1. TechEd 2013
12/28/ :16 PM
© 2013 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.

2. 12/28/ :16 PM
DBI-B311
Performance Tuning Microsoft SQL Server in Windows Azure Virtual Machines
Madhan Arumugam Ramakrishnan
Program Manager
Microsoft SQL Server
© 2013 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.

3. Performance Guidance for SQL Server in Windows Azure Virtual Machines
Whitepaper
Performance Guidance for SQL Server in Windows Azure Virtual Machines
Published: June 2013
Download it here:
nkId=306266

4. Performance tuning Analyzing performance impact
Is it a problem with my SQL application? or
Is it a problem with how I configured the Azure IaaS platform for my usage
SQL Performance considerations
Platform characteristics
Best practices
Monitoring
Troubleshooting

5. SQL performance considerations
KPIs
Throughput
Response time (aka latency)
Dimensions
Know what is the Platform impact vs Application impact
Metric
OLTP DW
Log
Read/Write mix
Mostly reads, smaller # of rows at a time
Scan intensive, large portions of data at a time, bulk loading
Mostly writes, requires low latency
IO size and pattern
Between 8 and 64K, mostly random
1 64KB read per 8 512KB reads, Mostly sequential
MB/s a critical metric
Highly sequential
# users
high
low
n/a

6. Windows Azure IaaS Characteristics

7. … … … Windows Azure Cluster 1 Cluster 2 Cluster n Datacenter 1
Datacenter n
Datacenter network
Cluster 1
Cluster 2 …
Cluster n …
Region 2
Region 1

8. Windows Azure Infrastructure Services
Implicit Cloud Service
Virtual Machine VM
Cloud Service
Worker Role
Web Role
VM1
VM2
VM3
VM4
VM5
VM…
Instances
Roles
Cloud Service
Virtual Machine VM
Cloud Service is a management, configuration, security, networking and service model boundary
Stateless Roles : Web/Worker Role – requires 2 or more instances
Persistent Roles : Virtual Machine can work with single instance

9. Windows Azure Infrastructure Services
VM2
LB/IP
VM1
PORT 3389
PORT 5586
PORT 5587
Single Public IP Per Cloud Service
Cloud Service
Endpoint
Public Port
Local Port
Protocol (TCP/UDP)
Name
Cloud Service
VM1
LBHTTP
VM2
VM3 LB
Endpoint Set
Public Port
Local Port
Protocol (TCP/UDP)
Name
PORT 80
Port Forwarding
Load Balanced Sets
Custom Load Balancer Health Probes -Health check with probe timeouts, HTTP based probing, allowing granular control of health checks

10. Inside the Windows Azure VM
No standard cached images for IaaS
OS is faulted in from blob storage during boot
On first boot
Windows Sysprep /specialize
SQL Sysprep completion
Virtual Machine
C:\
OS Disk
E:\, F:\, etc.
Data Disks
D:\
Temporary Disk Dynamic VHD
RAM Cache
Local Disk Cache
Blobs
Blob
Striped Volume

11. VM Disk Types & Configurations
OS disk (persistent)
Dynamic 127GB disk optimized for OS access patterns (e.g. boot up)
Data disk (persistent)
A VHD you can attach to a VM to store app data
Up to 1TB in size
Up to 16 disks for XL VMs
Temporary local disk (non-persistent)
Used for transient/temporary data storage & OS page files
Hosted in attached disks on physical host
Cleaned up in case of a VM failure or recycling
Physical disks shared across other VMs on same physical machine
Not recommended for user or system database files

12. Windows Azure VM Size & Bandwidth
Virtual Machine Size
CPU Cores
Memory
Disk Space for Virtual Machines
Allocated Bandwidth (Mbps)
Maximum data disks (1 TB each)
Maximum IOPS (500 maximum per disk)
ExtraSmall
Shared
768 MB
20 GB 5 1
1x500
Small
1.75 GB
70 GB
100 2
2x500
Medium
3.5 GB
135 GB
200 4
4x500
Large
7 GB
285 GB
400 8
8x500
ExtraLarge
14 GB
605 GB
800 16
16x500 A6
28 GB
1,000 A7
56 GB
2,000
Source:

13. VM Disk IO subsystem
Disks implemented as a shared multi-tenant service
Built-in triple redundancy, optional geo-redundancy
Performance more variable than on-prem
Host machines, storage services, network bandwidth shared between subscribers
Perf can depend on where and when VM is provisioned
Subject to maintenance operations
Granular control & configurability vs. cost, simplicity, out of box redundancy
Storage Stamp
Stream Layer
Partition Layer
Front-ends LB
Intra-stamp replication
Geo-replication
Storage Location Service

14. VM Disk Caching Caches VM data inside physical host machine
Can reduce disk I/O latency by reducing # transactions against Windows Azure Storage
2-tier cache
Recently accessed data stored in host RAM cache - space shared by all VMs on machine
Less recently accessed data stored on local hard disks of physical machine.
Reserved cache space for VM “OS Disk” and “Data Disks” based on the VM size

15. Default VM cache settings
Read Only: All requests cached for future reads. All writes persisted directly to Windows Azure Storage
Read Write: Reads and writes cached for future reads. Non-write-through writes persisted to local cache first. For SQL Server, writes are persisted to WA storage because it uses Write-through
- Lowest disk latency for light workloads
None (disabled): Bypasses cache. All disk transfers persisted to Windows Azure Storage
Highest I/O rate for I/O intensive workloads
Also consider TX cost
Disk type
Read Only
Read Write
None (disabled)
OS disk
Supported
Default mode
Not supported
Data disk
Supported (up to 4)
Temporary disk
Implemented using local attached storage

16. Best Practices

17. VM Disk cache setting recommendations
OS Disk
“Read Write” (default) reduces read latency for IO intensive workloads with smaller DBs (<=10GB)
Working set can fit in disk cache or memory, reducing blob storage IO
Data disks
Recommended for DBs > 10GB.
Cache setting depends on the IO pattern and workload intensity
Use default of “None” (disable) for Higher rate of random IOs (e.g. OLTP) & higher throughput
Bypasses physical host local disks, maximizing IO rate
Consider enabling read cache for latency sensitive read heavy workloads

18. Single Data Disk Configuration
Recommended for <1TB storage
Acceptable performance
Minimal complexity, simpler recovery
Sample SQL IO Measurement tests for single disk:
Random I/O (8KB Pages)
Sequential I/O (64KB Extents)
Sequential I/O (256KB Blocks)
Reads
Writes
IOPS
500
300
Bandwidth
4 MB/s
30 MB/s
20 MB/s
70 MB/s

19. Multiple Disk Configuration
Recommended for >1TB DB files & higher IOPS/bandwidth
Config options
Use Database files and Filegroups and place DB files across multiple data disks
This showed the best performance in our testing
Create OS volume on top of multiple data disks (e.g. OS striped volume or WS2012 storage space)
Storage spaces recommended over OS striped volumes

20. Aggregated Measurement Samples
4 disks
Random I/O (8KB Pages)
Sequential I/O (64KB Extents)
Sequential I/O (256KB Blocks)
Reads
Writes
IOPS
2000
1300
700
1100
Bandwidth
20 MB/s
120 MB/s
80 MB/s
170 MB/s
270 MB/s
8 disks
Random I/O (8KB Pages)
Sequential I/O (64KB Extents)
Sequential I/O (256KB Blocks)
Reads
Writes
IOPS
4000
2500
2600
700
2200
Bandwidth
30 MB/s
150 MB/s
160 MB/s
170 MB/s
550 MB/s
16 disks
Random I/O (8KB Pages)
Sequential I/O (64KB Extents)
Sequential I/O (256KB Blocks)
Reads
Writes
IOPS
8000
2500
5000
700
2400
Bandwidth
60 MB/s
150 MB/s
300 MB/s
170 MB/s
600 MB/s

21. Demo - 1 data disk vs 3 data disks

22. What else affects VM data disk performance?
Disk warm-up
NTFS Allocation Unit Size
Single vs. multiple storage accounts with a single VM
Data in different blobs not written at the same time
BLOBs that make up the stripe set could be out of sync
Spread the load across multiple VHDs instead
Note a storage account has a limit of 20K tps

23. Data Compression
IO intensive workloads: fewer pages ->reduced IO

24. Reduce IO with Instant File Initialization
Not default in Azure VM images
Reduces IO for
Creating a DB
Restoring a DB
Adding files to a DB
Extending file size
Autogrow, etc.
Add SQL service account to Perform Volume Maintenance Tasks security policy
Restart SQL Server

25. Should tempdb go on D: drive?
Short answer: No
Why?
Predictable performance: OS or data disk can provide same or better performance but D: drive can be more variable, being a physical disk sharing IO with other VMs on the host. Size and performance also depends on VM size
Configuration overhead: SQL Server has to recreate tempDB in D: if VM goes down – SQL Server service account requires Admin privileges. If stored in a separate folder this needs to be created at startup.
tempDB can be critical to application performance
Follow tempDB IO best practices

26. Demo - TempDB Contention and Resolution

27. Monitoring

28. Windows Azure Storage Analytics Metrics
Tracks aggregated storage usage for Blobs, Tables and Queues
Capacity – e.g. #containers, total #blobs
Requests - #requests, total ingress/egress, average E2E latency and server latency, total # failures by category, etc.
Access via storage account namespace
Enable in portal or using Set Blob Service Properties (REST API)
Set retention policy
See Windows Azure Storage Metrics: Using Metrics to Track Storage Usage

29. Performance Charts on the WA Portal
VM Dashboard
Monitor tab for storage account
Enabled under the “configure” tab
VM read and write to their VHDs using GetBlob and PutPage commands respectively

30. Other platform monitoring tools
Tools to determine network latency impact
Psping (free download at technet )
Traceroute (tracert)

31. SQL tools Tools to determine IO capacity of VM configurations
SQLIO – Disk Subsystem Benchmark Tool
Performance metrics
DMVs

32. Key SQL Perf counters Typical SQL KPIs Typical Web App KPIs
Max val for \Process(SQLServ)\% Processor Time
Avg val for \Process(SQLServ)\% Processor Time
Max val for \Processor(_Total)\% Processor Time
Avg val for \Processor(_Total)\% Processor Time
Max val for \SQLServer:SQL Statistics\Batch Requests/sec
Avg val for \SQLServer:SQL Statistics\Batch Requests/sec
Typical Web App KPIs
Max val for \ASP.NET Applications (_Total_)\Reqests/sec
Avg val for \ASP.NET Applications (_Total_)\Reqests/sec
Avg val for \Memory\Available Mbytes
Max val for \Processor(_Total)\% Processor Time
Avg val for \Processor(_Total)\% Processor Time
Avg val for \ASP.NET\Request Wait Time
Avg val for \ASP.NET\Request Execution Time
Avg val for \ASP.NET\Requests Queued
Avg val for \ASP.NET\Requests Rejected
Avg val for \ASP.NET\Requests Current
Typical User/test characteristics
Number of concurrent users
Average/Max request execution time
Number of web servers
Ramp up period, test method
Start and end time of test

33. Troubleshooting

34. Classic SQL Server Performance Factors
Plan change/choice issues
Software/hardware configuration
Locking & latching
Multi-user operations and blocking
Checkpoint & system operations

35. High-level Troubleshooting Steps
Define KPIs to monitor resource utilization
Monitor KPIs to track utilization over time
Examine trends and patterns as workload increases
Monitor DMVs to understand resource contention/waits
Monitor spinlock and back-off events

36. Troubleshooting Common VM Issues
KPIs To Monitor
Actions to Consider
CPU at or near 80%
% Processor Time (_Total)
SOS_SCHEDULER_YIELD waits
Increase instance size
Identify top consuming queries and tune
Load balance (e.g. move DB to another instance)
Near I/O capacity limits or
IO Latency Increases
Average disk reads per second
Average disk writes per second
Disk reads per second
Disk writes per second
io_virtual_file_stats
PAGEIOLATCH waits
SQL Server: Buffer Manager\Page Life Expectancy
Check Page Life Expectancy counter, for mem pressure. Increase instance size
Identify which DB and log files have I/O bottleneck
Add more data disks and separate data files if near IOPS limits per disk
Tune queries to reduce reads and writes
Consider enabling row or page compression
Memory resource pressure
Memory: Available Bytes
Memory: Pages per second
Process: Working Set (for SQL Server)
RESOURCE_SEMAPHORE waits
Check max server memory setting for SQL Server.
Use high memory instance
Identify SQL component (such as, CLR, high memory grants for app queries, et.), tune appropriately.

37. Key takeaways
Evaluate SQL Server in Windows Azure Infrastructure Services – SQL Server 2014 CTP1 available now!
Read the Performance Guidance for SQL Server in Windows Azure Virtual Machines white paper and follow best practices described there
Identify optimal VM size for your workload
Optimize for reduced IO and network round trips
Plan and test for IO perf variability
Identify your KPIs to monitor
Revisit optimization decisions as workload grows

38. Further Reading TechEd EMEA 2013 Sessions
White paper: Performance Guidance for SQL Server in Windows Azure Virtual Machines
SQL IaaS Basics
SQL Server in Windows Azure Virtual Machines
SQL Server HA/DR on IaaS
Windows Azure Storage
Windows Azure’s Flat Network Storage and 2012 Scalability Targets
Windows Azure Storage: A Highly Available Cloud Storage Service with Strong Consistency
Erasure Coding in Windows Azure Storage
SQL Server Backup and Restore with Windows Azure Blob Storage Service
SQL Server Performance
Analyzing I/O Characteristics and Sizing Storage Systems for SQL Server Database Applications
Compilation of SQL Server TempDB IO Best Practices
Windows Azure SQL Database and SQL Server -- Performance and Scalability Compared and Contrasted
TechEd EMEA 2013 Sessions
MDC-B212: Infrastructure Services on Windows Azure: Virtual Machines and Virtual Networks with Mark Russinovich Today 10:15AM, Hall 7.2
DBI-B206: Private Cloud – Better Together: Microsoft SQL Server 2012, Windows Server 2012 and System Center 2012 Today 5PM, N104
WAD-B402: Windows Azure Internals with Mark Russinovich Today 3:15PM, Hall 7.2
DBI-B329: Overview, Best Practices and Lessons Learned on Microsoft SQL Server in Windows Azure Virtual Machines June 27 10:15AM, N102
DBI-B306: Microsoft SQL Server High Availability and Disaster Recovery on Windows Azure June 27 3:15PM, N104

39. Resources Learning TechNet msdn http://channel9.msdn.com/Events/TechEd
12/28/ :16 PM
Resources
Learning
Sessions on Demand
Microsoft Certification & Training Resources
TechNet
msdn
Resources for IT Professionals
Resources for Developers
© 2013 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.

40. Evaluate this session Scan this QR code to evaluate this session.
12/28/ :16 PM
Required Slide
*delete this box when your slide is finalized
Your MS Tag will be inserted here during the final scrub.
Evaluate this session
Scan this QR code to
evaluate this session.
© 2013 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.

41. 12/28/ :16 PM
© 2013 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.
© 2013 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.