DAT309

Implementing Scale-Out Solutions with Microsoft SQL Server 2005

Don VilenProgram ManagerSQL ServerMicrosoft Corporation – Redmond

Who Am I?

Program Manager on SQL Server’s Storage EngineWorking on SQL Server at Microsoft for 12+ yearsResponsible for Database Mirroring and High AvailabilityPreviously:

PM for Filestream, other features

Training for PSS Engineers on SQL Server around the world

DonV@microsoft.com

Technologies to Implement Scale-out

Which Technology to Use?

Classes of Data and Scale-out

Deployment Scenarios

Use Cases

Agenda

Scaling Database Systems

This presentation is about scalingSecond 10,000 users as fast as first 10,000With twice as much horsepower – or less

Two ways to scaleScale up: bigger hardware, softwareScale out: more small boxes

Scale-out has potential savingsScale-up: redundant big serverScale-out: 100% redundancy not required

Scale-Out For Everyone?

Scale-out is not the answer to every problemSimilar to dividing processing among threadsIf scale is not required, solution will just be:

Unnecessarily more complex

Slower

Some scaling technologies require changingPrograms

Procedures

Best to plan this into the design ahead of time

We’ll assume you’ve addressed basic tuningAppropriate indexes, stored procs, etc.

The Technologies

Linked Servers and Distributed Queries

DPV Distributed Partitioned View

SSD Scalable Shared Database

Replication

P2P Peer-to-Peer Replication

SSB Service Broker – Queuing

QN Query Notifications – Caching

DDR Data Dependent Routing

Linked Servers and Distributed Queries

Define Linked Servers and connection attributesUse 4-part names to reference <server>.<database>.<schema>.<object>Query optimizer will split query or DML into appropriate queries to be sent to each linked serverEach linked server processes its query and returns result setResult sets are combined / joined Distributed transactions for DML

Linked Servers and Distributed QueriesExecuting a Query

1. User issues query / DML with DPV

2. Sends appropriate query / DML to linked servers, eliminating those with no relevant data

3. Queries run asynchronously on involved linked servers

4. Data is returned by linked servers and combined

Distributed Partitioned View – DPV

Define sub-tables with appropriate constraints on the various Linked ServersCreate a UNION-ALL view SELECTing from all the sub-tablesQuery optimizer will split query or DML into appropriate queries to be sent to each linked server, with dynamic elimination of uninvolved partitionsEach linked server processes its query and returns result setResult sets are combined / joined Distributed transactions for DML

Distributed Partitioned View – DPV Setting it up

1 - 999 1000 - 1999 2000 - 2999

1. Create tables with range constraints

2. Create linked servers

3. Create view with UNION ALL

Distributed Partitioned ViewsExecuting a Query

1 - 999 1000 - 1999 2000 - 2999

1. User issues query / DML with DPV

2. Sends appropriate query / DML to linked servers, eliminating those with no relevant data

3. Queries run asynchronously on involved linked servers

4. Data is returned by linked servers and combined

Distributed Partitioned View

Performs the work across involved nodes asynchronouslyAdd additional nodes as needed

How it Provides Scale-out

1 - 999 2000 - 29991000 - 1999 3000 - 3999

Distributed Partitioned Views

Available since SQL Server 7.0SQL Server 2005 adds the asynchronous processing

Usable for read or writeRequires MSDTC for distributed transactionsConsiderations for scalability

All queries and updates are affectedResponse is determined by the slowest partLocks are held for whole transactionNot feasible across security/organization boundaries

Scalable Shared Database – SSD

One copy of database attached to multiple instancesNew in SQL Server 2005 – Enterprise Edition onlyRequires a SANDatabase itself is read-only

Refresh on a schedule

Rolling refresh possible

Sweet spot is connection by 2 to 8 instances

Scalable Shared Database – SSD Setting it up

1. Create database, fill with data, and detach

2. Set volumes Read Only

3. Attach the same database to multiple instances

4. Periodically attach refreshed database to instances, perhaps via SAN operations

Scalable Shared Database – SSD Executing a Query

User queries on multiple instances use same reference data from single copy

Scalable Shared Database – SSD Refreshing the Data

1. Create a Clone of hot data, with recoveryHot Data

2. Set Clone volume to Read Only

3. For each node: Detach Instance from Vn, and attach Vn+1

Clone

Vn+1 Vn 4. Refresh Vn, and create Vn+2

Scalable Shared Database

Multiple instances access the same database filesAdd additional nodes as needed

How it Provides Scale-out

Setting Up SSD

Build processUnmask SAN unit number to make volume accessibleMount volume as read-write with DISKPARTBuild or refresh one copy of the database

Attach processDetach the database from the SQL Server instance that built it, using the DETACH command

Mark each volume as read-only by using DISKPART and dismount the disk volume from the production server

Unmask the SAN logical unit number to make the database available to each reporting server

Mount the disk volume on each reporting server by using DISKPART

Attach to one or more instances of SQL Server

Replication

Provides for distribution of all or subset of dataAvailable with various capabilities since SQL Server 7.0All use Publisher – Distributor – Subscriber modelAll incur latency; data changes appear some time, perhaps seconds, minutes, hours, later

Snapshot ReplicationMerge ReplicationTransactional Replication

Snapshot Replication

Periodic refreshing of whole sets of dataEquivalent to BCP-out from Publisher and BCP-in to Subscriber(s)Perhaps some rows haven’t changed at all; others have changed perhaps millions of times

Publisher / Distributor

Subscriber(s)

Mechanism: BCP to copy whole sets of data

Merge Replication

Periodic exchange of changed rows in both directionsChanged rows are tracked; net changes are sent

No guarantee of transactional consistency

Can update in multiple placesBuilt-in conflict detection

Publisher / Distributor

Subscriber(s)

Mechanism: Track changed rows, then exchange net changes

Transactional ReplicationPeriodically sends individual changes, maintaining transactional integrityChanges are read from log, converted into individual T-SQL statements, and sent preserving order and transaction grouping

Guarantees transactional consistency

Publisher / Distributor

Subscriber(s)

Mechanism: Translate each log record to T-SQL, send in same order, and with similar transactions

Peer-to-Peer Replication – P2P

Variety of transactional replicationNew in SQL Server 2005 – Enterprise Edition only

All Publish their ‘owned’ data and subscribe to others’ data

Peer-to-Peer Replication – P2P

Variety of transactional replicationNew in SQL Server 2005 – Enterprise Edition only

Allows ring-type replication configurationMore flexible than hierarchical topologies

Allows Read and Write operations

More replication paths than bus-typeNo built-in conflict detection

Each row must have an "owner"

Replication Topologies

P2P and Connectivity

Peer-to-Peer Replication – P2P

Clients work with a server with a ‘local’ copy Add additional nodes to increase capacityAlso increases availability

How it Provides Scale-out

All Publish their ‘owned’ data and subscribe to others’ data

SQL Server Service Broker – SSB

Transactional messaging, asynchronous processing, and disconnected systems

Separates input from executionRouting among instances is built inProvides service-oriented infrastructure

Available in all SQL Server 2005 editionsEven SQL Server Express

Requires application changes / programming

Service Broker – SSB Setting it up

1. Enable Service Broker

2. CREATE QUEUE

3. Rather than performing all the processing, it SENDs a message

4. Message is routed to receiver who RECEIVEs it from the queue and processes it

Service Broker – SSB

Distributes the workProcessing is done elsewhere and/or laterCan specify an Activation procedure and number of instancesCan further divide the work across multiple servicesAdd more resources as needed

How it Provides Scale-out

Query Notifications – QN Notify my client / cache when row is changed

Notification only; client must refresh cached data

Implemented in-server using:Indexed View-based notification technologyService Broker for communications

Requires SQL Server 2005 – any editionRequires "new" client stackUsed in ASP.NET "Cache Sync"

Client cache

1. Client caches data and registers for notification

3. SSB message notifies client of a change to result set

2. Data is changed on server

Query Notifications – QN

Intelligent cache notifications avoid continuous polling to check for changesEffectively “scales-out” to the clients

How it Provides Scale-out

Client cache

Data Dependent Routing – DDR More architectural concept than featureNot built in to SQL Server 2005Can be implemented:

By application

With SSB using built-in routing service with custom code

With middle-tier component

Can be very helpful with: Transactional replication

Distributed Partitioned Views

Which Technology?

DPV SSD P2P SSB QN

Special Hardware No Yes No No No

Program Changes No No No Yes Maybe

OpsChanges Maybe Yes Yes No No

Read-Write Yes No Per-Row Yes Yes

IncreaseRobustness No Maybe Yes Yes No

DDRUseful Yes Maybe Yes Maybe Maybe

SQL ServerEditions No Express EE only EE only All* All

Technologies to Implement Scale-out

Which Technology to Use?

Classes of Data and Scale-out

Deployment Scenarios

Use Cases

Agenda

Data Usage-Based Scale-Out

Scaling choice is based on data usageClassify the dataDetermine the database deployment topologyDetermine the application topologyChoose the appropriate technology

Classifying Data

Four main data classificationsResource data: customers

Traditional "data processing" data

Reference data: catalogs, price listsLookup tables

Service interaction data Original order forms

Activity data Pick lists

Shopping carts

Traditional Deployment

Single-tier or client-serverAll data together in databaseAll data treated the sameAll data updated directlyService interaction is part of transaction

Deployment in Tiers

Multi-tiersReference Data can be cached and replicatedActivity and Service Interaction coupled

Reside in application servers

Direct interaction between activity, service, and resourceMulti-operation transaction

Deployment with Services

Some data separated into servicesActivity data internal but asynchronous

Not tightly coupled to resource data

Managed through messaging

Service interactions can be disconnectedManaged through messaging

Some data replicatedTraditional transactional replicationSmart, active cache

Only resources need synchronous processing

Scaling and Data Types

Techniques differ with dataReference data

Scalable Shared Database

Snapshot replication

Query notifications

Activity data Dynamic partitioning

Data dependent routing

Service interaction data Inherently asynchronous

Resource data Static partitioning

Peer-to-peer or bi-directional replication

Storage and Data Types

Techniques differ with dataReference data

Master copy in central database

Replicas online

Activity dataKeep copies for activity tracking, trending, mining

Service interaction dataKeep EXACT copies of request and response

For retry and recovery

Sometimes a legal requirement

Resource dataDivide processes

Updates can be async with other categories

Use CasesDistributed Partitioned Views

“SQL Server Megaservers: Scalability, Availability, Manageability” (Jim Gray and Richard Waymire)http://www.microsoft.com/technet/prodtechnol/sql/2000/plan/ssmsam.mspx

Scalable Shared DatabaseCorporate Phone Book, Product Catalog

Peer-to-Peer Replication“Inside Microsoft.com: Managing SQL Server 2005 Peer-to-Peer Replication” (David Lindquist)http://www.microsoft.com/technet/technetmag/issues/2006/07/InsideMSFT/default.aspx

Service Broker“Inside MSDN: Building the MSDN Content Aggregation System”(John Mollman)http://msdn.microsoft.com/msdnmag/issues/06/07/InsideMSDN/default.aspx

Query NotificationsCache information about server status, Windows mobile server update

Summary

Scale-out more achievable with SQL Server 2005Multiple features geared toward scale-outService-oriented, replicated, or distributed

Which technologies to use depends on:Deployment scenariosWhether app or process can be changedData classes used

Scale-out technologies can be combined Data Dependent Routing useful adjunct to all

Resources“Scaling Out SQL Server 2005” Roger Wolterhttp://msdn2.microsoft.com/en-us/library/aa479364.aspx

“Planning, Implementing, and Administering Scaleout Solutions with SQL Server 2005” (Bob Beauchemin)http://www.microsoft.com/technet/prodtechnol/sql/2005/implementingscaleout.mspx

Bob Beauchemin's Blog: http://www.SQLskills.com/blogs/bobb

“Architecting Service Broker Applications” (Roger Wolter)http://msdn2.microsoft.com/en-us/library/aa964144.aspx

“Achieving Scalability and Availability with Peer-to-Peer Transactional Replication” (Michael Blythe)http://www.microsoft.com/technet/prodtechnol/sql/2005/p2ptranrepl.mspx

Scalable Shared Databases Are Supported by SQL Server 2005http://support.microsoft.com/default.aspx?scid=kb;en-us;910378

“Service-Oriented Data Architecture: App-Server:Lite?” (David Campbell)http://research.microsoft.com/research/pubs/view.aspx?tr_id=983

“Data on the Outside vs. Data on the Inside: An Examination of the Impact of Service Oriented Architectures on Data” (Pat Helland)http://msdn2.microsoft.com/en-us/library/ms954587.aspx

Q&A

Resources

Technical Communities, Webcasts, Blogs, Chats & User Groupshttp://www.microsoft.com/communities/default.mspx

Microsoft Developer Network (MSDN) & TechNet http://microsoft.com/msdn http://microsoft.com/technet

Trial Software and Virtual Labshttp://www.microsoft.com/technet/downloads/trials/default.mspx

Microsoft Learning and Certificationhttp://www.microsoft.com/learning/default.mspx

SQL Down Under Code Camp

Biggest SQL Server community event in Australia14 SQL Server sessions over two daysSpecial guest presentation from Kevin Kline

President of PASS, Director of Technology – Quest Software

When?Weekend of the 13th and 14th of October

Where?Charles Sturt University, Wagga Wagga

What does it cost?This event is FREE

http://www.sqldownunder.com

SQL Server User Groups

Best place to learn about SQL Server for FREEUser Groups in most capital citiesGreat networking opportunityhttp://www.sqlserver.org.au

Call to Action

Visit the TLC Area to talk to SQL Server experts

Download the latest SQL Server “Katmai” CTP from: http://www.microsoft.com/sql/prodinfo/futureversion/default.mspx

SQL Server TechCenter: http://www.microsoft.com/technet/prodtechnol/sql/default.mspx

SQL Server DevCenter:http://msdn.microsoft.com/sql

SQL Server Best Practices:http://www.microsoft.com/technet/prodtechnol/sql/bestpractice/default.mspx

© 2007 Microsoft Corporation. All rights reserved. Microsoft, Windows, Windows Vista and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market

conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.