Upload
others
View
12
Download
0
Embed Size (px)
Citation preview
Database Solutions Engineering
Dell Product Group
Mayura Deshmukh
April 2013
Dell Reference Configuration for a 12TB
Microsoft SQL Server 2012 Fast Track
Data Warehouse
A Dell Technical White Paper
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
ii
This document is for informational purposes only and may contain typographical errors and
technical inaccuracies. The content is provided as is, without express or implied warranties of any
kind.
© 2013 Dell Inc. All rights reserved. Dell and its affiliates cannot be responsible for errors or omissions
in typography or photography. Dell, the Dell logo, and PowerEdge are trademarks of Dell Inc. Intel and
Xeon are registered trademarks of Intel Corporation in the U.S. and other countries. Microsoft,
Windows, and Windows Server are either trademarks or registered trademarks of Microsoft Corporation
in the United States and/or other countries. Other trademarks and trade names may be used in this
document to refer to either the entities claiming the marks and names or their products. Dell disclaims
proprietary interest in the marks and names of others.
February 2013 | Rev 1.0
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
iii
Contents Executive Summary .................................................................................................... 4
FTDW Reference Architectures Using PowerEdge R720xd Server .............................................. 4
12TB Dell R720XD FTDW Reference Architecture ................................................................. 5
Hardware Components .............................................................................................. 5
Internal Storage Controller (PERC H710P Mini) Settings ...................................................... 7
Application Configuration .......................................................................................... 9
Capacity Details .................................................................................................... 10
Performance Benchmarking ...................................................................................... 11
Conclusion ............................................................................................................. 13
References ............................................................................................................. 14
Tables Table 1: Dell Fast Track Reference Architectures for PowerEdge R720xd Server ............................. 4
Table 2: Tested Dell FTDW Reference Architecture Components ................................................ 5
Table 3: Mount Point Naming and Storage Enclosure Mapping .................................................... 9
Table 4: Capacity Metrics .............................................................................................. 10
Table 5: Performance Metrics ......................................................................................... 11
Figures Figure 1: Proposed Dell Fast Track Reference Architecture ....................................................... 5
Figure 2: Memory Slot Locations ........................................................................................ 7
Figure 3: Virtual Disk Settings ........................................................................................... 7
Figure 4: Internal Storage Controller Settings ....................................................................... 8
Figure 5: RAID Configuration ............................................................................................ 8
Figure 6: Storage System Components ................................................................................. 9
Figure 7: SQLIO Line Rate Test from Cache (Small 5MB File) .................................................... 12
Figure 8: SQLIO Real Rate Test from Disk (Large 25GB File) .................................................... 12
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
4
Executive Summary
The performance and stability of any data warehouse solution is based on the integration between
solution design and hardware platform. Choosing the correct solution architecture requires balancing
the application’s intended purpose and expected use with the hardware platform’s components. Poor
planning, bad design, and misconfigured or improperly sized hardware often lead to increased costs,
increased risks and, even worse, unsuccessful projects.
This white paper provides guidelines to achieve a compact, balanced, optimized 12TB Microsoft® SQL
Server® 2012 data warehouse configuration for Dell™ PowerEdge™ R720 and R720xd servers using
Microsoft Fast Track Data Warehouse (FTDW) principles. Benefits of implementing this reference
architecture include:
Achieve a balanced and optimized system at all levels of the stack by following hardware and
software best practices.
Avoid over-provisioning hardware resources to reduce costs.
Implement a tested and validated configuration with proven methodologies and performance
behaviors to help avoid the pitfalls of improperly designed and configured systems.
Easily migrate from a small- to medium-sized data warehouse configuration (5TB) to a large
data warehouse configuration (12TB).
Data center space comes at a premium. This configuration provides a compact, high-performance
solution for large data warehouses with 12TB of data or more.
FTDW Reference Architectures Using PowerEdge R720xd Server
The Microsoft FTDW reference architecture achieves an efficient resource balance between SQL Server
data processing capability and realized component hardware throughput to take advantage of improved
out-of-the-box performance.
As most data warehouse queries scan large volumes of data, FTDW system design and configuration are
optimized for sequential reads and are based on concurrent query workloads. Understanding
performance and maintaining a balanced configuration helps reduce costs by avoiding over provisioning
of components.
Dell provides various Fast Track reference architectures for SQL 2012 built using the Dell PowerEdge
12th Generation servers. These solutions are differentiated depending on the data warehouse capacity
and scan rate requirements. Table 1 summarizes FTDW configurations with Dell R720XD server.
Table 1: Dell Fast Track Reference Architectures for PowerEdge R720xd Server
Solution ID Server CPU Data Drives Rated Capacity
2457176 R720XD (2) Intel® Xeon® E5-2643 CPU @3.3GHz
(18) 900GB 10K SAS
12TB
The 12TB R720XD configuration described in this white paper is also available as a rapid deployment,
with hardware, software, and services included in the Dell™ Quickstart Data Warehouse Appliance 2000
(QSDW 2000). This configuration provides a low-cost and easier migration path for customers who want
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
5
to go from a 5TB to 12TB solution. For more information on Dell QSDW 2000, see Dell Quickstart Data
Warehouse Appliance.
12TB Dell R720XD FTDW Reference Architecture
The following sections of this paper describe the hardware, software, capacity, and performance characteristics of a 12TB Microsoft SQL Server 2012 FTDW solution with scan rates of about 2GBps using PowerEdge R720XD servers.
Hardware Components
Redundant and robust tests have been conducted on PowerEdge servers to determine best practices
and guidelines for building a balanced FTDW system. Table 2 provides the detailed hardware
configuration of the reference architecture.
Figure 1: Proposed Dell Fast Track Reference Architecture
Tested Dell Fast Track Reference Architecture Component Details
Table 2: Tested Dell FTDW Reference Architecture Components
Component Details
Server PowerEdge R720xd
CPU (2) Intel® Xeon® E5-2643 CPU @3.3GHz (HT Enabled)
Number of sockets used 2
Total Number of CPU Cores 8
Memory 128GB RAM (8 X 16GB DDR3 DIMMs @1600MHz)
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
6
Internal Hard Drives
22x 900GB 10K 2.5” SAS (18 data, 2 logs, 2 staging)
2x 900GB 10K 2.5” SAS (2 Hot Spares)
2x 900GB 10K 2.5” SAS (2 drives with OS) rear bay
Operating System Microsoft Windows® Server 2008 R2 SP1 Enterprise Edition
Database Software Microsoft SQL Server 2012 Enterprise Edition
PowerEdge R720xd Server
The PowerEdge R720xd server is a two-socket, 2U high-capacity, multi-purpose rack server offering an
excellent balance of internal storage, redundancy, and value in a compact chassis. For technical
specifications of the R720xd server, see the Power Edge R720xd Technical Guide.
Processors
The Fast Track Data Warehouse Reference Guide for SQL Server 2012 describes how to achieve a
balance between components such as storage, memory, and processors. To balance available internal
storage and memory for the PowerEdge R720xd, the architecture uses two Intel Xeon E5-2643 four-core
processors operating at 3.3GHz.
Memory
For SQL Server 2012 reference architectures, Microsoft recommends using 128GB to 256GB of memory
for dual-socket configuration. Selection of memory DIMMS will also play a critical role in the
performance of the entire stack.
This configuration was tested with various memory sizes running at different speeds—for example,
192GB running at 1333MHz, 192GB running at 1600MHz, 112GB running at 1600MHz, and so on. Using
DIMMs with memory rate of 1600MHz showed significant performance improvement (about 400MBs/s)
over DIMMS with memory rate of 1333MHz. In the test configuration, the database server is configured
with 128GB of RAM running at 1600 MHz to which create a well-balanced configuration.
To achieve 128GB of RAM on the PowerEdge R720xd server, place eight 16GB RDIMMS in slots A1-A4 and
B1-B4 (white connectors). See Figure 2: Memory Slot LocationsFigure 2 for memory slot locations.
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
7
Figure 2: Memory Slot Locations
Internal Storage Controller (PERC H710P Mini) Settings
The Dell PERC H710P Mini is an enterprise-level RAID controller that provides disk management
capabilities, high availability, and security features in addition to improved performance of up to
6GB/s throughput. Figure 3 shows the management console accessible through the BIOS utility.
Figure 3: Virtual Disk Settings
Stripe element size
By default, the PERC H710P Mini creates virtual disks with a segment size of 64KB. For most workloads,
the 64KB default size provides an adequate stripe element size.
Read policy
The default setting for the read policy on the PERC H710P Mini is “adaptive read ahead.” This
configuration was tested with “adaptive read ahead,” “No read ahead,” and “Read Ahead” settings.
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
8
During testing, it was observed that the default setting of “adaptive read ahead” gave the best
performance.
Figure 4: Internal Storage Controller Settings
RAID configuration
When deploying a new storage solution, selecting the appropriate RAID level is a critical decision that
impacts application performance. The FTDW configuration proposed in this paper uses RAID 1 disk
groups for database data files and database log files, nine RAID 1 data disk groups, and one RAID 1 log
disk group, each created with a single virtual disk. Additionally, two drives in RAID 0 are assigned as a
staging area. Figure 5 shows the proposed RAID configuration.
Figure 5: RAID Configuration
RAID 1Data 1
RAID 1Data 2
RAID 1Data 3
RAID 1Data 4
RAID 1Data 5
RAID 1Data 6
RAID 1Data 7
RAID 0Stage
RAID 1Data 9
RAID 1Data 8
RAID 1Logs
Hot Spares
OS
H710P Mini Monolithic
Rear Bay Drives
Drive slot configuration:
Slots 0-17: Nine RAID 1 disk groups were created, each configured with a single virtual disk
dedicated for the primary user data
Slots 18-19: One RAID 1 disk group created from two disks and a single virtual disk dedicated to
host the database log files
Slots 20-21: RAID 0 disk group created from two disks dedicated for staging
Slots 22-23: Remaining two disks assigned as global hot spares
Slots 24-25 (rear bay drives): One RAID 1 disk group for operating system
For FTDW architectures, it is recommended to use mount-point rather than drive letters for storage
access. It is also important to assign the appropriate virtual disk and mount-point names to the
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
9
configuration to simplify troubleshooting and performance analysis. Mount-point names should be
assigned in such a way that the logical file system reflects the underlying physical storage enclosure
mapping. Table 3 shows the virtual disk and mount-point names used for the specific reference
configuration and the appropriate storage layer mapping. All of the logical volumes are mounted to the
C:\FT folder.
Table 3: Mount Point Naming and Storage Enclosure Mapping
Disk Group
Virtual Disk
Virtual Disk Label Logical Label
Full Volume Path Capacity
1 1 Cage1-Card1-vData1 Data1 C:\FT\PRI\Cage1-Card1-vData1 837.75 GB
2 2 Cage1-Card1-vData2 Data2 C:\FT\PRI\Cage1-Card1-vData2 837.75 GB
3 3 Cage1-Card1-vData3 Data3 C:\FT\PRI\Cage1-Card1-vData3 837.75 GB
4 4 Cage1-Card1-vData4 Data4 C:\FT\PRI\Cage1-Card1-vData4 837.75 GB
5 5 Cage1-Card1-vData5 Data5 C:\FT\PRI\Cage1-Card1-vData5 837.75 GB
6 6 Cage1-Card1-vData6 Data6 C:\FT\PRI\Cage1-Card1-vData6 837.75 GB
7 7 Cage1-Card1-vData7 Data7 C:\FT\PRI\Cage1-Card1-vData7 837.75 GB
8 8 Cage1-Card1-vData8 Data8 C:\FT\PRI\Cage1-Card1-vData8 837.75 GB
9 9 Cage1-Card1-vData9 Data9 C:\FT\PRI\Cage1-Card1-vData9 837.75 GB
10 10 Cage1-Card1-vLog Log C:\FT\LOG\Cage1-Card1-vLog 837.75 GB
11 11 Cage1-Card1-Stage Stage C:\FT\Stage\Cage1-Card1-Stage 1675.5 GB
Figure 6 represents the storage system configuration for the proposed FTDW reference architecture.
Figure 6: Storage System Components
Data file 1 - 9
User database Temp DB
SQL SERVER
Logs
Data file 1-9
Non-DB Stage
INTERNAL STORAGE
RAID 1
RAID 1
Virtual disk group 1-9
Virtual disk group 10
RAID 0Virtual disk group 11
The production, staging, and system temp databases are deployed per the recommendations provided
in the Fast Track Data Warehouse Reference Guide for SQL Server 2012.
Application Configuration
The following sections explain the settings applied to operating system and database layers.
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
10
Windows Server 2008 R2 SP1
Enable Lock Pages In Memory to prevent the system from paging memory to disk. For more
information, see How to: Enable the Lock Pages in Memory Option.
SQL Server Configuration
The following startup options were added to the SQL Server Startup options:
-E: This parameter increases the number of contiguous extends that are allocated to a
database table in each file as it grows to improve sequential access.
-T1117: This trace flag ensures the even growth of all files in a file group when auto growth is
enabled. It should be noted that the FTDW reference guidelines recommend pre-allocating the
data file space rather than allowing auto-grow.
SQL Server Maximum Memory: FTDW for SQL Server 2012 guidelines suggest allocating no more than
92% of total server RAM to SQL Server. If additional applications will share the server, then adjust the
amount of RAM left available to the operating system accordingly. For this reference architecture, the
maximum server memory was set at 119808 MB (117GB).
Resource Governor: For SQL Server 2012, Resource Governor provides a maximum of 25% of SQL Server
memory resources to each session. The Resource Governor setting can be used to reduce the maximum
memory consumed per query. While it can be beneficial for many data warehouse workloads to limit
the amount of system resources available to an individual session, this is best measured through
analysis of concurrent query workloads. This configuration was tested with both 25% and 19% memory
grant, and the 25% setting was found to be optimal for the proposed configuration. For more
information, see Using the Resource Governor.
Max Degree of Parallelism: The SQL Server configuration option Max degree of parallelism controls
the number of processors used for the parallel execution of a query. For the configuration, settings of
12 and 0 were tested. The default setting of 0 provided maximum performance benefits. For more
information, see Maximum degree of parallelism configuration option.
Capacity Details
Table 4Table 4 shows the capacity metrics reported for the recommended reference configuration.
Table 4: Capacity Metrics
Metric Value Description
Raw Data Space (GB) 7695 Raw mirrored/striped space allocated for data
Raw User Database Space (GB)
5771.3 Raw user space (without compression) available
after allocating space for tempdb
Maximum User Database Capacity (TB)
19
Raw user space with compression (compression factor=3.5).
This is an estimate for the largest amount of user data the system will hold.
FTDW Rated Data Warehouse Capacity (TB)
12 This capacity rating is based on “up-to” capacity
but adjusted to account for FTDW rated I/O.
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
11
Performance Benchmarking
Microsoft FTDW guidelines help to achieve optimized database architecture with balanced CPU and
storage bandwidth. Table 5 shows the performance numbers reported for the recommended reference
configuration.
Table 5: Performance Metrics
Metric Value Description
FTDW Rated I/O (MB/s) 1909 Core performance metric for validation; average
of physical and logical I/O
Benchmark Scan Rate Logical (MB/s)
2164 Reflects actual user query throughput, which
includes reads from RAM/Buffer cache
Benchmark Scan Rate Physical (MB/s)
1654 Reflects physical I/O read from disk during
benchmark
FTDW Peak I/O (MB/s) 3481 Maximum observed I/O rate
FTDW Rated CSI (MB/s) 4337.5 Represents potential throughput using
Columnstore Index
The following sections describe the detailed performance characterization activities carried out for the
validated Dell Microsoft FTDW reference architecture.
Baseline Hardware Characterization Using Synthetic I/O
The goal of hardware validation is to determine actual baseline performance characteristics of key
hardware components in the database stack to ensure that system performance is not bottlenecked in
intermediate layers.
The disk characterization tool, SQLIO, was used to validate the configuration. The results in Figure 7
show the maximum baseline that the system can achieve from a cache (called Line Rate). A small file is
placed on the storage, and large sequential reads are issued against it with SQLIO. This test verifies the
maximum bandwidth available in the system to ensure no bottlenecks are within the data path.
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
12
Figure 7: SQLIO Line Rate Test from Cache (Small 5MB File)
POWER EDGE R720-XD
Windows Server 2008 R2 SP1
SQL Server 2012
Intel E5-26434 core CPU
Single RAID 1 Disk GroupSynthetic I/O rate: 375 MB/s
PERC H710P Mini ControllerSynthetic I/O rate: 2674 MB/s
SQL Server 2012 EnterpriseDual Socket Intel Quad core E5-2643
Aggregate Synthetic I/O rate: 2674 MB/s Intel E5-26434 core CPU
PERC H710P Mini
Controller
INTERNAL STORAGE
RAID 1RAID 1RAID 1RAID 1RAID 1RAID 1RAID 1RAID 1RAID 1
RAID 0RAID 1
The second synthetic I/O test with SQLIO was performed with a large file to ensure reads are serviced
from the storage system hard drives instead of from cache. Figure 8 shows the maximum real rate that
the system is able to provide with sequential reads.
Figure 8: SQLIO Real Rate Test from Disk (Large 25GB File)
POWER EDGE R720-XD
Windows Server 2008 R2 SP1
SQL Server 2012
Intel E5-26434 core CPU
Single RAID 1 Disk GroupSynthetic I/O rate: 294 MB/s
PERC H710P Mini ControllerSynthetic I/O rate: 2616 MB/s
SQL Server 2012 EnterpriseDual Socket Intel Quad core E5-2643
Aggregate Synthetic I/O rate: 2613 MB/s Intel E5-26434 core CPU
PERC H710P Mini
Controller
INTERNAL STORAGE
RAID 1RAID 1RAID 1RAID 1RAID 1RAID 1RAID 1RAID 1RAID 1
RAID 0RAID 1
FTDW Database Validation
The performance of a FTDW database configuration is measured using two core metrics: Maximum CPU
Consumption Rate (MCR) and Benchmark Consumption Rate (BCR).
MCR - MCR indicates the per-core I/O throughput in MB or GB per second. This is measured by
executing a pre-defined query against the data in the buffer cache, and then measuring the
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
13
time taken to execute the query against the amount of data processed in MB or GB. For the
validated configuration with two Intel E5-2643 four-core processors, the system aggregate MCR
was 2488 MB/s. The realized MCR value per core was 311 MB/s.
BCR - BCR is calculated in terms of total read bandwidth from the storage hard drives—not
from the buffered cache as in the MCR calculation. This is measured by running a set of
standard queries specific to the data warehouse workload. The queries range from I/O
intensive to CPU and memory intensive, and provide a reference to compare various
configurations. For the validated FTDW configuration, the aggregate BCR was 1909 MB/s.
During the evaluation cycle, the system configuration was analyzed for multiple query variants
(simple, average, and complex) with multiple sessions and different degrees of parallelism
(MAXDOP) options to arrive at the optimal configuration. The evaluation results at each step
were validated and verified jointly by Dell and Microsoft.
FTDW Database Validation with Column Store Index (CSI)
SQL Server 2012 implements CSI technology as a nonclustered indexing option for pre-existing tables.
Significant performance gains are often achieved when CSI query plans are active, and this
performance can be viewed as incremental to the basic system design.
After the test configuration was validated, CSI was added. Then, the same set of I/O and CPU-intensive
queries were executed to compare throughput achieved using CSI. Throughput rating of 4337.5 MB/s
was achieved for CSI-enhanced benchmarks. These numbers can be used to approximate the positive
impact to query performance expected under a concurrent query workload.
Conclusion
The Dell Microsoft FTDW architecture provides a uniquely well-balanced data warehouse solution. By
following best practices at all stack layers, a balanced data warehouse environment can be achieved
with a greater performance benefits than traditional data warehouse systems.
Dell Reference Configuration for 12TB Microsoft SQL Server 2012 Fast Track Data Warehouse
14
References
Dell SQL Server Solutions
www.dell.com\sql
Dell Services
www.dell.com\services
Dell Support
www.dell.com\support
Microsoft Fast Track Data Warehouse and Configuration Guide Information
www.microsoft.com/fasttrack
An Introduction to Fast Track Data Warehouse Architectures
http://msdn.microsoft.com/en-us/library/dd459146.aspx
How to: Enable the Lock Pages in Memory Option
http://go.microsoft.com/fwlink/?LinkId=141863
SQL Server Performance Tuning & Trace Flags
http://support.microsoft.com/kb/920093
Using the Resource Governor
http://msdn.microsoft.com/en-us/library/ee151608.aspx
Maximum degree of parallelism configuration option
support.microsoft.com/kb/2023536
Power Edge R720xd Technical Guide
http://www.support.dell.com/support/edocs/systems/per720/en/index.htm