Tài liệu SQL Server 2008 Database Administration– Monitoring and Performance pdf

Chapter 9: Monitoring and Performance Course 157: SQL Server 2008 Database Administration Presented by Scott Whigham... • Chapter 9 is about performance and exception monitoring your se

Chapter 9: Monitoring and

Performance

Course 157: SQL Server 2008 Database Administration

Presented by Scott Whigham

• PerfMon

• Data Collector

What We’re Going to Cover

• Chapter 9 is about performance and

exception monitoring your server

• We’ll start with the basics (using Profiler) and end with new features in SQL Server

2008 (Data Collector)

Chapter Introduction

• PerfMon

• Data Collector

What We’re Going to Cover

• There are many built-in tools to monitor

performance and health:

– Activity Monitor and sp_who

• Covered in Chapter 5!

– SQL Server Profiler

• Detailed view of statements issued against instance

– Dynamic Management Views (DMVs)

• “Live” view of server health and activity

– Performance counters and objects (“PerfMon”)

• “Live” view of server health and activity

– Data Collector

Tools of Performance Monitoring

Profiler DMVs Perf

Mon

Data Coll.

Provides “live”, real-time reporting?

Can do trend analysis and historical reporting? Requires

manual load

Requires manual load

Requires manual load

Can view SQL issued?

Provides detail-level view or summary? Detail Depends… Summary Summary

Can be used to diagnose performance problems?

Tools of Performance Monitoring

“But Mr Scott, how do I know when to use each one?”

• Let’s play with Profiler!

In the next video…

Chapter 9: Monitoring and

Performance

Course 157: SQL Server 2008 Database Administration

Presented by Scott Whigham

• PerfMon

• Data Collector

What We’re Going to Cover

• SQL Server Profiler is a tool for

monitoring real-time queries issued

against your server

– “Old Faithful” - been around for years

– Requires sysadmin or ALTER TRACE server

permission

SQL Server Profiler

SQL Server Profiler

• DBAs use traces to monitor statements

• SQL Server Profiler is the GUI for

defining, viewing, and replaying traces

– It does not have to be running to create a trace

– Many DBAs create their traces via Profiler and

then script out to jobs

SQL Server Profiler

• Running Profiler on a “live” system adds

to an already stressful situation

– Particularly if run interactively on the server

• If you must run Profiler against a “live”

server, run it remotely

• Adds network overhead but this is almost never an issue today

SQL Server Profiler

• Trace data can be stored:

– In a trc file (fastest)

– In a SQL Server table

• trc files can be configured to rollover to

a new file after a certain threshold

• Can load trc file contents into SQL Server table after trace for maximum flexibility!

SQL Server Profiler

• After trace is complete, you can export

results

SQL Server Profiler

• Tips on working with Profiler

– Traces can have filters to help minimize the noise

• The better you filter, the less of an impact you put on the server

– Being able to write queries against the trace

results can help you identify many common trends and problems

• This requires you load the trace data into a table!

SQL Server Profiler

• Let’s play with Profiler!

In the next video…

Chapter 9: Monitoring and

Performance

Course 157: SQL Server 2008 Database Administration

Presented by Scott Whigham

• PerfMon

• Data Collector

What We’re Going to Cover

• DBAs use Dynamic Management Views

(“DMVs”) and Dynamic Management

Functions (“DMFs”) to peek under the

covers

– In the sys schema and begin with dm_

– Used to get up-to-the-second performance

information during critical time

– Often need to JOIN many together to come up

Dynamic Management Views

Profiler DMVs Perf

Mon

Data Coll.

Provides “live”, real-time reporting?

Can do trend analysis and historical reporting? Requires

manual load

Requires manual load

Requires manual load

Can view SQL issued?

Provides detail-level view or summary? Detail Depends… Summary Summary

Can be used to diagnose performance problems?

Tools of Performance Monitoring

Reminder…

• There are too many DMxs to cover each

one so we’ll cover concepts and core

DMxs

– Over 130 DMVs and DMFs in SQL Server 2008!

• It is important for DBAs to maintain a

script library

Dynamic Management Views

• The easiest way to get to know the

• Intellisense can help you greatly!

Dynamic Management Views

• Object Explorer can also help

Dynamic Management Views

• You can also query system views for full

DMV and DMF Query Results

• We’ll take a look at a few categories and examples

In the next video…

Chapter 9: Monitoring and

Performance

Course 157: SQL Server 2008 Database Administration

Presented by Scott Whigham

• PerfMon

• Data Collector

What We’re Going to Cover

• Reminder: the more popular categories

for performance monitoring

Dynamic Management Views

• From a performance standpoint, you will spend a lot of time in the Database

• Common index-related DMVs in the

Database Category

– sys.dm_db_missing_index_details

returns information about indexes that could

potentially help speed up the system

you whether an index is being used and, if so, in

what way

Database Category

• Common index-related DMFs in the

• We’ll take a look at writing queries using the DMVs and DMFs in the Database

category

In the next video…

Chapter 9: Monitoring and

Performance

Course 157: SQL Server 2008 Database Administration

Presented by Scott Whigham

• PerfMon

• Data Collector

What We’re Going to Cover

• Let’s look at the other categories now…

Dynamic Management Views

• The Execution/Session category will help you spy on people

– That’s what you really want to do, isn’t it?

– Can also view specific query information

• Plans, text, stats, caching information

Execution/Session Category

• There is at least one DMF in this category that you definitely want to know

– sys.dm_exec_sql_text allows you to

see the query text a user submits

• By itself, it is not that useful but combined with other Execution/Session DMVs it is very powerful

• Accepts one parameter - the memory location of the query plan -

Execution/Session Category

• A common DMV in this category is:

metrics about queries submitted to server

• Reads, writes, waits, execution count, etc

• plan_handle column is the column to pass to

sys.dm_exec_sql_text to return SQL

Execution/Session Category

SELECT TOP(50) sql.text

Return 50 most-run queries

• Other common DMVs in this category are

info

the cached query plans

• Is it compiled or adhoc?

• How many times has it been used?

Execution/Session Category

• Spying on users is easy!

authenticated session

Returns all non-internal sessions

SELECT * FROM sys.dm_exec_sessions

WHERE session_id>50

Execution/Session Category

• We’ll write several queries and see the

Execution/Session category in action

In the next video…

Chapter 9: Monitoring and

Performance

Course 157: SQL Server 2008 Database Administration

Presented by Scott Whigham

• PerfMon

• Data Collector

What We’re Going to Cover

• Let’s look at the IO and OS categories

now to help diagnose disk pressure

Dynamic Management Views

• The most common item in this category

is

– sys.dm_io_virtual_file_stats()

returns information about usage of the database

files

• Reads, writes, IO waits, size

• Accepts two parameters:

– database_id - use NULL for all databases – file_id - the file_id from sys.database_files; use NULL for

IO Category

• sys.dm_io_virtual_file_stats

returns several columns:

IO Category

Column Description

sample_ms Milliseconds since server was restarted

num_of_reads Physical reads of objects in database file

num_of_bytes_read Large textual-based columns will affect this greatly

io_stall_read_ms Total time a user process waited for IO to this file

num_of_writes,

num_of_bytes_written,

Database write activity

• Here are some things to monitor with

dm_io_virtual_file_stats()

– Look at total reads+writes to find “most active”

– Look at io_stall to identify files that might

benefit from new disks

• All databases have stalls; the important metric is “What percentage of the uptime of the server am I having

stalls?”

• Compare read/write stalls on log and data file to

IO Category

View IO stats for all DBs on instance:

IO DMF

Aggregate total IO since last restart:

GROUP BY vfs.database_id

ORDER BY IO_MB DESC

IO DMF

• The OS category is one of the largest

• You can use this category to find waiting tasks:

dozens and dozens of wait types

waits for a specific resource

• Hint: use dm_os_wait_stats to identify the areas of contention and then dm_os_waiting_tasks to view

OS Category

• These two DMVs are accumulative

– Their statistics are the totals since SQL Server

service has been started

• To track effectively over time, it is

necessary to reset periodically

– Common to store in a table for historical

perspective

– DBCC SQLPERF(‘<dmv_name>’, CLEAR)

OS Category

Current wait info:

SELECT wait_type, wait_time_ms / 1000 as wait_time_s

FROM sys.dm_os_wait_stats

ORDER BY wait_time_ms DESC

Clear out stats:

DBCC SQLPERF ('sys.dm_os_wait_stats', CLEAR)

Run this at a later time to view waits since CLEAR

SELECT wait_type, wait_time_ms / 1000 as wait_time_s

FROM sys.dm_os_wait_stats

OS Category

OS Category

• To identify disk pressure, you can look in

sys.dm_os_wait_stats for specific wait

types

– Order by wait_time_ms DESC to identify most often waits

– If anyone of these are in the top two positions, you

have disk I/O pressure:

• Let’s look further at the OS category and others

In the next video…

Chapter 9: Monitoring and

Performance

Course 157: SQL Server 2008 Database Administration

Presented by Scott Whigham

• PerfMon

• Data Collector

What We’re Going to Cover

• The OS category has lots of info

of CPUs, start time of service

loaded DLL that SQL Server is using

• Cryptography Manager

• Kerberos Security Package

• Microsoft SQL Server Native Client 10.0

OS Category

• Many OS category DMVs can help

diagnose memory pressure

– Performance Counters can help too

– We’ll cover Perf Counters in an upcoming section

of this chapter

OS Category

• The main areas for memory pressure are:

– Database page cache (used to cache table/index)

– Query Workspace memory (for Hash and Sort ops)– Plan cache

– There are others…

OS Category

Find out what has the most entries

OS Category

Top 10 consumers of Buffer Pool

SELECT TOP 10 [type]

OS Category

• One area that many servers have trouble

with is the TokenAndPermUserStore cache

– An internal security system that generates security

tokens for sessions

– Particularly troublesome area for DBs with lots of

ad-hoc SQL

– The problem can be that the

TokenAndPermUserStore cache takes up so

much memory and is so frequently accessed that

when there are a high number of items in the cache,

access takes a long time

OS Category

1 Queries for diagnosing

issues:

What is the size? The larger, the slower!

SELECT SUM(single_pages_kb + multi_pages_kb)

AS ‘CurrentSizeOfTokenCache(kb)’

FROM sys.dm_os_memory_clerks

OS Category

2 Queries for diagnosing

Scary!

• There are multiple ways to prevent

• access check bucket count

• access check cache quota

•

OS Category

• There is one way to fix a “live”

– Free the system cache

WARNING: Do not use on production without

a full understanding!

After running, there will not be an entry

in sys.dm_os_memory_clerks for TAPUS!

DBCC FREESYSTEMCACHE('TokenAndPermUserStore')

OS Category

• Let’s look at blocking and locking…

In the next video…

Chapter 9: Monitoring and

Performance

Course 157: SQL Server 2008 Database Administration

Presented by Scott Whigham

• PerfMon

• Data Collector

What We’re Going to Cover

• To maximize coverage, you want to use

both DMVs/DMFs along with

Performance Monitor

– a.k.a PerfMon and System Monitor

• PerfMon gathers stats about SQL Server

as well as the rest of the environment

– Disk subsystems, non-SQL resource usage

Performance Monitor

• PerfMon uses objects and counters

– Objects are for categorization

– Objects contain one or more counters

• When you install SQL Server, it installs a

set of instance-specific objects

Performance Monitor

• You can view the SQL Server objects and counters using the DMV

sys.dm_os_performance_counters

– Careful - more than 1,100 rows

Performance Monitor

• A few things to know about

sys.dm_os_performance_counters

– Only displays SQL Server instance’s counters

– Any counter that has “/sec” in the name is cumulative since time SQL Server service was started

• Transactions/sec

• Bulk copy rows/sec

– To accurately read these values, compare

intervals

• Compare 0800 reading to 0900 reading, for

Performance Monitor

• Example query to return all

deprecated features usage

– Many feature usages are actually SQL Server internals and tools!

Performance Monitor

• Querying for specific counters allows

ad-hoc views of point-in-time information

– But no graphs, charts, pretty colors

• For trend analysis, you have a few

choices:

– Store the query results in a table for later review

– Use PerfMon to store logs for later analysis

Performance Monitor

• Some DBAs will store certain values in a

table for later querying or for historical

perspective

– Watch the rest of the chapter prior to attempting

this!

Create a new table (”PerfCounters”)

and store current counters

SELECT *

INTO PerfCounters

FROM sys.dm_os_performance_counters

Performance Monitor

• All DBAs will use PerfMon

Performance Monitor

• Let’s take a look at specific objects and

counters

In the next video…

Chapter 9: Monitoring and

Performance

Course 157: SQL Server 2008 Database Administration

Presented by Scott Whigham

• PerfMon

• Data Collector

What We’re Going to Cover

• There are a few specific areas you will

want to monitor

– Sometimes you can find that “Aha!” moment

using ad-hoc monitoring

– Sometimes you can only find that “Aha!” moment using trend analysis

Performance Monitor

• A few topics before we go further:

– PerfMon, counters with “/sec” in the name are

not cumulative

• Makes it easy to do ad-hoc monitoring

– PerfMon’s GUI defaults to taking measurements

every one second

• Too often for trend analysis

• 30-seconds is generally “good enough” for ad-hoc troubleshooting

Performance Monitor

• A few areas DBAs want to monitor:

• A few areas DBAs want to monitor:

– Disk usage

– CPU usage

– Memory usage

Performance Monitor

• Disk usage monitoring

– Disk I/O is commonly a performance stealer

– While SQL Server manages the “when” and “how

often” of disk I/O, it is the operating system that

performs I/O

– When monitoring disk activity, you need to

monitor both SQL Server’s disk activity and the

disk activity of the server as a whole

Disk Usage

• Disk I/O cannot be isolated as a metric

– Example: You are monitoring disk I/O and notice

that you have waits for disk reads/writes Does

this always mean that you need faster/more

disks?

• No! Could be that you are paging constantly and memory is the real bottleneck

Disk Usage

• Important system objects and counters:

Disk Usage

Object Counter When to worry

PhysicalDisk % Disk Time Consistently > 90%

PhysicalDisk Current Disk Queue

PhysicalDisk Avg Disk Queue Length In many configs, a disk has 1 spindle Start

worrying when this value is more than 1.5x the number of spindles

• PhysicalDisk: % Disk Time

– Tells us how “busy” your array is

– With this counter, you want to worry about

sustained activity, not bursts

– A % Disk Time of more than 75% for several

continuous minutes occurring several times a day

can likely be resolved:

• By adding more disks to the array

• By using faster drives

Disk Usage