Microsoft Azure SQL Database Performance and Scale Out Deep DiveJan EngelsbergJoseph Idziorek

DBI-B313

Agenda

Predictable Performance + Elastic Scale

DEMO

Scalability options in Azure SQL Database

Scalability options in Azure SQL DatabasePredictable Performance: Scale-up or scale-down (Vertical)• Change service-tiers for a given database as capacity/performance needs fluctuate

Elastic Scale: Scale-out or scale-in (Horizontal)• Add or remove databases as more or less capacity/performance as needed

Basic

Standard

Premium

Basic Basic Basic Basic Basic Basic

PremiumStandar

d

Scale out/in

Sca

le u

p/d

ow

n

Predictable PerformanceCase for Change

ConceptsChoosing Performance LevelsQuota Management

Case for Change

Web & Business• Price/storage• Throttling• Performance varies based on

neighboring tenants • Deprecated (September 2015)

Basic, Standard & Premium• Price/performance & features• Resource Governance• Predictable Performance • General Available (September

2014)• Backwards Compatible (except

federations)

Based on customer feedback we are moving to a model based on Predictable Performance

Predictable PerformanceWeb / Business Basic / Standard /

PremiumMachineCompute

Wri

tes R

ead

s

Memory

DB 1DB 2

DB 3

DB 4

DB 7

DB 5 DB 6

DB 8DB 9

MachineCompute

Wri

tes R

ead

s

Memory

DB 1DB 2

DB 3

DB 4

DB 7

DB 5 DB 6

DB 8DB 9

DB 1

DB 2DB 3

DB 4

DB 7

DB 5DB 6

DB 8

DB 9

Noisy neighbor!

Bounding boxes

eliminates noisy

neighbors

Roles and ResponsibilitiesAzure SQL Database

• Assure resource are available when they are requested

• Provide insights into and tools to measure resource consumption

• Provide guidance on how to use the new Service Tiers

Customers

• Chose the right Service Tier (features)

• Chose the right Performance Level

• Monitor and react to performance needs

Database Throughput Unit – DTU

Represents the relative power (resources) assigned to the database

Blended measure of CPU, memory, and read and write rates Compare the power across performance levels Simplifies talking about performance, think IOPS vs. %

Monitoring% of current Performance Level

Utilization

75

%

Read

50%

Write

50%

CPU

60%

Memory

Compute

Wri

tes R

ead

s

Memory

DB workload

Bounding Box

Azure SQL Database Benchmark – ASDB An example representing meaningful OLTP-

workloadUses six tables of varying sizes some of which are always larger than available memory and scale with the throughputUses nine transaction typesA transaction is a combination of multiple SELECT, DELETE, INSERT, UPDATE statements

New Performance Levels Summary

Basic Standard Premium

Performance Levels (DTU)

5 S0: 10S1: 20S2: 50

P1: 100 P2: 200 P3: 800

ASDB results 16,600 tx/hour S1: ,520 tx/minuteS1: ,940 tx/minuteS2: 2,570 tx/minute

P1: 105 tx/secondP2: 228 tx/secondP3: 735 tx/second

Maximum DB size 2GB 250GB 500GB

Price* per hour (month) $0.0069 (~$5) S0: $0.0208 (~$15)S1: $0.0417 (~$30)S2: $0.1042 (~$75)

P1: $0.6458 (~$465)P2: $1.292 (~$930)P3: $5.167 (~$3,720)

*Starting November 1st

RESULT

CPU

Writes

Reads

GovernorCPU

Writes

Reads

P1 Governor

Resource Governance

SELECT * FROM a JOIN b ON …

S2

Resource requests are being not rejected but being queued

Overloaded can result in long running transactions and command timeouts

CPU

Writes

Reads

Governor

master.sys.resource_statsBased on 5 minute averages

userdb.sys.dm_db_resource_statsBased on 15 second averages

Percentages relative to performance levelAccessible though Azure Portal

Allows to configure alerting!

Resource Monitoring

Migration from on-premise installationsYou have to test!

Migration from Web&BusinessUse master.sys.resource_stats

Switching between levelsUse userdb.sys.dm_db_resource_stats

Query tuningsys.dm_exec_query_statssys.dm_exec_query_sql_text()sys.dm_exec_query_plansys.dm_exec_requestssys.dm_exec_sessions

Choosing Performance Levels

5 10 20 50 100 200 800DTUs

Changing Performance LevelPowerShell

Set-AzureSqlDatabase

RESTUpdate Database / ServiceLevelObjectiveId

.NETMicrosoft.WindowsAzure.Management.Sql.SqlManagementClientclient.Databases.Update(…)

T-SQLALTER DATABASE … MODIFY (EDITION = …)

Quota ManagementServer Quota

6 Servers per Subscription

Per Server DB Quota1600 DTU150 databases

Examples2 P3 = 2x800 DTU = 1600 DTU160 S0 = 160x10 DTU = 1600 DTU200 S2 = 200x50 DTU = 10000 DTU1 P3, 14 S2,10 S0 … = 1600 DTU / 25 DBs

DEMO

Performance PredictabilityMonitoring

Elastic ScaleCustomer scenarios; Elastic Scale Public Preview

Customer Scenario

• Application exceeds the limits of a single database: capacity, performance, geo-locality, isolation

• Application Types• Single tenant: Each tenant’s data is stored in a different database• Multi-tenant: Multiple tenants share the same database• Hybrid model: Some tenants share databases, others get their own

database• Temporal Data: Telemetry ingestions/Internet of Things (IOT)

scenarios• Geo-location/Geo-political: Keeping data in geographical regions

for performance and/or policy reasons

Terminology

(1) Database

(2) Sharding Key

. . .

(3) Shard Map Manager

(4) Shard

(4) Shard

(5) Shard Set

(6) Sharded Table

(7) Reference Table

(8) Shardlet

Customer ID

Name

1 Alice

2 Bob

Customer Table

Data Center ID DC Name

1 Boston

2 Miami

Data Center Table

Elastic Scale Overview

Application

Developer

.NET Application

Shard Map Management (SMM)• Define groups of shards for your application• Manage mapping of routing keys to shards

Data Dependent Routing (DDR)• Route incoming requests to the correct

shard, e.g., given a customer ID • Ensure correct routing as tenants move • Cache routing information for efficiency

Multi-Shard Query (MSQ)• Interactive processing across several shards • Same statement executed on all shards with

UNION all semantics

.NET Client APIs

Management Service

Admin/DevOps

Elastic Scale Overview

Application

Developer

.NET Application

Admin/DevOps

Management Service

Split Merge Service (SMS)• Grow or shrink capacity by adding or

removing scale units • Dynamically adjust scale factor of scale

unit• Trigger adjustment dynamically through

policies

Federations Migration Utility• Migrate from Federations to Elastic Scale

Shard Elasticity (SE)• Policy-based horizontal and vertical

sharding using Azure Automation

Management

• Two types of shard maps• Range: contiguous values• List: explicit values

• Four types of sharding keys• INT, BIGINT, GUID, VARBINARY

Elastic Scale Overview[shards_global]sid smi

d server databa

se

1 1 Server1

DB1

2 1 Server1

DB2

[shardmaps_global] smid name

1 RangeShardMap

[shard_mappings_global]mid smi

d min

max

Sid

1 1 0 100 1

2 1 100

200 2

Shard Set

DB1

[0-100) . . .

DB2

[100-200)

DB3

[200-300)

DB4

[300-400)

DB5

[400-500)

DB6

[500-600)

DBn

[n – n+100)

Shard Map Manager

Data Dependent Routing (DDR)

Client AppDDR APIs

Scenario: query a shard with a specific sharding key value

SELECT * FROM customers WHERE customer ID = 104

DB1

[0-100) . . .

DB2

[100-200)

DB3

[200-300)

DB4

[300-400)

DB5

[400-500)

DB6

[500-600)

DBn

[n – n+100)

Application

Developer

Shard Map Manager

Data Dependent Routing (DDR)

// Get a routed connection for a given shardingKeyusing (SqlConnection conn = ShardMap.OpenConnectionForKey(

shardingKey, connectionString /* Credentials Only */,ConnectionOptions.Validate /* Validate */ ));

{using (SqlCommand cmd = new SqlCommand()

{cmd.Connection = conn;cmd.CommandText = "SELECT * FROM customers WHERE customer ID = 104";

SqlDataReader sdr = cmd.ExecuteReader();

// Now consume results from the data reader…}

}

Caching: improve performance of shard operations • Global Shard Map (GSM) – state of all shards in the Shard Map• Local Shard Map (LSM) – state of all shards on a particular shard• Client Cache (eager/lazy) – state of all shards in the Shard Map/known

shards

Data Dependent Routing (DDR)

Client AppDDR APIs

Cache

Application

Developer

Shard Map Manager

GSM

DB

LSM

Multi-Shard Query

Scenario: execute a query across a set of shards (returns a UNION ALL result set)

Client AppMSQ APIs

SELECT count(*) FROM customers

UNION ALL result set

DB1

[0-100) . . .

DB2

[100-200)

DB3

[200-300)

DB4

[300-400)

DB5

[400-500)

DB6

[500-600)

DBn

[n – n+100)

Application

Developer

Shard Map Manager

Multi-Shard Query APIsusing (MultiShardConnection conn = new MultiShardConnection(m_shardMap.GetAllShards(null), MultiShardTestUtils.GetTestSqlCredential())){ using (MultiShardCommand cmd = conn.CreateCommand()) { cmd.CommandText = "SELECT count(*) FROM customers"; cmd.CommandType = CommandType.Text; cmd.Policy = MultiShardPolicy.PartialResults;

using (MultiShardDataReader sdr = cmd.ExecuteReader(includeShardNameColumn: true)) { while (sdr.Read()) {

// Now consume results from the data reader… }

} }}

Split Merge Service

Scenario: perform a split or merge action• Split: create two distinct shards from one• Merge: create one shard from two distinct shards• Point move: move a single tenant to an individual shardSplitMerge

Management Service

Admin/DevOps

DB1

[0-100) . . .

DB2

[100-200)

DB3

[200-300)

DB4

[300-400)

DB5

[400-500)

DB6

[500-600)

DBn

[n – n+100)

DB2.1

[0-200)

DB5.1

[400-450)

DB5.2

[450-500)

Scale Options

ALL shards are regular Azure SQL Databases• Scale up and down individual shards, predictable

performance• Feature support: backup, auditing, etc.• Tooling integration: SSMS, SSDT, portal

DB1

[0-100)

. . .

DB2

[100-200)

DB3

[200-300)

DB4

[300-400)DB5

[400-500)

DB6

[500-600) DBn

[n – n+100)

Basic Standard

Premium

Premium

Basic Standard

DEMO

Elastic Scale with Azure SQL DB

Related contentFind Me Later At. . .

Thursday, October 30 - 6:30 PM-8:00 PM

Ask the Experts, Hall 5

Azure SQL Databasehttp://azure.microsoft.com/en-us/documentation/services/sql-database/

Track resources

Elastic Scale Documentation Maphttp://azure.microsoft.com/en-us/documentation/articles/sql-database-elastic-scale-documentation-map/

Elastic Scale Getting Started Samplehttps://code.msdn.microsoft.com/vstudio/Elastic-Scale-with-Azure-a80d8dc6

Near Real-Time Performance Metricshttp://azure.microsoft.com/blog/2014/09/11/azure-sql-database-introduces-new-near-real-time-performance-metrics/

27 Hands on Labs + 8 Instructor Led Labs in Hall 7

DBI Track resources

Free SQL Server 2014 Technical Overview e-book

microsoft.com/sqlserver and Amazon Kindle StoreFree online training at Microsoft Virtual Academy

microsoftvirtualacademy.com Try new Azure data services previews!Azure Machine Learning, DocumentDB, and Stream Analytics

Resources

Learning

Microsoft Certification & Training Resources

www.microsoft.com/learning

TechNet

Resources for IT Professionals

http://microsoft.com/technet

Sessions on Demand

http://channel9.msdn.com/Events/TechEd

Developer Network

http://developer.microsoft.com

Please Complete An Evaluation FormYour input is important!TechEd Schedule Builder CommNet station or PC

TechEd Mobile appPhone or Tablet

QR code

Evaluate this session

© 2014 Microsoft Corporation. All rights reserved. Microsoft, Windows, 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.