Increasing Database

Performance Using the Dell

PERC H800 with Solid State
Drives
A Dell Technical White Paper

Joe Noyola

Systems Performance Analysis

Increasing Database Performance Using the PERC H800 with Solid State Drives

THIS WHITE PAPER 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.

© 2011 Dell Inc. All rights reserved. Reproduction of this material in any manner whatsoever
without the express written permission of Dell Inc. is strictly forbidden. For more information,
contact Dell. Dell, the DELL logo and the DELL badge are trademarks of Dell Inc. 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 Inc. disclaims any proprietary interest in
trademarks and trade names other than its own. TPC and the benchmark name TPC-E are
registered trademarks of the Transaction Processing Performance Council. TPC-E results were
simulated and do not represent actual TPC benchmark results.

July 2011

Page ii
Increasing Database Performance Using the PERC H800 with Solid State Drives

Table of Contents
Executive Summary ........................................................................................................... 4
Introduction .................................................................................................................... 4
Key Results ..................................................................................................................... 4
Test Methodology .............................................................................................................. 5
Test Results..................................................................................................................... 5
Database Performance - Transactions Per Second (tps) .......................................................... 5
Database Performance - User Scalability with acceptable average query latency........................... 6
Storage Performance - I/Os per second (IOPS) ..................................................................... 7
Price/Performance ...................................................................................................... 7
Conclusion ...................................................................................................................... 8
Appendix A – System Cost Details (from Dell.com) ...................................................................... 9
Appendix B - Server Hardware Configuration .......................................................................... 10
Appendix C – Storage Hardware Configurations ........................................................................ 11
Appendix D - Database Software / Configuration / Benchmark Factory Configuration Details ............... 13

Tables
Table 1. Storage Configuration .......................................................................................... 5

Figures
Figure 1. Database Performance Comparison .......................................................................... 6
Figure 2. Database Concurrent User Comparison...................................................................... 6
Figure 3. IOPS Scalability Comparison ................................................................................... 7
Figure 4. Performance, Price/Performance Comparison............................................................. 8

Page iii
Increasing Database Performance Using the PERC H800 with Solid State Drives

Executive Summary
Dell’s High-Performance SAS Solid State Drives (SSD) significantly improve I/O performance
compared to traditional hard drives (HDDs). With the introduction of CacheCade™ functionality,
Dell’s PERC (PowerEdge™ RAID Controller) H800 with the PowerVault™ MD1220 enclosure now
supports two SSD implementation schemes: SSD Caching and SSDs implemented as primary data
storage. This white paper evaluates the performance improvements and cost of these SSD usage
cases against traditional hard drives using a simulation of the I/O workload of a small-to-medium
sized SQL Server TPC-E-like OLTP database.

Introduction
Many IT professionals today are considering SSDs as an option to improve storage performance to
mitigate traditional storage I/O bottlenecks. Solid state storage is flash-based and thus does not
rely on slow-moving mechanical parts to access data. This makes them ideal for workloads with
read-intensive data I/O access patterns. SSDs, while providing maximum performance, do require
tradeoffs: given that this is a fairly new technology, they are currently more expensive than
traditional HDDs.

Compared to all-SSD implementations that require multiple SSDs per volume, CacheCade with one
SSD provides a lower-cost solution which still harnesses the performance of SSDs. CacheCade
software (SSD-caching) running on the PERC H800 1GB NV Cache RAID controller improves the
performance of hard drive storage by identifying frequently-accessed read operations and caching
them on solid-state storage. Application read performance is improved over time, as the
CacheCade cache volume is populated and subsequent reads are serviced from SSD cache.

Unlike CacheCade, where I/O performance is achieved as the SSD cache is populated, all-SSD
implementations see performance improvements immediately. Applications can utilize the full
potential of SSDs, as both read and write operations are accelerated. If maximum storage
performance supersedes storage capacity and cost concerns, the PERC H800 combined with
multiple SSDs provides a new level of storage I/O performance traditional storage cannot match.

To measure the performance improvements these SSD implementations offer, Quest’s Benchmark
Factory® was used to build and simulate an industry-standard SQL Server OLTP database. Compared
to hard drive storage alone, CacheCade and all-SSD implementations significantly improve OLTP
database performance.

Key Results
OLTP Database Performance - Transactions per second (tps)
• Enabling an SSD as a CacheCade volume increased hard drive database performance by 5x 1
• All-SSD storage achieved 15x more database performance than HDD storage alone1
Database User Scalability Support with Acceptable Response time of 100ms
• CacheCade volume allows hard drive storage an additional 3x more users 2
• All-SSD storage supports up to 350 users, an 11x increase in database user support over hard drive
storage2
Storage Performance – I/Os per Second (IOPS)
• CacheCade provided an increase of 4x more IOPS than hard drive storage alone 3

1
For detailed performance results see Figure 1. Testing configurations are detailed in the appendices.
2
For detailed performance results see Figure 2. Testing configurations are detailed in the appendices.
Page 4
Increasing Database Performance Using the PERC H800 with Solid State Drives

• All-SSD storage achieved 14x more IOPS than hard drive storage3

Price per IOPS (lower is better)

• Dollars per IOPS of hard drive storage costs $1.54/IOPS 4
• An SSD enabled as a CacheCade volume decreases dollars per IOPS to $0.64/IOPS, a 58 percent
savings in price/performance4
• All-SSD storage achieved 14x more IOPS than hard drive storage, which drops overall dollar per IOPS
to only $0.31/IOPS, an 79 percent dollars/IOPS savings compared to hard drive storage4

Test Methodology
Quest’s Benchmark Factory for Databases (BMF) is a database build and workload simulation tool
that measures database performance. To compare the performance that CacheCade and all-SSD
storage achieve over that of traditional hard drives, a SQL Server 2008 100GB database modeled
after the industry-standard TPC-E benchmark was built on a RAID 10 volume made up of fourteen
146GB 10k RPM hard drives. An additional database was built on a 4-drive SAS SSD RAID 10 volume.
Table 1 lists the hardware of each storage configuration.

Table 1. Storage Configuration

Database Storage RAID Configuration Drive Type / Speed
Configuration
Hard Drive baseline 14-drive RAID 10 SAS HDD /10k RPM – 6Gbs
Hard Drive + 14-drive RAID 10 + 1 SAS SSD SAS HDD /10k RPM – 6Gbs +
1 CacheCade volume 1 SSD

All-SSD baseline 4-drive RAID 10 SAS SSD

Baseline performance was collected from the database built on hard drive storage alone. A single
SAS SSD was then enabled as a CacheCade volume and measurements were repeated. Metrics were
then collected from the database built on all-SSD storage. During each measurement, BMF
simulated users issuing TPC-E like SQL transactions that exercised each database’s storage sub-
system with a random read and write I/O data access pattern.
BMF simulated an increasing number of virtual users until the average query response time met an
acceptable response time of 100 milliseconds. To ensure all I/O requests were serviced from
physical storage, the SQL Server 2008 default instance was limited to run with only 1GB of RAM.
Limiting the amount of memory available to SQL Server ensures that storage sub-systems are
exercised to their full I/O capacity.

Test Results
Database Performance - Transactions Per Second (tps)
A baseline of 40 tps was measured from the hard drive database. Enabling a single SAS SSD as a
CacheCade volume increased database performance to over 200 tps, a 5x gain. All-SSD database
performance reached a staggering 600 tps; a 15x gain in tps compared to the database on all-HDD
storage. Figure 1 shows the achieved database performance CacheCade and all-SSDs provide.

3
For detailed performance results see Figure 3. Testing configurations are detailed in the appendices.
4
For detailed performance results see Figure 4. Testing configurations are detailed in the appendices.
Page 5
Increasing Database Performance Using the PERC H800 with Solid State Drives

700
600 tps
600
Transactions Per Second (tps)

500

400
15x
300
200 tps more
200 tps
5x
100 more
40 tps tps

0
14-HDDs 14-HDDs + 1-drive 4-SSDs
CacheCade Volume

Figure 1. Database Performance Comparison

Database Performance - User Scalability with acceptable average query latency

BMF increased the number of virtual users simulating an OLTP workload to each storage
configuration while measuring average query latency. Acceptable average query latency was set to
100ms, thus ensuring the simulated users an acceptable quality of service. Hard drive storage
supported up to 30 simultaneous users accessing the database while maintaining the acceptable
average query response time. Enabling CacheCade allows hard drive storage alone to sustain an
additional 3x user load. Database built on solely SSDs supports up to 350 concurrent users, 11x
more users than hard drives.

1
0.9
Average I/O Latency (seconds)

0.8
0.7
0.6
0.5 30 100 350
users users users
0.4
0.3
0.2
Acceptable Average Latency = .1
0.1
0
25 50 75 100 125 150 175 200 225 250 275 300 325 350
14-HDDs 14-HDD+ 1 CacheCade Volume 4-SSDs

Figure 2. Database Concurrent User Comparison

Page 6
Increasing Database Performance Using the PERC H800 with Solid State Drives

Storage Performance - I/Os per second (IOPS)

The hard drive storage configuration achieved a maximum of 5,500 IOPS while 30 users
concurrently queried the database. Configuring an SSD as a CacheCade volume allowed 100 users to
query the database and increased total IOPS of hard drives to 20,450, a 3x improvement. All-SSD
storage supported 350 concurrent database users and achieved 76,270 IOPS, 11x more IOPS
compared to all-HDD storage. Figure 3 illustrates the IOPS achieved by hard drives, CacheCade, and
all-SSD storage.

90,000

80,000 76,270 IOPS

70,000

60,000

50,000
IOPS

40,000

30,000
20,450 IOPS
20,000
5,500 IOPS
10,000

0
14-HDDs 14-HDDs + 1-drive 4-SSDs
CacheCade Volume

Figure 3. IOPS Scalability Comparison

Price/Performance
Traditional storage offers a combination of low cost and storage capacities that solid state storage
currently cannot match. Due to the electro-mechanical rotating circular design, however, the
number of IOPS hard drives can achieve is physically limited. As seen in Figure 4, price per IOPS of
a database with hard drive storage is $1.54, a 58 percent higher cost per IOPS than CacheCade at
$0.64 per IOPS and 79 percent more expensive than $0.31 per IOPS that all-SSD storage provides.

Page 7
Increasing Database Performance Using the PERC H800 with Solid State Drives

90,000 $1.80

80,000 $1.60

Price ( $) / Performance ( IOPS)

$1.54 / IOPS
70,000 $1.40

60,000 $1.20

50,000 $1.00
IOPS

40,000 $0.80
$0.64 / IOPS
30,000 $0.60
$0.31/
20,000 IOPS $0.40

10,000 $0.20

0 $0.00
14 HDDs 14 HDDs + 1-drive CacheCade 4 SSDs
Volume

Figure 4. Performance, Price/Performance Comparison

Conclusion
With the introduction of flash-based Solid State Disks, customers are faced with balancing performance
versus cost. CacheCade allows the best of both worlds, where SSDs can be used as a read cache, but
traditional HDDs can be used for primary data storage for a lower total cost of ownership. Although
carrying the highest purchase price, SSDs implemented at primary data storage differ from CacheCade as
both read and write I/O performance is accelerated and available instantly to the application versus SSD-
caching.

The results of this Benchmark Factory OLTP database simulation clearly shows that both CacheCade
and SSDs used for primary database storage improve database performance while decreasing the
overall cost per IOPS. CacheCade provides customers a low-cost SSD solution that improves
database performance while continuing to benefit from the best features traditional hard drives
offer: storage capacity and cost. If performance is paramount, database applications built on all-
SSD storage fully harness the I/O capability of Dell’s enterprise SAS SSD and achieve a new era of
OLTP database performance above and beyond that of traditional rotating media.

Page 8
Increasing Database Performance Using the PERC H800 with Solid State Drives

Appendix A – System Cost Details (from Dell.com)

Storage Configuration Components Price Dollars/IOPS

All-HDD storage 1- PERC H800 1GB NVRAM controller $8,478 $1.54/IOPS
1- PowerVault MD1220 storage enclosure
14 - 10k SAS 146GB drives - data
4 - 10k SAS 146GB drives – log
All-HDD + 1-drive All-HDD configuration + $13,073 $0.64/IOPS
CacheCade 1 - Pliant 149GB SLC SSD as CacheCade
All-SSD 1- PERC H800 1GB NVRAM controller $23,288 $0.31/IOPS
1- PowerVault MD1220 storage enclosure
4 - 149GB SLC SSDs – data
4 - 10k SAS 146GB drives - log

Page 9
Increasing Database Performance Using the PERC H800 with Solid State Drives

Appendix B - Server Hardware Configuration

Server Hardware Dell PowerEdge R710

Processor Dual Intel Xeon X5670 2.93Ghz 12MB Cache 6 core
processors
Total RAM in system (GB) 16
Vendor and model number Samsung ST9146852SS
Memory Type PC3L-10600R
Speed (MHz) 1333
Speed in system as tested 1066
Number of RAM modules 16 x 4 GB
Rank organization Dual Rank
Vendor and model number Dell MBE2147RC
Number of disks in system 2
Size (GB) 146
RPM 15,000
Type SAS 6 Gbps
RAID Type RAID 1
Controller PERC Integrated H700 512MB
F/W:12.1.0091 / Driver - 4.31.1.64
Name Microsoft® Windows Server® 2008 R2 Enterprise
File system NTFS
Language English
Vendor and model number Broadcom® BCM5709S NetXtreme® II
Type Integrated

Page 10
Increasing Database Performance Using the PERC H800 with Solid State Drives

Appendix C – Storage Hardware Configurations

Storage Configuration: All-HDD Baseline

PowerVault Storage Enclosure MD1220
Total Number of Available 24
Drive Slots
Number of Drives Installed 14-drive RAID 10 volume for database
4-drive RAID 10 volume for database log
Drive Type 6Gb/s SAS
HDD Speed (RPM) 10k HDD
HDD Capacity (GB) 146GB
Controller Dell PERC H800 1GB NV Cache
Controller Firmware / Driver F/W 12.1.0091 / 4.31.1.64
PERC Cache Settings Write Back
Adaptive Read Ahead
Disk Firmware HT64

Storage Configuration: All-HDD + 1 CacheCade Volume

PowerVault Storage Enclosure MD1220
Total Number of Available 24
Drive Slots
Number of Drives Installed 14-drive RAID 10 volume for database
4-drive RAID 10 volume for database log
1 Pliant SSD as CacheCade volume
Drive Type 6Gb/s (HDD) / 3Gb/s (SSD)
HDD Speed (RPM) 10k HDD
SSD / HDD Capacity (GB) 149GB (SSD) / 146GB (HDD)
Controller Dell PERC H800 1GB NV Cache
Controller Firmware / Driver F/W 12.1.0091 / 4.31.1.64
PERC Cache Settings Write Back
Adaptive Read Ahead
Disk Firmware D801

Page 11
Increasing Database Performance Using the PERC H800 with Solid State Drives

Storage Configuration: All-SDD

PowerVault Storage Enclosure MD1220
Total Number of Available 24
Drive Slots
Number of Drives Installed 4-drive Pliant SSD RAID 10 volume for database
4-drive RAID 10 volume for database log
Drive Type 3Gb/s (SSD) / 6Gb/s (HDD)
HDD Speed (RPM) 10k
SSD / HDD Capacity (GB) 149GB (Database) / 146GB (Log)
Controller Dell PERC H800 1GB NV Cache
Controller Firmware / Driver F/W 12.1.0091 / 4.31.1.64
PERC Cache Settings Write Back
Adaptive Read Ahead
Disk Firmware D011

Page 12
Increasing Database Performance Using the PERC H800 with Solid State Drives

Appendix D - Database Software / Configuration / Benchmark Factory

Configuration Details

Database Software Microsoft® SQL Server® 2008 R2 Enterprise

Memory Configuration 1GB
Database Container Size 150GB
Log Size 100GB
Database Instance Memory Limited to 1GB, ensured all I/O be serviced from disk
Primary DB File Location HDD Database: 14-drive 10K RAID 10 volume
SSD Database: 4-drive SAS SSD RAID 10 volume
Benchmark Software Quest’s Benchmark Factory for Databases
Version 6.1.1
Industry Standard Benchmark TPC-E
Scale Factor 50
Complete Database Size 126GB
Number of Virtual Users Unlimited
Data Collection Enabled Yes, with additional logical disk metrics collected by BMF from
Windows Performance Monitor

Page 13