Documentos de Académico
Documentos de Profesional
Documentos de Cultura
1. Executive summary
Microsoft SQL Server 2012 Fast Track Data Warehouse (FTDW) Reference Architecture from
Huawei enables IT architects and managers to build optimized data warehouse solutions with
tested and tuned server, storage, network, and software configurations. The reference
architectures present implementation guidelines and best practices for a variety of data warehouse
configurations. This approach significantly reduces time of deployment, service and costs while
maximizing performance and reliability.
Leveraging decades of experiences in producing mission critical Information and Communication
Technology systems, Huaweis enterprise class servers, storage and networking products form a
solid foundation for deploying enterprise critical IT services such as data warehouses and business
analytics.
IT managers, database architects, CXOs interested to explore and deploy data warehouses and BI
applications could take advantage of Huaweis engineering, integration and optimization effort to
build their next data warehouses.
Proven Configurations: Microsoft SQL Server 2012 Fast Track Data Warehouse
Reference Architectures from Huawei documents the design, components, settings, and best
practice to deploy a Microsoft SQL Server 2012 Fast Track Data Warehouse. Users can now
procure the exact configuration of hardware and software and turn on new data warehouse
services in the shortest amount of time, with minimal risk.
Performance: The Reference Architectures provide the best practice implementation that has
been tested and tuned to deliver best performance. It also provides database management
suggestions to keep the data warehouse at its best performance level.
Costs: The Reference Architectures design has taken considerations of selecting the right
hardware components to maximize performance while keeping costs low.
Three configurations are included in this Reference Architectures document covering a broad
range of capacity and performance requirements.
enterprise.huawei.com
Published 1/11/2013
Page 1 of 21
Workload Assumptions
The Microsoft SQL Server Fast Track Data Warehouse for Huawei reference architectures
described in this document target scan-intensive, large-request, analytical data warehousing
workloads. Contrary to OLTP databases, this type of data warehouse workloads typically involve
much larger data requests and benefit from the increased total throughput potential of sequential
disk scans. With Fast Track data warehousing workloads an emphasis is placed on achieving
consistently high I/O scan rates (measured in MB/s) rather than the more traditional focus on
operations per second (measured in IOPS).
In order to enable data warehouses to support large volumes of data and wide-range of queries,
traditional use-case specific optimizations will only work to a certain extent but very often increase
complexity, costs, maintenance overhead and perhaps even reduce performance. Fast Track
Data Warehouse is a different way of optimizing for data warehouse workloads. By aligning
database files and configuration with efficient disk scan (rather than seek) access, performance
achieved from individual disks can be significantly higher. The resultant per-disk performance
increase reduces the number of disks needed to generate sufficient I/O throughput to satisfy the
ability of SQL Server to process data for a given workload. Furthermore, you can avoid some
index-based optimization techniques used to improve disk seek.
enterprise.huawei.com
Published 1/11/2013
Page 2 of 21
Tecal RH2288 V2 rack server is an enterprise class 2U rack server that is designed to
provide best performance and scalability in its class of servers. The RH2288 V2 can support up to
16 computing cores, a memory capacity of 768 GB, and a storage capacity of 24 TB. The RH2288
V2 is an ideal product for mission critical enterprise applications.
Published 1/11/2013
Page 3 of 21
The RH2288 V2 has multiple sensors at strategic internal locations that monitor the temperature of
key components and send the data to the basic input output system (BIOS) to dynamically adjust
the fan speed to cool the server precisely. This ensures that the server runs reliably while
consuming a minimum level of power.
The dynamic configuration technology on maximum power can limit the power of the data center
server within a preset range, without affecting the performance of the server.
Half-Length PCIe card that delivers maximum read IOPS Performance of 600K@4KB and
enterprise.huawei.com
Published 1/11/2013
Page 4 of 21
Industry-leading Reliability
Dynamic RAID algorithm used in all the channels, to share resources, and keep the reliability
with the multiple chip failure
Support TRIM & sophisticated wear algorithms, to enhance the recovery efficiency, reduce the
write wear, extend service life
Easy to Maintain
With solid state storage for user data, the FTDW reference architectures have many more
advantages besides higher performance, including simplified management, I/O resiliency, and
lower operational cost.
Published 1/11/2013
Page 5 of 21
state storage does not require fragmentation management. The complex steps in the DW load
process that often involves single threaded loading to avoid page fragmentation can be simplified,
and this will bring customers faster DW load time.
In addition, larger disk groups and lower volume counts can be used to achieve a given throughput
target, which further simples DW management in terms of filegroup creation and maintenance.
I/O Resiliency: With its exceptional random I/O performance, solid state storage has minimal
performance degradation under high concurrency (large user numbers) or page fragmentation.
Mixed random read (seek) workload does not negatively impact large request (scan) I/O patterns.
It enables real-time DW updates while the DW is in use with minimum query response time
degradation.
Lower Operational Cost: Less management efforts outlined above translates into lower
operational cost. Additionally, solid state storage offers a more efficient balance of I/O throughput
to capacity per unit. Usually fewer SSDs are required to achieve a given DW throughput target
when compared with HDDs, which reduces the power consumption, generates less heat (thus less
HVAC capacity), and often supports higher density solutions.
More about solid state storage in FTDW including configuration guidelines can be found in
Microsoft Fast Track DW Reference Guide for SQL Server 2012 at
http://msdn.microsoft.com/en-us/library/hh918452.aspx.
enterprise.huawei.com
Published 1/11/2013
Page 6 of 21
enterprise.huawei.com
Published 1/11/2013
Page 7 of 21
Descriptions
Server
Tecal 2288 V2
CPU
Memory
96GB DRAM
(4x 16GB DDR3 DIMMs and 4x 8GB DDR3 DIMMs @
1600MHz)
RAID Controller
Operating System
Database Software
RAID Configuration
Huawei SR620 is an enterprise-class RAID controller that provides disk management capabilities,
high-availability, as well as improved storage performance. The RAID configurations play an
important role of the data warehouse performance, and the following shows how the disk drives
are configured in Huawei 5TB FTDW RA.
18x 300GB disks for user data and TempDB: Nine RAID 1 groups are created, and each
configured with a single virtual disk.
4x 300GB disks for database logs: One RAID 10 group is created and configured with a single
enterprise.huawei.com
Published 1/11/2013
Page 8 of 21
virtual disk.
2x 600GB disks for Windows Server and SQL server: One RAID 1 group is created and
configured with a single virtual disk for OS and software installation.
Two additional 600GB disks are reserved as hot spare to further improve system availability.
enterprise.huawei.com
Published 1/11/2013
Page 9 of 21
Descriptions
Server
Tecal 2288 V2
CPU
Memory
Internal Storage
enterprise.huawei.com
Published 1/11/2013
Page 10 of 21
Huawei SR120
Operating System
Database Software
Huawei ES3000 PCIe SSD card exhibits industry-leading reliability with features including
embedded ECC engine and RAID5 engine combined to provide two-dimensional error
detection/correction mechanism. However, for customers who are accustomed to traditional RAID
protection, a second PCIe SSD card can be added in this FTDW RA to mirror the original one.
Published 1/11/2013
Page 11 of 21
Descriptions
Server
Tecal 2288 V2
CPU
Memory
Internal Storage
HBA
Huawei SR120
Operating System
Database Software
enterprise.huawei.com
Published 1/11/2013
Page 12 of 21
To fully utilize the potential of two PCIe SSD cards, the following FTDW configuration effectively
boost the Fast Track rated capacity from 4TB to 10TB while doubling the query throughput at the
same time.
Descriptions
Server
Tecal 2288 V2
enterprise.huawei.com
Published 1/11/2013
Page 13 of 21
CPU
Memory
Internal Storage
HBA
Huawei SR120
Operating System
Database Software
RAID Configuration
All the PCIe SSD based FTDW RAs share the same RAID configurations for the hard disks.
4x 300GB disks for database logs: One RAID 10 group is created and configured with a single
virtual disk.
2x 600GB disks for Windows Server and SQL server: One RAID 1 group is created and
configured with a single virtual disk for OS and software installation.
One additional 600GB disk is reserved as hot spare to further improve system availability.
In Huawei 4TB FTDW RA with redundant 2.4TB PCIe SSDs, the two SSDs are mirrored by
Windows Server Disk Management in Server Manager.
enterprise.huawei.com
Published 1/11/2013
Page 14 of 21
Enable Lock Pages in Memory: This Windows policy determines which accounts can use a
process to keep data in physical memory preventing the system from paging the data to virtual
memory on disk. For more information, see How to: Enable the Lock Pages in Memory Option
(http://go.microsoft.com/fwlink/?LinkId=141863).
Startup Options
o
-E must be added to the start-up options. This increases the number of contiguous
extents in each file that are allocated to a database table as it grows. This improves
sequential disk access. For more information about this option, see Microsoft
Knowledge Base Article 329526 (http://support.microsoft.com/kb/329526). It is
important to ensure that the -E option has taken effect at database startup. The option is
case-sensitive and format-sensitive. White spaces before or after the option can prevent
initialization.
-T1117 should also be added to the start-up options. This trace flag ensures even
growth of all files in a file group in the case that autogrow is enabled. The standard
FTDW recommendation for database file growth is to pre-allocate rather than autogrow
(with the exception of TempDB).
-T834 should be evaluated on a case-by-case basis. This trace flag can improve
throughput rates for many data warehousing workloads. This flag enables large page
allocations in memory for the SQL Server buffer pool. For more information about this
and other trace flags, see Microsoft Knowledge Base Article 920093
(http://support.microsoft.com/kb/920093).
Note: At this time SQL Server 2012 does not support the use of -T834 if ColumnStore
indexes are in use on the database. If you plan to use ColumnStore indexes, do not use
this trace flag. It was not used in the Fast Track validation testing.
SQL Server Maximum Memory: For SQL Server 2012 no more than 92 percent of total
server RAM should be allocated to SQL Server. If additional applications are to share the
server, the amount of RAM left available to the operating system should be adjusted
enterprise.huawei.com
Published 1/11/2013
Page 15 of 21
accordingly. This setting is controlled by the max server memory option. In the Fast Track
validation testing, the maximum server memory was set at 92% of total server RAM.
Resource Governor: In default settings for SQL Server 2012, Resource Governor provides a
maximum of 25 percent of SQL Server memory resources to each session. This setting can be
used to reduce the maximum memory consumed by each query, and there is a trade-off
between lowering constraints that offer higher performance for individual queries and more
stringent constraints that guarantee the number of queries that can run concurrently. In the
Fast Track validation testing, Resource Governor memory grant was set at 19%.
Maximum Degree of Parallelism (MaxDOP): This setting controls the number of logical
cores that SQL Server uses for the parallel execution of a query. The default value of
MaxDOP is 0, which dictates SQL server will dynamically allocate work up to the total number
of CPU cores seen by the SQL Server service. In the Fast Track validation testing, MaxDOP
was set at 8 for one-socket systems and set at 16 for two-socket systems. Also
Hyper-Threading was turned on in the tests.
As with all Fast Track performance benchmarks, actual results for customers data and workload
may vary. Testing at various Resource Governor and MaxDOP settings can reveal best results for
specific situations.
enterprise.huawei.com
Published 1/11/2013
Page 16 of 21
Fast Track Data Warehouse Validation with Column Store Index (CSI)
Microsoft SQL Server 2012 features a new non-clustered indexing feature based on columnar
storage technology, called xVelocity in-memory Column Store Index (CSI). It could significantly
improve query performance for a wide range of queries. The FTDW validation testing is based on
non-CSI enabled workloads, since FTDW systems are designed to run effectively even without
columnar optimization. After a Fast Track configuration is validated, CSI will be added to the test
database and the throughput performance will be recorded.
During the Fast Track validation testing, we used simulated data warehouse workloads to gauge
performance of each Fast Track configuration. These tests use workloads with varying sizes
ranging from 5 to 40 concurrent queries, running from 30 minutes to 1 hour each.
The following table lists the performance metrics of the Fast Track reference architectures
presented in this paper.
enterprise.huawei.com
Published 1/11/2013
Page 17 of 21
Benchmark
FTDW RA
Logical
Physical
Model
Scan Rate
Scan Rate
MB/s
MB/s
1181
Fast Track
Fast Track
Rated IO
Rated CSI
MB/s
IO MB/s
772
976.5
2766.5
1380
1443.5
1081.5
1262.5
2714
2004
1498
1133
1315.5
2749
2003
2852.5
2053
2452.75
4411
3865
Peak Scan
Rate MB/s
5TB
RH2288V2 w/
SAS HDD
4TB
RH2288V2 w/ 1x
PCIe SSD
4TB
RH2288V2 w/ 2x
Mirrored PCIe
SSD
10TB
RH2288V2 w/ 2x
PCIe SSD
Benchmark Logical Scan Rate measures how fast data can be read from buffer cache.
Benchmark Physical Scan Rate measures how fast data can be read from physical disks.
Fast Track Rated IO is the average of the benchmark logical and physical scan rates.
Fast Track Rated CSI IO is the average of the logical and physical scan rates when Column
Store Indexing is enabled.
Peak Scan Rate is the maximum scan rate observed during testing.
enterprise.huawei.com
Published 1/11/2013
Page 18 of 21
TempDB Space
Maximum User
Data Capacity
(with Compression)
(with Compression)
5TB
RH2288V2 w/
2565 GB
769 GB
5 TB
6.1 TB
2280 GB
684 GB
4 TB
5 TB
2280 GB
684 GB
4 TB
5 TB
4560 GB
1048 GB
10 TB
12 TB
SAS HDD
4TB
RH2288V2 w/ 1x
PCIe SSD
4TB
RH2288V2 w/ 2x
Mirrored PCIe
SSD
10TB
RH2288V2 w/ 2x
PCIe SSD
Maximum User Data Capacity is the raw data space minus the recommended space for
TempDB database files, multiplied by the Fast Track compression factor, which is 3.5 for SQL
Server 2012. It does not include any disk headroom that is practically required.
Fast Track Rated User Data Capacity is the optimal, Fast Track validated user data capacity.
It takes into account the Fast Track Rated IO throughput, as well as the available storage
capacity. The rated performance of the Fast Track system is valid when user data does not
exceed this capacity.
enterprise.huawei.com
Published 1/11/2013
Page 19 of 21
5. Conclusion
Huawei Microsoft Fast Track Data Warehouse reference architecture aims to document a set of
hardware and software optimized configurations that are tested and tuned to deliver high
performance and cost effective solutions.
This document describes two classes of configurations. The cost optimized configuration based on
enterprise class SAS HDDs delivers 5TB of rated data capacity at superior performance as
compared to recently published alternative solutions. The performance optimized configurations
based on PCIe SSD cards deliver up to double the rated capacity and more than double the rated
IO throughput as compared to the SAS HDD based configuration.
Huaweis decades of experiences in designing and manufacturing mission critical Information and
Communication Technology solutions allow us to offer a broad range of data warehouse solutions
based on Microsoft SQL Server 2012 that are cost effective, highly available and higher in
performance than alternative SMP solutions on the market.
enterprise.huawei.com
Published 1/11/2013
Page 20 of 21
enterprise.huawei.com
Published 1/11/2013
Page 21 of 21