

! Choosing a Storage Mode

! Setting Aggregation Options

! Determining the Level of Aggregation

! Finishing Up

The Storage Design Wizard is the interface that lets you specify the storage mode and aggregation design The next section introduces the steps involved in using the Storage Design Wizard to design storage modes and aggregations:

! Choosing a storage mode

! Setting aggregation options

! Determining the level of aggregation

! Finishing up There are two entry points into the Storage Design Wizard:

! After building or modifying a cube, you are prompted to set storage options

You start the wizard by clicking Yes

! Right-click a cube or a partition in a cube, and then click Design Storage

The user interface of the Storage Design Wizard differs depending on:

! Whether storage has been designed previously for the cube

! Whether the cube contains partitions

Topic Objective

To introduce the steps

involved in using the

Storage Design Wizard

Lead-in

The Storage Design Wizard

is the interface that lets you

specify the storage mode

and aggregation design

Delivery Tip

Integrate this content into

the demonstration

Designing Storage for the

Sales Cube as an effective

method of presenting the

material

Choosing a Storage Mode

At the first step of the wizard, you must specify the storage type: MOLAP, ROLAP, or HOLAP The selected storage mode determines query performance, processing performance, and cube storage Therefore, you must determine the appropriate storage mode before starting the Storage Design Wizard

Topic Objective

To review storage options

Lead-in

At the first real step of the

wizard, you must specify the

storage type

Setting Aggregation Options

The next step of the Storage Design Wizard allows you to design aggregations for the cube or partition Three design options are available as radio button choices The design options are as follows:

! Estimated storage reaches

! Performance gain reaches

! Until I click Stop

The following characteristics apply to all three options:

! Regardless of aggregation option, the dimensional hierarchies are navigated and aggregations are built where the greatest performance benefits will be realized based on the number of members at a given level

! Aggregations are calculated when the cube is processed or refreshed and when incremental data is added to the cube—not during the aggregation design

! Aggregation design is per partition A single cube can consist of multiple partitions, and each partition can have a different aggregation design

At the stage in which you design aggregations in the Storage Design Wizard, aggregations are designed, not calculated The aggregation design is stored in the metadata repository Only when the cube is processed are actual aggregations built

Topic Objective

To show the aggregation

design step of the Storage

Design Wizard

Lead-in

The next step of the Storage

Design Wizard allows you to

design aggregations for the

cube or partition

Key Points

The Performance Gain

Reaches option is the most

helpful method for designing

aggregations

No one specific percentage

works best for all cubes

Every cube is different and

requires testing to find the

optimal aggregation

percentage The best

method of determining the

optimal percentage is to test

various settings for their

effects on the query and

cube processing times

Note

Estimated Storage Reaches The Estimated storage reaches option is determined by disk space usage,

specified in megabytes (MB) With this option, aggregations are designed until they consume the specified amount of disk space or until 100 percent

The more aggregations a cube contains, the greater the amount of time required to process the cube Excessive aggregation of a large cube can lead to unacceptable processing times

As you increase the performance percentage, you quickly reach a level of diminishing returns, consuming large amounts of disk storage in exchange for nominal query performance benefit For example, the number of aggregations required to reach 10 percent optimization is fewer—sometimes by an order of magnitude—than the number of aggregations required to improve from 10 percent to 20 percent

In general, you gain little by setting the Performance gain reaches option

higher than 50 percent Ten percent to 30 percent is adequate for most applications

The following is recommended practice for implementing Performance gain reaches:

1 Enter a low percentage, record the processing time, and then gauge retrieval performance through testing

2 Enter a higher percentage, again recording processing time and testing performance

3 Repeat this procedure until an optimum balance between processing time and retrieve performance is achieved for your business need

Note

Until I Click Stop With the Until I click Stop option, the system designs aggregations until one of

the following two situations occurs:

! You literally click Stop as you watch a simulation of your aggregation design take place in the Performance vs Size box

! A 100-percent performance gain is reached without clicking Stop You should click Stop when the curve of the line in the Performance vs Size

box starts to level at an acceptable performance gain

The Until I click Stop and Estimated storage reaches options are not

very helpful and potentially misleading when designing aggregations Use the

Performance gain reaches option for the best results

Note

Determining the Level of Aggregation

Because Analysis Server uses aggregations intelligently, it is unnecessary to create aggregations for all possible combinations of data A common mistake is over-aggregation

The following are two important guidelines to follow when designing aggregations:

! Due to the speed of MOLAP data retrieval, MOLAP generally requires less aggregation

! Due to the relative slowness of ROLAP retrieval, ROLAP generally requires more aggregation

The key to defining aggregations is identifying when you have reached a point

of diminishing returns Testing is necessary to find that point for each cube

Topic Objective

To explain how much

aggregation is necessary

Lead-in

Here are some guidelines

to help you determine how

much aggregation is

generally necessary

Finishing Up

You have four options at the last step of the Storage Design Wizard:

! Process now Writes the aggregation metadata to the repository, and then

processes the cube

! Save, but don’t process now Writes the aggregation metadata to the

repository, but does not process the cube or build the actual aggregations

! Cancel Discards all inputs to the Storage Design Wizard

! Back Allows you to go back to previous steps to change inputs

Topic Objective

To show the options offered

at the last step of the

Storage Design Wizard

Lead-in

At the last step of the

Storage Design Wizard, you

have four options