HIGH AVAILABILITY OF SQL SERVER IN THE CONTEXT OF SLA

Tobiasz Janusz Koprowski

SELECT {BIO}

• Polish SQL Server User Group Leader

• Microsoft Certified Trainer

• MCP, MCSA, MLSS, MLSBS, MCTS, MCITP, MCT

• SQL Server MVP from 2010

• Friends of RedGate PLUS

• PASS SQL Azure Virtual Chapter Co-Founder

• Blogger, Influencer, Technical Writer

• Last 7 years (living) in Data Center in Wrocław

• Generally about 12 years in IT/banking area

• GITCA Technical Lead & Vice-Chair EMEA Board

• Speaker at SQL Server Community Launch, Time for SharePoint, CodeCamps, SharePoint Community Launch, CISSP Day, InfoTRAMS, SQLSaturday, SQLBits, CarreerCon,

• Autor of few articles on TechNet (PL) and WSS.pl portal

• Deep Dives Co-Author:

High availability of SQL Server in the context

of Service Level Agreements (Chapter 18th)

• Working for MS Subject Matter Expert and MS Terminology community (Windows 7, 8 & Visualstudio 2010,2011

Agenda

• Back to the school:− What is High Availability− What is Service Level Agreement

• Using HA in SQL Server 2008• HA solutions in SQL Server 2008 that means: Enterprise, Enterprise• Why SLA and DBA• Dependency of SLA and HA• Case Studies• Q&A

What is High Availability?

• High Availability (HA) to ensure the continuedoperation of equipment and systems for thepurposes of (usually) in an enterprise productionenvironment.

• Is designed to prevent data loss as a result of:− software bugs,− manufacturing defects− hardware failure− natural disasters− human error− other unforeseen events

What is Disaster?

What is Disaster?

Two kinds of monster:

PSO > USO > SLA

• PSO Planned System Outages – Planned System Unavailability

− Minimum planned unavailability, due to the need to carry outmodernization work, installing patches, replacement / extension ofhardware,

− Agreed/accepted by/with the client and not affecting theprovisions of the HA, and SLA, until

• ...USO Unplaned System Outages – Unplanned System Unavailability

− an error that prevents a partial or total work environment in atangible, measurable customer

− resulting in high costs if you need repairs, as well as penaltypayments for non-SLA

Performance metrics (HA)

• What it really is the availability of the order of 99.99%?

• Availability 99.99% to 0.01UNAVAILABILITY in a requested period (eg annual), which ...

• How much is that in terms of the unavailability of the server / environment / database:

Availability = MTBF / MTBF + MTTR

− MTBF -> Mean Time Between Failures

− MTTR -> Mean Time To Repair

Unavailability in minutes, hours, days, weeks...

Availability %Downtime

per year

Downtime

per month*

Downtime

per week

90% 36.5 days 72 hours 16.8 hours

95% 18.25 days 36 hours 8.4 hours

98% 7.30 days 14.4 hours 3.36 hours

99% 3.65 days 7.20 hours 1.68 hours

99.5% 1.83 days 3.60 hours 50.4 min

99.8% 17.52 hours 86.23 min 20.16 min

99.9% ("three nines") 8.76 hours 43.2 min 10.1 min

99.95% 4.38 hours 21.56 min 5.04 min

99.99% ("four nines") 52.6 min 4.32 min 1.01 min

99.999% ("five nines") 5.26 min 25.9 s 6.05 s

99.9999% ("six nines") 31.5 s 2.59 s 0.605 s

What is SLA?

• SLA - Service Level Agreement.

• The origins date back to 1980 and the agreementsbetween operators and end customers.

• Mutually negotiable contract for the provision of services(not just IT, but these in particular)

• It must be concluded formally, though legally permissibleis an informal agreement

• Including the level and range of services provided bymeans of measurable indicators (level of accessibility,usability, performance)

• The contract should have specified minimum andmaximum range for each subject to its services

Metrics of SLA

There is no specific SLA measurement WITHOUT indicators!

SAMPLE CALL CENTER / SERVICE DESK:

• ABA (Abandonment Rate): Percentage of calls abandoned while waiting for a response.

• ASA (Average Speed to Answer): Average time (usually in seconds) required for the connection of boards help.

• TSF (Time Service Factor): Percentage of calls answered in precise time frame, such as 80% in 20 seconds.

• FCR (First Call Resolution): Percentage of calls where the problem was solved without having to switch to another expert

• TAT (Turn Around Time): The time it takes to complete certain tasks.

High Availability in SQL Server 2008

Microsoft SQL Server 2008/2008R2/2012:

• Database Mirroring

• Database Snapshots

• Windows Clustering

• SQL Server Replication

• Hot-add memory and CPU

• Online Index Operations

• Table and Index Partitioning

• Failover Clustering

• Peer-To-Peer Replication

• Always On

Solutions for HA for SQL Server

AREADATABASE

MIRRORING

FAILOVER

CLUSTERING

TRANSACTIONAL

REPLICATIONLOG SHIPPING

Data Loss no data loss no data loss some data loss possible

some data loss

possible

Automatic Failover YES (in HA mode) YES no no

Transparent To Client YES, autodirect

YES, connect to same

IP no, NLB helps no, NLB helps

Downtime < 3 seconds

20 seconds or more +

time to recovery seconds

seconds plus time to

recovery

Standby Ready Access Yes, with db snapshots no data loss YES

Data Granularity DB only all systems and db's table or view DB only

Masking of hdd failure YES No, shared disk YES YES

Special hardware

NO, duplicate

recommended Cluster HCL

NO, duplicate

recommended

NO, duplicate

recommended

Complexity Some More More More

Why High Availability?

• Businesses need to work around the clock to meet customer demands

• When systems are not running, businesses are losing revenue, opportunities, customers and reputation

• High availability reduces the impact of required maintenance on day-to-day operations and helps recover quickly from disasters

• Businesses need flexibility to easily build high availability solutions that meet business and technology needs

Prevent Unplanned Downtime

Reduce Planned Downtime

Multiple instance clustering

Database snapshots

Peer-to-peer replication

Online operations

Automatic page repair with

database mirroring

Hot-add CPU and RAM

High Availability

Live Migration

MS

FT

SLID

E

Prevent Unplanned Downtime

Because of the critical nature of the G4S application, CASON sets up the servers in a failover cluster to ensure high availability.

− —CASON Case Study

OfflineFailover Active

Applications &

Business Logic 110010100101110010100

101

110010

110010100101 110010100101 110010

110010100101 110010100101 110010

Active Active

Multiple-Instance Database

Clustering

• More than one passive node is

available to host instances from

multiple failovers on active nodes

• Having multiple failover nodes

provides greater availability

• Multiple instances can share the

same failover node, which reduces

hardware costs

• Simplified setup reduces

administrative costs

High Availability

MS

FT

SLID

E

Enhanced Database Mirroring

“This is a really powerful enhancement because prior to this… you would have to run DBCC CHECKDB... and that would likely mean taking downtime… With SQL Server

2008 Database Mirroring you can avoid the effort and downtime.”

— Glenn Berry, Database Architect, NewsGator Technologies

• High Performance Mirroring• Increase performance through

asynchronous mirroring

• Automatic Page Repair• Automatically detects page corruption

and retrieves data from the mirror

• Reduces downtime and management costs

• Minimizes application changes to correctly handle I/O errors

• Reporting from Mirror• Increase utilization of mirror server

• Reduce need for reporting servers

Applications &

Business Logic

Principal Mirror

High Availability

MS

FT

SLID

E

Help Recover From User Errors

“Database snapshots allow you to create read-only databases for reporting and can also be useful in your data recovery efforts in the event of a disaster.”

—Tim Chapman, SQL Server Database Administrator

• Database Snapshots• Provide a read-only static view of

the database at a point in time

• Revert to a point in time before user error

• Data loss is limited to changes after the snapshot

• Run reports from a snapshot created on the mirror server in a mirror to better utilize resources

Snapshot Source

110010100101 110010100101 110010

Applications &

Business Logic

110010100101 110010100101 110010

110010100101 110010100101 110010

High Availability

MS

FT

SLID

E

Maintain Databases Without Downtime

We recommend performing online index operations for business environments that operate 24 hours a day, seven days a week, in which the need for concurrent user

activity during index operations is vital.

— SQL Server Books Online

• Online Operations• Allow routine maintenance without

corresponding downtime

• Online index operations

• Online page and file restoration

• Online configuration of peer-to-peer nodes

• Users and applications can access data while the table, key, or index is being updated

Applications &

Business Logic

IndexTable

7

110010100101 110010100101 110010

0

2

3

4

5

1

5

2

3

4

0

1

5

4

2

7

0

Deleted

3

5

110010100101 110010100101 110010

6

Deleted

IndexTable

0

3

4

Deleted

7

5

6

Deleted

High Availability

MS

FT

SLID

E

Minimize Planned Downtime and Increase Efficiency

“This server already runs on our cluster solution with high availability, but after we have tested live migration on the new hardware, we’ll move it over to ensure optimal performance and reliability”

—Rodrigo Immaginario, IT Manager, Universidade Vila Velha

• Live Migration• Move running instances of VMs

between host servers

• Virtual machines can be moved for maintenance or to balance workload on host servers

• Perform maintenance on physical machines without any downtime

• Requires Windows Server 2008 R2 Hyper-v

Applications &

Business Logic11001010

0101

110010100101

110010

110010100101

110010100101

110010

110010100101

110010100101

110010

110010100101

110010100101

110010

110010100101

110010100101

110010

110010100101

110010100101

110010

110010100101

110010100101

110010

110010100101

110010100101

110010

High Availability

MS

FT

SLID

E

Minimize Planned Downtime

Hot-add CPU is the ability to dynamically add CPUs to a running system. Adding CPUs can occur physically by adding new hardware, logically by online hardware partitioning, or

virtually through a virtualization layer.

—SQL Server Books Online

• Hot-Add CPU and RAM• Dynamically add memory and

processors to servers without incurring downtime

• Requires hardware support for either physical or virtual hardware

Applications &

Business Logic110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

110010100101

110010100101 110010

High Availability

MS

FT

SLID

E

Access Data Seamlessly Across Servers

“[Microsoft] SQL Server 2008 replication proved to be very predictable and reliable in our testing. This helps us to create flexible and scalable replication solutions. Reliability must be at the

foundation of all that we do.”

— Sergey Elchinsky, Leading System Engineer, Baltika Breweries

• Peer-to-Peer Replication• Increases reliability by replicating

data to multiple servers

• Provides higher availability in case of failure or to allow maintenance at any of the participating nodes

• Offers improved performance for each node with geo-scale architecture

• Add and remove servers easily without taking replication offline, by using the new topology wizard

Applications &

Business Logic 1100101001011001011001011001

01

110010100101

110010100101

110010

110010100101

110010100101 110010

110010100101

110010100101

110010

High Availability

MS

FT

SLID

E

Database Mirroring

• Mirroring, which is a mirror image of the data− Available only for two bases (principal, mirror)

− The desired function of a witness (witness)

• Requirements:− principal, mirror - only SQL Server Enterprise

− witness - can be SQL Server Express

• Availability for the database:− copy of the database on a different physical server and / or virtual

• Availability for the system:− A copy of the entire environment on a different physical server and

/ or virtual

Database Mirroring Refresher Synchronous Mode

Commit

Write to

local log

Transmit to mirror

Write to

remote log

Log

Acknowledge

Committed

in log

Constantly

redoing on

mirror

Acknowledge

DBDB

Log

1

2

2

3

4

5

6

7

KEY POINT: mirror

database is an EXACT

copy of the principal

Hot-add memory and CPU

• In SQL Server 2005 added the ability to use memory to be added "on the fly"

• In SQL Server 2008 extends the dynamic capabilities of SQL Server work, allowing you to hot add CPU

• "Hot-add" is the ability to connect the RAM / CPU to the computer while the computer is running, and then by refreshing the SQL Server to use the new equipment ONLINE

• The equipment must support hot-add (of course!)− Supported only in the Enterprise Edition running on a 64-bit version of Windows

Server 2008 Datacenter / Enterprise

− SQL Server does not automatically start using the new processor / memory

− The need to reconfigure run

− Already running query will not use the newly added memory / processor.

Hot-Add CPU: Affinity Masks

• Affinity masks control which CPUs are used by SQL Server, and for what purpose

• Any affinity masks will need to be updated after hot-adding new CPUs

− If the affinity mask is set to non-zero, you will need to update it so that SQL Server knows it can use the new CPUs.

− On systems with > 32 CPUs, you will need to set the affinity64 mask to pick up the new CPUs

− If you want to use the new CPUs for IO only, you must add the relevant bits to the affinity I/O (or affinity64 I/O) mask

• If questioned about affinity masks− All zeroes means that Windows decides which CPUs are used

− Non-zero: single bit per CPU, if bit is 1, SQL Server will use it

− bit cannot be set in affinity AND affinity I/O mask

Fast Manual Failover

• High Security mode (synchronous mirroring without witness), manual failover is always used

• SQL Server 2005, if there is an emergency situation, the database on the mirror is closed and restarted to force the to recover non-commited transaction log − This can greatly increase the failover time

− Consider a database with hundreds of files, which all have to be opened to start the sequence database

• SQL Server 2008 removes this step, thus speeding up and reducing the use of emergency shutdown

SEND and REDO queues

Amount of log not sent to mirror

Represents possible data loss

• SEND Queue

• Unsent log

Log to redo on mirror

Represents failover time

• REDO Queue

Mirror

Time

Amount of log sent to mirror

Peer-to-Peer Topology (?)

• In SQL Server 2005 introduces the ability to use solution peer-to-peer (or "two-way") Transactional Replication

− A great way to scale the resources needed to work

− Partialy as a way to have "undue copy"

• One major drawback - changing the topology of peer-to-peer required to stop ALL activity on the servers in the topology tree

• In SQL Server 2008,

− these restrictions have been removed (in most cases),

− Setup Wizard also upgraded peer-to-peer network in SSMS

− Switching partitions can be repeated

Topology Wizard

• The wizard now is graphical, with drag-n-drop functionality for making topology connections

SQL Server 2012 & AlwaysOn | marketing

• Help reduce planned and unplanned downtime with the new integrated high availability and disaster recover solution, SQL Server AlwaysOn.

• Simplify deployment and management of HA requirements using integrated configuration and monitoring tools.

• Improve IT cost efficiency and performance using Active Secondary.

• Reduce planned downtime with Windows Server Core.

SQL Server 2012 & AlwaysOn | technical

AlwaysOn Failover Cluster InstancesAs part of the SQL Server AlwaysOn offering, AlwaysOn Failover Cluster Instances leverages Windows Server FailoverClustering (WSFC) functionality to provide local high availability through redundancy at the server-instancelevel—a failover cluster instance (FCI). An FCI is a single instance of SQL Server that is installed acrossWindows Server Failover Clustering (WSFC) nodes and, possibly, across multiple subnets. On the network, anFCI appears to be an instance of SQL Server running on a single computer, but the FCI provides failover fromone WSFC node to another if the current node becomes unavailable.

AlwaysOn Availability GroupsAlwaysOn Availability Groups is an enterprise-level high-availability and disaster recovery solution introducedin SQL Server 2012 to enable you to maximize availability for one or more user databases. AlwaysOnAvailability Groups requires that the SQL Server instances reside on Windows Server Failover Clustering(WSFC) nodes.

Database mirroringAvoid using this feature in new development work, and plan to modify aplications that currently use thisfeature. We recommend that you use AlwaysOn Availability Groups instead. Database mirroring is a solutionto increase database availability by supporting almost instantaneous failover. Database mirroring can be usedto maintain a single standby database, or mirror database, for a corresponding production database that isreferred to as the principal database. For more information, see Database Mirroring (SQL Server).

Log shippingLike AlwaysOn Availability Groups and database mirroring, log shipping operates at the database level. Youcan use log shipping to maintain one or more warm standby databases (referred to as secondary databases)for a single production database that is referred to as the primary database. For more information about logshipping, see About Log Shipping (SQL Server).

SQL Server 2012 & AlwaysOn

SLA - what does this have to do with the DBA

• Production hours:− Hours in which the partition / table / database must be available

− May be different for different parts of a database, for example, depending on the application

• The percentage of time the service:− The percentage of time within (time range) when the service / partition / table /

database is available

• Hours reserved for downtime:− These advance hours of downtime (technical break) facilitate the work of users

− Methods Customer Support

− The response time from the HelpDesk

− DBA response time for an event

SLA - what does this have to do with the DBA

• Number of users on the system− Number of transactions processed per unit of time

• Acceptable performance levels for access to the various operations− Minimum time required to replicate the different servers

• Deadline for data recovery from failures− Accidental deletion of data

− Damage to the database

− SQL Server Crash

− OS Server Crash

• Time it takes to read the data on the web (eg read / write table sales) so that it was possible to continue the sale− Maximum amount of space

− Maximum amount of tables / databases

− Number of users in specific roles

Why SLA is so important?

• In fact, it's more than just a signed agreement between the client and your boss.

• It is also a contract that YOU need to meet

• If it's signed an agreement to zero downtime and zero data loss (abstraction?) Then you need to make sure that if corruption can fulfill this contract (change / delete data on purpose by the authorized user).

• If you can not meet the SLA, the business is exposed to downtime and data loss

• The end result is to submit your CV to a recruitment agency ...

Do you think you can meet your Service Level Agreement?

• You need to know what are the conditions / requirements for SLA if you meet them

• As you can accomplish if you do not know that there is an SLA?

• As you review the contract if you did not invite anyone to the meeting on the creation of a Service Level Agreement?

• The end result is to submit your CV to a recruitment agency ...

Do you think you can meet your SLA?

• The recovery plan looks great on paper - but if ever you test it?

• Suppose this situation:− We allow 15 minutes is not available for database size of 100 GB.

− We are able to within the last 15 minutes substitute a copy of the user database

− What will you do in case of damage to the database?

− What will you do in the event of disk failure?

− What will you do in case of burning the motherboard?

− What do you do when cutting the cable FC?

− How much time it will take to recover from a backup?

− How much time it will take to bring tape with backup from a second location 25 kilometers away in the city center at 14?

Do you still meet the SLA 15 minutes of downtime?

Summary

• Database mirroring

• Log Shipping

• Hot-add CPU

• Transactional Replication

• Failover clustering enhancements

• Peer-to-peer replication enhancements

• Clouds (Google, Azure, Amazon...)

Summary

• You need to know about the existence of SLA

• You must take part in a Service Level Agreement (requirements / features / technology)

• You need to have contingency plans - TESTED

• You must have knowledge of their responsibilities

• You must be able to meet the technical SLA

Resources

• Database mirroring

− http://www.sqlskills.com/blogs/paul/2007/10/11/SQLServer2008PerformanceBoostForDatabaseMirroring.aspx

− http://www.sqlskills.com/blogs/paul/2007/10/01/SQLServer2008NewPerformanceCountersForDatabaseMirroring.aspx

− http://www.sqlskills.com/blogs/paul/2007/09/27/SQLServer2008AutomaticPageRepairWithDatabaseMirroring.aspx

• Backup compression

− http://www.sqlskills.com/blogs/paul/2008/01/09/SQLServer2008BackupCompressionCPUCost.aspx

− http://www.sqlskills.com/blogs/paul/2007/09/20/SQLServer2008BackupCompression.aspx

• Hot-add CPU

− http://www.sqlskills.com/blogs/paul/2008/01/10/SQLServer2008HotAddCPUAndAffinityMasks.aspx

• DBCC CHECKDB

− http://www.sqlskills.com/blogs/paul/CategoryView,category,CHECKDB%2BFrom%2BEvery%2BAngle.aspx

• Failover clustering

− http://www.microsoft.com/windowsserver2008/failover-clusters.aspx

• Peer-to-peer replication

− http://www.sqlskills.com/blogs/paul/2007/12/07/SQLServer2008ConfiguringPeertoPeerReplication.aspx

AFTER SESSION {next contact}

• MAIL: KoprowskiT@windowslive.com

• MSG: KoprowskiT@windowslive.com

• SKYPE: tjkoprowski

• TWITTER @KoprowskiT

• SlideShare (post-sessions): http://www.slideshare.net/Anorak

BLOGS:

• ITPRO Anorak’s Vision: http://itblogs.pl/notbeautifulanymore/ [PL/EN]

• Volume Licensing Specialites: http://koprowskit.eu/licensing/ [PL/EN]

• My MVP Blog: http://koprowskit.eu/geek/ [PL/EN/ES]

PLEASE RATE MY SESSION

THANK YOU