HP LeftHand SAN Solutions Support Document

Application Notes NSM 2120-G2 Performance

Legal Notices Warranty The only warranties for HP products and services are set forth in the express warranty statements accompanying such products and services. Nothing herein should be construed as constituting an additional warranty. HP shall not be liable for technical or editorial errors or omissions contained herein. The information contained herein is subject to change without notice. Restricted Rights Legend Confidential computer software. Valid license from HP required for possession, use or copying. Consistent with FAR 12.211 and 12.212, Commercial Computer Software, Computer Software Documentation, and Technical Data for Commercial Items are licensed to the U.S. Government under vendor's standard commercial license. Copyright Notices © Copyright 2009Hewlett-Packard Development Company, L.P.

2

Contents

NSM 2120-G2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Maximum Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Performance Tuning Tips. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Performance Test Environment for Realistic Application Benchmarks . . . . . . . . . . . . . . . . . . 7 Real-World Application Benchmarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 NSM2120 Performance Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Additional References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3

4

Th is pa ge is in te nt io

na lly

le

ft

bl an

k

NSM 2120-G2 Performance

Overview The LeftHand Networks NSM 2120-G2 is an ideal platform to build robust storage clusters running the LeftHand Networks SAN/iQ® software. LeftHand’s SAN/iQ software platform delivers enterprise-class SAN functionality on industry standard hardware. This document provides an overview of the performance characteristics your applications can rely on with the LeftHand Networks NSM 2120-G2. The LeftHand Networks SAN architecture enables the aggregation of any number of storage nodes to create storage clusters. Increasing the number of storage nodes in a cluster increases the total number of disks, processors, controllers, network connections and cache used in the storage cluster, which allows for performance to scale along with capacity. In addition, SAN/iQ Network RAID dynamically load balances data across the storage clusters and eliminates bottlenecks as the cluster grows. This dramatically increases data availability since data is now striped and mirrored across multiple storage nodes. The disks, processors, controllers, network connections and caches in the NSM 2120-G2 were selected to maximize throughput, IOPS, and scalability.

NSM 2120-G2 The LeftHand Networks NSM 2120-G2 is an enterprise class storage node built on industry-standard Intel chipsets. The storage node can utilize either Serial Attached SCSI (SAS) or Serial ATA II (SATA) disk drives. The solution is available in multiple disk configurations yielding between 1.8 to 12.0 Terabytes (TB) of raw storage capacity. The LeftHand Networks NSM

5

2120-G2 provides enterprise-class reliability, high availability and performance. Specific product highlights include: Twelve hot swappable disk drives, dual NIC connectivity with automatic failover, dual power supplies, robust power and cooling, and battery-backed write cache.

Maximum Performance Metrics The maximum performance metrics data is used to show maximum performance limits of the SAN in an optimal scenario. While these numbers are often used as a basis of comparison between different storage vendors, they aren't representative of the workload typically found in production environments. The type of workload run against the SAN can dramatically affect performance numbers. The maximum performance metrics should be used for comparison against other SAN vendors rather than as a guideline for your specific needs. When sizing for your specific environment, LeftHand Networks advises using a benchmark that closely resembles your own data requirements.

Performance Tuning Tips SAN performance can be negatively impacted if the SAN is not configured properly. The following guidelines will help maximize the performance of the LeftHand Networks SAN. Network Best Practices

6

•

Use Gigabit Ethernet switching infrastructure.

•

Dedicate a separate IP subnet/VLAN for your iSCSI SAN.

•

Use a secondary Gigabit Ethernet NIC in servers attached to the Ethernet storage subnet for high availability and performance.

•

If possible attach all modules and application servers to the same Gigabit switches for the storage networking.

•

Make sure all switch ports and Ethernet adapters are configured to run at Full-duplex Gigabit speeds.

Disk Formatting Best Practices •

Use the default block size, or allocation unit size, unless a particular application specifies an alternate size. Microsoft Exchange should be on volumes formatted with a block size of 4096 bytes.

•

Partitions should be created starting at sector 128 to align the offset to 64KB using Diskpar or Diskpart.

Storage Node Configuration Best Practices •

Bond networks interfaces for network redundancy and more bandwidth using Adaptive Load Balancing (ALB).

•

Use RAID level 5 or 10 for disk redundancy within storage modules.

•

Enable SAN/iQ iSCSI Load Balancing on authentication groups for Windows servers

•

Install the DSM software driver to enable fault-tolerant paths to the SAN modules while also increasing available bandwidth to the SAN for maximum performance

Performance Test Environment for Realistic Application Benchmarks Performance data was gathered using enterprise class servers with the following attributes: •

2.8GHz dual core Intel Xeon processor

•

4 gigabytes of RAM

•

Dual gigabit Ethernet network adapters (bonded using balance ALB)

•

All equipment connected to the same gigabit Ethernet switch

The iSCSI SAN configurations included a dedicated group of three storage nodes comprised of: •

Single management group with a single cluster including all storage modules

•

Storage modules were configured in either RAID 5 with volumes further protected by SAN/iQ Network RAID Level-2 (2-way Replication).

•

iSCSI sessions were load balanced across the clusters using SAN/iQ iSCSI Load Balancing

7

The LeftHand Networks SAN is designed to work on your existing Ethernet infrastructure. Segmenting the storage network from your corporate network is recommended, and boosts overall performance of the solution. For performance testing, the SAN network was isolated to a single broadcast domain (subnet) and had dedicated NICs for the storage traffic. A single 24 port Gigabit switch was used for all tests. All disk volumes were configured using the Microsoft iSCSI Software Initiator Version 2.0.4. Volumes were mounted using the Windows Logical Disk Manager and formatted with NTFS using the default allocation unit size.

Real-World Application Benchmarks All of the following benchmarks were designed to help characterize the actual performance users can expect in similar production environments. Each benchmark measures different storage characteristics. Standard SQL server and Exchange benchmarks illustrate the expected performance in those application environments. Simple file copies show throughput in megabytes per second (MB/sec) and IOmeter simulates IOPS (inputs/outputs per second) and throughput. Microsoft SQLIOStress The SQLIOStress utility was written and is maintained by the Microsoft SQL Server Escalation staff. Designed to help detect possible I/O path problems that would lead to Microsoft SQL Server data corruption or loss, it emulates different SQL server database versions and types and reports on some basic benchmarks. At a minimum it runs 5 iterations. Only the third and fourth iterations should be used for comparing performance according to the SQLIOStress tool documentation. For this document the 4th iteration will be used. The SQLIOStress test runs from only a single host and typically becomes CPU bound. The tool is not capable of scaling at the same rate as the SAN. Some scaling is achieved on a single host due to the lower latency in the SAN. Best Practice - Keep the data files and logs for SQL databases on separate volumes.

8

Microsoft Exchange Jetstress The Jetstress tool allows administrators to verify the performance and stability of the disk subsystem before putting their Exchange server into a production environment. It reports back to the administrator with an html document that contains results and a PASS or FAIL for the overall test. A PASS or FAIL is strictly based on specific test results not exceeding Microsoft specified limits. Disk Latency is the primary metric used by Jetstress for PASS / FAIL criteria. The values to pass are the log avg. disk sec/write, database page fault stalls, and database avg. disk sec/read. Log average latency should be below 10ms, database page fault stalls should always be 0, and database average read latency should be less than 20ms. LeftHand Networks uses an automated command line version of Jetstress that will increase threads until the test no longer passes and records the IOPS of the last passing run. The highest IOPS achieved while still passing is the value reported in this paper. Best Practice – Keep the email database files and log files on separate volumes. File Copy Using Robust File Copy The file copy test is a straight-forward Windows file copy. Using robocopy.exe a file copy of one hundred 100MB files is done from direct attached storage (DAS) in the server to a volume on the SAN. For this test a simple measurement of megabytes per second is taken. Robocopy (Robust File Copy) is included in the Windows resource kit. While individual file copy speeds are limited by the nature of low queue depth disk operations, the architecture of a LeftHand Networks SAN solution is capable of handling multiple file copy jobs simultaneously without degrading performance. This is represented by the “Concurrent Jobs” set of data. Best Practice - When copying from DAS to the iSCSI SAN remember the file copy cannot go any faster than the DAS disk. Make sure the disk that the source files are on is the fastest disk available to you otherwise it might be limiting the file copy speed. Running multiple file copies at a time will increase the aggregate performance. IOmeter IOmeter is an I/O subsystem measurement and characterization tool. IOmeter is a good tool for simulating loads of different I/O sizes to demonstrate a system’s IOPS and throughput capabilities. These configurations are designed to show typical speeds as opposed to the maximum values.

9

Best Practices - Run IOmeter against raw disks to avoid file system cache interference.

NSM2120 Performance Results

Figure 1: Cache hit random IOPS

Figure 2: Cache hit sequential throughput

10

Figure 3: Single Host SQLIOStress IPS

Figure 4: Single Host Jetstress IOPS

11

Figure 5: Single Host File Copy Reads

Figure 6: Single Host File Copy Writes

12

Figure 7: 512 byte Random IOPS

Figure 8: 64K Sequential Workloads - MB/s

Conclusion The NSM 2120-G2 from LeftHand Networks delivers reliable, scalable and predictable performance for enterprise production environments. Utilizing SAN/iQ clustering the resources of multiple storage nodes are aggregated

13

together into a robust high performance storage cluster. The flexibility and power of SAN/iQ clustering paired with the NSM product line allows a customer to meet any performance and/or capacity need.

Additional References Exchange 2007 - Planning Storage Configurations– http:// technet.microsoft.com/en-us/library/bb124518(EXCHG.80).aspx Using Jetstress to Verify Storage System Performance - http:// technet.microsoft.com/en-us/library/aa996390.aspx Windows Server 2003 Resource Kit Tools (Robocopy) - http:// www.microsoft.com/downloads/ details.aspx?familyid=9d467a69-57ff-4ae7-96ee-b18c4790cffd&displayl ang=en App Note - Building High Performance High Availability IP Storage Networks with SAN/iQ - http://www.hp.com/ App Note - Best Practices for Enabling Microsoft Windows with SAN/iQ http://www.hp.com Iometer Downloads - http://www.iometer.org/doc/downloads.html

The Information in this Article applies to SAN/iQ versions 8.0+ Microsoft Windows Server 2003/2008 *Some of the performance claims represented in this document are extrapolated from benchmark data based on internally accepted performance scaling models used by LeftHand Networks

14