Reference Architecture Library

Reference Architecture Library Reference Architecture Overview for Microsoft® Exchange Server 2007 with Hitachi Adaptable Modular Storage Model AMS1000 White Paper

September 2007

Executive Summary The release of Microsoft® Exchange Server 2007 incorporated significant improvements, including changes to the Exchange workload profile and new options for data protection that influence the underlying storage infrastructure. This technical brief provides an overview of the ways some new features of Exchange Server 2007 impact the design and implementation of a robust storage infrastructure. The technical brief is intended to serve as a companion to the initial Hitachi Data Systems submission to the Exchange Solution Reviewed Program (ESRP) for Exchange Server 2007, and it describes solutions based on the Hitachi Adaptable Modular Storage model AMS1000 configuration tested for a range of 3,125 to 25,000 Exchange mailboxes. The paper discusses how data protection alternatives can affect performance and introduces a recommended building-block approach for designing and architecting Exchange Server 2007 storage infrastructures on the AMS1000. Specifically, the paper describes the reference test environment that was used during the development of the ESRP submission. It also provides an overview of some of the architectures available to provide varying degrees of Exchange 2007 data protection. In addition, the paper provides insight into the various hardware, software, and service components from Hitachi Data Systems required to implement a protected Exchange Server 2007 solution on the AMS1000 and address economy, performance, and growth. Although the recommendations documented here may generally represent good practices, configurations may vary. Please contact your Hitachi Data Systems representative, or visit Hitachi Data Systems online at http://www.hds.com for further information on solutions by Hitachi Data Systems.

Contents Perform an ce a nd Dat a Prot ec tion for Mi crosoft E xch a nge Serv er 2007 ................................................... 1 Hitachi Adaptable Modular Storage Model AMS1000 for Exchange............................................................................... 1 AMS1000 and Exchange Server 2007 ESRP Performance Solution Profile.................................................................... 2 Reference ESRP Test Topology ....................................................................................................................................... 3 Reference ESRP Test Environment Specifications........................................................................................................... 3 Reference ESRP Test Storage Configurations ................................................................................................................. 4 Reference ESRP Test Summary Results.......................................................................................................................... 8 Exch ang e Data Prote ction .................................................................................................................... 9 Performance Planning for Backup and Replication.......................................................................................................... 9 Designing for Performance and Protection .................................................................................................................... 11 Key Takeaways ............................................................................................................................................................... 12 Storag e- bas ed Prote ction for Exc hang e .............................................................................................. 13 Adaptable Modular Storage model AMS1000: Hitachi Data Systems Protected Architectures for Exchange Server 2007................................................................................................................................................................................ 13 Examples of Building-block-driven Hitachi Data Systems Protected Solution Architectures ........................................ 16 Sum mar y ........................................................................................................................................... 22 Ap p endi x: Sup pl eme nt al E xch ang e S erv er 2007 Infor mation ................................................................. 23

Reference Architecture Library Reference Architecture Overview for Microsoft® Exchange Server 2007 Hitachi Adaptable Modular Storage Model AMS1000 White Paper

Performance and Data Protection for Microsoft Exchange Server 2007 Microsoft Exchange Server 2007 changed the way that Exchange servers and their underlying storage interact. Since the release of Exchange 2003, servers have increased in both power and the ability to affordably support large amounts of memory. Microsoft took this into consideration when designing Exchange Server 2007, using 64bit processors with large internal cache memory as the new required platform. These changes led to a reduction in the I/O workload on the disk storage system since more operations are now cached in server memory, resulting in fewer I/Os to physical storage. These enhancements, along with other fundamental changes to the data architectures of Exchange, improve performance and availability while minimizing unnecessary expense.

Hitachi Adaptable Modular Storage Model AMS1000 for Exchange With those objectives in mind, the initial Hitachi Data Systems submission to the Microsoft Exchange Solution Reviewed Program (ESRP) used the AMS1000 as the foundation for the solution. The ESRP submission can be found at http://technet.microsoft.com/en-us/exchange/bb412164.aspx. The AMS1000’s high-availability characteristics and its ability to scale easily to support a large number of active Exchange users make it an ideal platform for a wide range of Exchange solutions. Microsoft offers the Exchange Solution Reviewed Program to help customers and partners design the right environment for their requirements. Visit the abovementioned Web page for additional information about this valuable program provided by the Exchange product group.

Fundamental Storage Considerations for Exchange Server 2007 Capacity, performance, and available budget are often at odds with each other when it comes to selecting a storage solution for an Exchange environment. Even with performance improvements, Exchange Server 2007 still places a significant demand on the underlying storage infrastructure, and this frequently translates into the need for more physical disk resources to provide acceptable response times and the highest availability. Generally, the design decisions for Exchange Server 2007 involve the following factors: • Making sure there is enough space to store all of the data • Making sure the solution provides an acceptable user experience as measured by response times • Making certain that all critical Exchange components are properly protected to ensure operations resume

following an outage, disaster, or other event that leads to service disruption The architect’s goal is to balance the above factors when designing an Exchange Server 2007 infrastructure that addresses cost while improving availability.

1

The Exchange Server 2007 reference architecture overview that follows is designed to accommodate a scalable Exchange Server 2007 deployment with a range of Hitachi Data Systems data protection options. It is strongly recommended that the Microsoft documents listed in the Appendix be referenced when implementing or migrating to a Microsoft Exchange Server 2007 environment.

AMS1000 and Exchange Server 2007 ESRP Performance Solution Profile The initial task was to develop and test a scalable, high-performance platform for Microsoft Exchange Server 2007. Exchange Server 2007 presents two distinct workload patterns to storage: a random workload consisting of 8KB read/write operations to the databases, and a sequential write workload of varying record size (between 512 bytes and the log buffer size) to the transaction logs. A key for performance design is an expandable storage architecture that supports varying I/O workloads while allowing for the preferred method of data protection. The AMS1000-based solution targets organizations that are seeking to consolidate their Exchange Server 2007 storage on a high-performance, high-reliability storage system that offers a range of data protection options. The tested ESRP configuration was designed to support up to 25,000 Exchange users in a modular, buildingblock fashion, using the following profile specifications: • Up to 8 Exchange Servers • 3,125 users per Exchange Server • 0.48 IOPS per user • 200MB mailbox size • 5 storage groups per server • 1 mailbox store per storage group • 625 mailboxes per mailbox store

2

Reference ESRP Test Topology The following diagram shows the hardware components used for the AMS1000 ESRP submission. Figure 1. Hardware Components for AMS1000 ESRP Submission Hitachi Adaptable Modular Storage AMS1000

IBM® BladeCenter Two (2) HS20 Blades

IBM BladeCenter Two (2) HS20 Blades Fibre Channel SAN IBM BladeCenter Two (2) HS20 Blades

IBM BladeCenter Two (2) HS20 Blades

Reference ESRP Test Environment Specifications The tested configuration is capable of supporting 25,000 users with a 0.48 IOPS per user profile and user mailbox size of 200MB. The guiding design philosophy of this solution is modularity, so that this configuration can be scaled in both directions, down to a minimum of 3,125 users, or up to 25,000 users. This also provides organizations with the ability to expand an Exchange Server 2007 storage configuration incrementally by deploying disk trays and disks as needed. See AMS1000 ESRP Submissions for additional information and details about this and other Hitachi Data Systems submissions to the ESRP program. The following tables summarize the testing environment: Table 1. ESRP Test Environment Specificatons—Simulated Exchange Configuration Simulated Exchange Configuration Number of Exchange mailboxes simulated

25,000

Number of hosts

8

Number of mailboxes/host

3,125

Number of storage groups/host

5

Number of mailbox stores/storage group

1

Number of mailboxes/mailbox store

625

Number of mailbox store LUNs/storage group

1

Simulated profile: I/Os per second per mailbox (IOPS, include 20 percent headroom)

0.48

Database LUN size

267GB

Log LUN size

30GB

Total database size for performance testing

5,468GB

Percentage of storage capacity used by Exchange database**

51 percent

3

Table 2. ESRP Test Environment Specifications—Primary Storage Hardware Primary Storage Hardware Storage Connectivity (Fibre Channel, server attached storage (SAS), SATA, iSCSI)

Fibre Channel

Storage model and OS/firmware revision

Hitachi AMS1000—0760/B-H

Storage cache

16GB

Number of storage controllers

2

Number of storage ports

8

Maximum bandwidth of storage connectivity to host

32Gbit [8x4Gbit host bus adapter (HBA)]

Switch type/model/firmware revision

Brocade 4 Port Enterprise Fibre Channel Switch for IBM® BladeCenter®—v4.4.1a

HBA model and firmware

QLogic QMC 2462S BIOS 1.47, FW 4.00.23

Number of HBAs per host

2 HBAs per host

Host server type

IBM HS20 Blade 2X3.2GHz Intel Xeon Processor—8GB RAM

Total number of disks tested in solution

200

Maximum number of spindles can be hosted in the storage

480

Reference ESRP Test Storage Configurations Table 3. ESRP Test Storage Configurations—Mailbox Store Disks Primary Storage Disk Configuration (Mailbox Store Disks) Disk type, speed, and firmware revision

Fibre Channel disk 146GB 15K 4C56

Raw capacity per disk (GB)

146GB

Number of physical disks in test

160

Total raw storage capacity (GB)

23,360GB

Disk slice size (GB)

N/A

Number of slices per LUN or number of disks per LUN

4 disks per LUN

RAID level

RAID-1+0 at storage level

Total formatted capacity

10,640GB

Storage capacity utilization

46 percent

Database capacity utilization

52 percent

4

Table 4. ESRP Test Storage Configurations—Transaction Log Disks Primary Storage Disk Configuration (Transaction Log Disks) Disk type, speed, and firmware revision

Fibre Channel disk 146GB 15K 4C56

Raw capacity per disk (GB)

146GB

Number of spindles in test

40

Total raw storage capacity (GB)

5,840GB

Disk slice size (GB)

N/A

Number of slices per LUN or number of disks per LUN

2 disks per LUN

RAID level

RAID-1 at storage level

Total formatted capacity

1200GB

The scalable design of the storage architecture for the AMS1000-based solution was shaped by the concept of building blocks, which produced the storage module configuration. A storage module, which is described in Table 5, is used as a fundamental element on which the protected solution architectures (described later) are based. Table 5. ESRP Test Storage Configurations—Single Storage Module Configuration Single Storage Module Configuration Number of supported users

3,125

Number of hosts

1

IOPS per user

0.48

Mailbox size

200MB

Number of storage groups

5

Number of disk trays

2

Total number of disks

27

Number of database disks

20

Number of log disks

6

Number of spare disks

1

Disk type

146GB 15K RPM Fibre Channel

Database RAID level

1+0

Log RAID level

1

Figure 2 shows a graphical illustration of the physical disk allocation for a single storage module configuration.

5

Figure 2. Single Storage Module Hitachi Adaptable Modular Storage AMS1000 Single Module (1 Host) RKA 2

RKA 1

DB SG1

DB SG2

DB SG3

DB SG4

DB SG5

Log SG1 Log SG2

Log SG3 Log SG4

Log SG5

For the targeted 25,000-user Exchange environment, eight storage modules are stacked together with a dedicated storage module allocated to each of the eight host servers. Figure 3 depicts the AMS1000 configuration, scaled to support 25,000 mailboxes by using single storage modules in a building-block fashion.

6

Figure 3. Eight Storage Modules Hitachi Adaptable Modular Storage AMS1000 Eight Modules (8 Hosts)

RKA 16 Host 8 RKA 15

RKA 14 Host 7 RKA 13

RKA 12 Host 6 RKA 11

RKA 10 Host 5 RKA 9

RKA 8 Host 4 RKA 7

RKA 6 Host 3 RKA 5

RKA 4 Host 2 RKA 3

RKA 2 Host 1 RKA 1

DB SG1

DB SG2

DB SG3

DB SG4

DB SG5

Log SG1 Log SG2

Log SG3 Log SG4

Log SG5

7

Reference ESRP Test Summary Results The ESRP storage performance tests exercise storage with a maximum, sustainable Exchange I/O workload for two hours using the Microsoft JetStress tool. The test is designed to show I/O response times for a storage infrastructure under a constant, intensive Exchange Server 2007 emulated workload. The test results below are the sum of all logical disk I/Os and average of all the logical disks’ I/O latency. The complete set of test results, logs, and reports can be found in the AMS1000 ESRP Submission for Exchange Server 2007.

Individual Server ESRP Performance Metrics The sum of I/Os and the average latency across all servers are provided in Table 6. Table 6. Individual Server ESRP Performance Metrics Blade 1

Blade 2

Blade 3

Blade 4

Blade 5

Blade 6

Blade 7

Blade 8

1,814

1,874

1,829

1,851

1,851

1,836

1,845

1,805

Average Database Disk Reads/sec (IOPS)

963

996

971

984

982

976

978

957

Average Database Disk Writes/sec (IOPS)

851

878

858

867

869

860

867

848

Average Database Disk Read Latency (ms)

17

16

16

16

16

16

16

16

Average Database Disk Write Latency (ms)

4

4

4

4

4

4

4

4

Average Log Disk Writes/sec (IOPS)

557

573

560

568

564

564

563

553

Average Log Disk Write Latency (ms)

0

0

1

0

1

1

1

0

Database I/O Average Database Disk Transfers/sec (IOPS)

Transaction Log I/O

Aggregate ESRP Performance Metrics Across all Servers The sum of I/Os and the average latency across all servers in solution are included in Table 7. Table 7. Aggregate ESRP Performance Metrics Across All Servers Aggregated Performance for Exchange Server 2007 Database I/O Average Database Disk Transfers/sec (IOPS)

14,707

Average Database Disk Reads/sec (IOPS)

7,808

Average Database Disk Writes/sec (IOPS)

6,899

Average Database Disk Read Latency (ms)

16

Average Database Disk Write Latency (ms)

4

Transaction Log I/O Average Log Disk Writes/sec (IOPS)

4,504

Average Log Disk Write Latency (ms)

1

8

ESRP Performance Test Summary Using the AMS1000 storage module building blocks, basic Exchange Server 2007 performance can easily scale between 3,125 and 25,000 mailboxes. The total expected IOPS based on the ESRP profile is 1,500 for one storage module (3,125 * 0.48) and 12,000 for eight modules (25,000 * 0.48), with an average response time of <= 20ms for database read and writes. The average storage module in the ESRP submission supported 1,838 database IOPS (+338) with 16ms and 4ms average read and write latencies, well below the critical 20ms benchmark established by Microsoft. As these latencies breach the 20ms+ levels, end-user experience begins to degrade significantly. The average ESRP storage module maintains roughly 20 percent of IOPS and latency headroom above the tested profile expected results. The scalability of the AMS1000 is validated as the same headroom factor is maintained when eight storage modules are scaled out to support 25,000 mailboxes. However, this built-in headroom may be needed to provide the resources necessary to support the workloads that correspond to certain protection requirements. The potential impact on performance of data protection options is covered in the following section.

Exchange Data Protection Performance Planning for Backup and Replication The performance summary in the preceding section showed results for a standalone environment with minimal data protection. Adding backup and replication functions to any application solution, including Exchange Server 2007, requires effective planning for these processes. Backup and replication place demands on one or more components of an Exchange Server 2007 infrastructure and require additional storage resources, regardless of the deployed technology. Some of the associated costs, benefits, and requirements that accompany host and storage based backup and replication solutions are summarized in Table 8. Table 8. Table Host and Storage-based Backup/Replication Considerations

Method

Host RAM and CPU Requirements

Added Workload to Production Exchange Required RPO (Recovery Storage Backup Window Point Objective)

RTO (Recovery Time Objective)

Host-based Backup

Medium to High

Medium to High

Hours

Hours to Days

Hours to Days

Minutes to Days

Minutes to Days2

Hours to Days

Minutes to Hours

Host-based Replication

Medium to High

Low to High

N/A1

Storage-based Backup/ Replication

Low

Low to Medium

Seconds to Minutes

1

Backup not performed, but can offload backup process on the passive replica

2

If restore is required

E-mail has become a critical resource for conducting business and is thus a high priority for any business continuity solution. Therefore, it is important to understand the areas where tradeoffs may be sought as the overall costs of the solution are evaluated against the business value provided.

9

Host-based Backup Software Traditional host-based backup applications for Exchange Server 2007 are a perfectly acceptable protection solution for some organizations, especially for small environments with less stringent data protection requirements. By nature, the streaming backup process for Exchange Server 2007 demands increased host resources during normal backup operations. Specifically, additional CPU cycles are required to perform the page checksum process, and RAM and disk resources must be sufficient to support the intensive sequential read and write workloads associated with the streaming backup. The amount of time required to perform this type of backup (commonly known as the backup window) is much greater than the time required for storage-based backup methods, which prohibits this type of solution from delivering attractive recovery point objective (RPO) and recovery time objective (RTO) levels. However, tape backups are a critical business requirement for many organizations and provide enhanced data protection when combined with storage-based replication as we will see later in this section. The AMS1000 ESRP Submission contains the complete details and test results for the AMS1000, which include metrics for the simulated streaming backup tests performed by JetStress during the various tests. Hitachi Data Protection Suite, powered by CommVault®, is described later in this document. It is a comprehensive backup solution that provides Microsoft’s Volume Shadow Copy service–based (VSS) and streaming API backup functionality that can be applied in a variety of designs to provide enhanced RPO and RTO objectives.

Host-based Replication Software—Including Microsoft LCR and CCR Exchange Server 2007 introduces new replication and availability features that may affect overall storage design. Local Continuous Replication (LCR) and Cluster Continuous Replication (CCR) enhance the recoverability of Exchange 2007 by complementing, not replacing, backup solutions. Each technology provides varying levels of protection by providing a passive copy for rapid recovery. However, a VSS-based or steaming API backup solution is still required for LCR or CCR implementations where true data protection is required. Please refer to Microsoft’s specific guidelines and limitations for LCR and CCR. Local Continuous Replication LCR provides functionality that maintains a second copy of a storage group locally on a standalone Exchange 2007 server, which can facilitate recovery when the second copy is manually activated. In essence, when deployed using modern, high-availability storage systems from Hitachi Data Systems, the value of LCR is limited to offloading the backup process to the second copy and potential recovery capabilities from some logical corruptions. LCR uses the Exchange Server 2007 Replication Service, which requires additional server hardware resources, specifically processor cycles and memory address space. Standalone Exchange 2007 servers commonly have the Hub Transport role configured, in addition to the Mailbox role, which also increases the server CPU and RAM requirements to accommodate the additional workload. The recommendations provided by Microsoft for additional resources for CPU and RAM, and other server and storage hardware requirements, should be followed. Adding LCR on the AMS1000 entails allocating additional storage group storage modules (described previously) for the passive copy. This satisfies two key storage recommendations for LCR-enabled storage groups: • Isolate the physical disks for the active and passive copies • Ensure that the disks allocated to the passive copies have the same performance capabilities as the disks used

to support the active copy Cluster Continuous Replication CCR leverages Exchange Server 2007 asynchronous log-shipping technology to replicate storage groups between the active and passive nodes of a cluster. In addition, CCR is integrated into Microsoft Cluster Services to facilitate Exchange service management and failover. While still resource intensive, many of the key processes and their associated performance impacts occur on the passive Exchange node. The primary Exchange server does

10

incur a resource penalty due to the overhead of managing the replication service, a network share, and other components. Because log shipping technologies inherently operate at a point in time that is behind the production system, the RPO and RTO levels delivered through host-based replication, in most cases, cannot match those of a zero-data-loss storage-based replication solution. The storage architecture for CCR on the AMS1000 using the storage module design from the ESRP submission entails allocating storage modules for the passive copy on an additional, often remotely located, AMS1000. This architecture, combined with the storage group and mailbox store design of the AMS1000-based storage module, ensures that the disks allocated to maintain the passive copies have the same performance and capacity capabilities as the disks on which the active copy resides. Native host-based replication for Exchange Server 2007 is new Microsoft technology; testing of those new features was not conducted during the initial ESRP submission for the AMS1000. However, one can use the ESRP results for the AMS1000 along with Microsoft’s replication guidelines to estimate the performance impact in terms of IOPS, which may be significant for larger Exchange environments.

Storage-based Backup/Replication The performance impact associated with using storage-based replication technologies of the AMS1000 will vary based on a number of factors outlined in detail in AMS1000 documentation. However, as a rule of thumb, expect to incorporate a distance-related performance allowance into the design to account for the increased data protection workload. The allowance may need to be increased for some of the larger, higher IOPS Exchange environments targeted at the upper range of the 25,000-user scale when using one of the advanced replication architectures. Similarly, performance impacts may be negligible for smaller environments using only one to four storage modules and which would likely use more basic replication architectures. The protected Hitachi replication architectures will be discussed in a later section. Detailed planning and design are critical for Exchange Server 2007 environments that can be characterized as having high IOPS, aggressive RPO/RTO, demanding business continuity requirements or otherwise possessing advanced levels of technical or design complexity. These solutions may benefit significantly from the assistance of performance and business continuity consultants from Hitachi Data Systems Global Solution Services (GSS).

Designing for Performance and Protection Specific production requirements will rarely be identical to a test environment due to differing module and option configurations, additional applications, and the specifics of the physical environment, including the distance between sites and the type of connectivity. In this case, the strength of the scalable AMS1000 Exchange design is apparent because it provides IOPS and response-time benchmarks that one can apply as building blocks in conjunction with other technologies to meet specific demands. Using the Exchange Server 2007 test profile described in the ESRP submission, each reference storage module processed between 2,350 and 2,450 IOPS (including database and log I/Os). When eight storage modules were stacked together, the AMS1000 supported a simulated Exchange Server 2007 environment consisting of eight servers that maintain 25,000 mailboxes. The average database read and write latencies measured 16ms and 4ms respectively, which equates to 25 percent and 75 percent below the 20ms response time target. Sustained IOPS that occur within an acceptable response time (which is the primary validation goal of JetStress) are the key metrics one needs to consider when building Hitachi Data Systems protected architectures for Exchange Server 2007. Designing a high-performing and resilient solution from the storage module building blocks requires allowing for the performance impact associated with the backup and/or replication processes in terms of IOPS and response time. Regardless of the backup and/or replication technology, additional workload requirements are placed on the

11

server and/or the storage resources composing the Exchange Server 2007 infrastructure. It is necessary to incorporate the increased resource demands of backup and replication functions into the storage design to ensure application performance does not degrade during times of high activity. Increasing the amount of memory (RAM) on the Exchange Server 2007 hosts can provide significant performance benefits. Specifically, an increased number of IOPS, reduced response times, and/or reduction in host-based replication impact can be realized with increased levels of RAM. Microsoft indicates that 32GB of RAM is the practical, not technical, limit for Exchange Server 2007 hosts, as the benefit of increasing RAM beyond this level does not provide enough return to justify the significant costs. From a storage perspective, two possible methods of adapting a reference storage module from the ESRP submission for use in building protected solution architectures involve: • Decreasing the number of mailboxes allocated to the mailbox store in each storage group for the Exchange

instance. This decreases the total number of mailboxes housed by the Exchange instance, thereby reducing the I/O workload to the physical storage resources associated with those storage groups. • Increasing the number of RAID-1+0 disk spindles allocated to the Adaptable Modular Storage system RAID

groups, which comprise the production LUNs for the database file for a given storage group. This allocates the mailbox store file over an increased number of physical disks from the larger RAID-1+0 RAID group, which provides the ability to support an increased number of IOPS within an acceptable average response time. By decreasing the number of mailboxes we are simply accommodating the IOPS and response time requirements for the backup and/or replication process with the IOPS and response time requirements of the reduced number of mailbox stores for the Exchange instance. The benchmark IOPS (2400, including logs) and response time metrics (16ms) from the average reference storage module should be the targeted maximum levels when the impact of backup and/or replication processes is included. However, ensure that a 10 percent to 15 percent factor is built into the calculations in order to account for growth and spikes in latency. Alternatively, by increasing the number of physical disk spindles that compose the RAID groups for the database LUNs for a storage group we can account for backup/replication overhead. This method stripes the mailbox store file (database) over an increased number of physical disks, providing the ability to support an increased number of IOPS within an acceptable average response time. The benchmark IOPS (2400, including logs) and response time metrics (16ms) from the average storage module now become the expected minimum levels supported when additional physical spindles are added to the Adaptable Modular Storage system RAID groups for the mailbox store LUNs. Again, ensure that a 10 percent to 15 percent factor is built into the calculations in order to account for growth, spikes in latency-producing, nontransactional I/O, and maintenance activities. This calculation can easily get complex and requires restructuring the reference storage module to account for the additional spindles. Therefore, be sure to involve Hitachi Data Systems Global Solution Services when using this alternate technique.

Key Takeaways The following list summarizes the key Exchange Data Protection takeaways and provides additional guidance for designing and validating Exchange Server 2007 solution architectures: • The AMS1000-based storage module architecture provides a benchmark building-block performance metric for

Exchange-solution construction. However, each situation is unique and the test results should be used only as a guideline rather than as an absolute predictor of performance. • When material changes to the basic storage module architecture are required, it is strongly recommended that

additional Microsoft JetStress testing be conducted against all anticipated I/O and non-transactional workloads, and that the appropriate Hitachi Data Systems technical resources be engaged.

12

• Be conservative when developing workload and resource estimates and when mitigating peak periods of heavy

I/O and nontransactional workloads. • JetStress testing is just one of the Microsoft recommended steps to establish readiness and supportability for

an Exchange Server 2007 environment. However, it is crucial for validating a storage configuration, especially when designing for data protection. • LoadGen validation testing that includes the architected backup and/or replication processes and overhead is

highly recommended.

Storage-based Protection for Exchange The following components are used in the building blocks shown for Hitachi-protected Exchange Server 2007 solutions.

Hardware • Hitachi Adaptable Modular Storage model AMS1000

Software • Hitachi HiCommand® Device Manager 5.5—provides centralized management of distributed Hitachi

environments • Hitachi HiCommand Dynamic Link Manager 5.8—provides robust path failover and load balancing for Hitachi

storage in open systems operating environments • Hitachi HiCommand Protection Manager 5.7—protects mission-critical applications by automating split-mirror

backups and enabling rapid recovery using proven Hitachi replication technologies • Hitachi Command Control Interface—allows scripted control of replication procedures • Hitachi ShadowImage® Replication—provides nondisruptive, host-independent data replication for creating

copies of any customer accessible data within a single Hitachi storage system • Hitachi TrueCopy® Synchronous—provides continuous, nondisruptive, host-independent remote data

replication for disaster recovery • Hitachi Data Protection Suite (or other third-party backup software)—provides scheduled disk-to-tape or disk-

to-disk-to-tape backup of data volumes or files for offsite storage or local recovery

Adaptable Modular Storage model AMS1000: Hitachi Data Systems Protected Architectures for Exchange Server 2007 In order to more clearly present the protection alternatives that follow, it may be useful to first map some of the common elements between Exchange Server 2007 and the AMS1000. This paper first recaps the elements associated with a single Exchange server and the disks that make up the volumes for a single Exchange Server 2007 storage group. Figure 4 highlights the physical layout of the first tray of a single storage module. For a single storage group, note that two volumes are represented consisting of six physical disks—four disks for the mailbox store and two for the logs and checkpoint file.

13

Figure 4. Single Storage Group Disks Extracted from a Single Storage Module Hitachi Adaptable Modular Storage AMS1000 Single Module (1 Storage Group)

RKA 1

SG1 Mailbox Store (.edb)

Allocated

SG1 Log & Checkpoint (.log & .chk)

Spare

Up to this point, discussions of the AMS1000-based storage modules have focused on the production Exchange Server 2007 LUNs, RAID groups and physical disks. In order to build protected architectures that use Hitachi storage–based replication solutions, storage modules must be allocated to hold one or more replication volumes, commonly known as secondary volumes (S-VOLs), and then the S-VOLs must be mapped to the primary physical disk configuration on the AMS1000. Additional disk trays (RKAs) and racks will be required to expand the AMS1000 to house the S-VOL physical disks. Also, remember that the logical drawings that accompany the Hitachi Data Systems protected reference architectures introduced in this section represent only one Exchange storage group, which is composed of two dedicated LUNs—one for the Exchange database and one for the logs. The straightforward method for developing the S-VOL storage modules is to allocate one additional storage module for each S-VOL you want to maintain, providing S-VOL LUN construction and allocation that identically matches the characteristics of the P-VOLs. This design provides optimal replication performance by isolating workloads and storage resources and avoids disk contention areas also referred to as hot spots. Figure 5 illustrates a single storage module for the P-VOLs, and the additional storage module architecture required for two copies of S-VOLs. Additional S-VOLs are required to provide multiple recovery points in time, which are used to rapidly restore and recover from multiple scenarios. The number of S-VOLs and the frequency of refresh via the backup process are key factors in determining the recovery service level objectives that a storage-based backup and replication solution can support. Each additional copy provides an enhanced level of data protection, but an additional set of storage resources are required to maintain each backup copy.

14

Figure 5. Primary and Secondary Volume Storage Modules for Replication Hitachi Adaptable Modular Storage AMS1000 Single Module (S-VOL #2) RKA 20

RKA 19

Hitachi Adaptable Modular Storage AMS1000 Single Module (P-VOLs) DB SG1

DB SG2

DB SG3

DB SG4

DB SG5

Log SG1 Log SG2

Log SG3 Log SG4

Log SG5

Spare

RKA 2

RKA 1

Hitachi Adaptable Modular Storage AMS1000 Single Module (S-VOL #1) DB SG1

DB SG2

DB SG3

DB SG4

DB SG5

Log SG1 Log SG2

Log SG3 Log SG4

Log SG5

Spare

RKA 20

RKA 19

DB SG1

DB SG2

DB SG3

DB SG4

DB SG5

Log SG1 Log SG2

Log SG3 Log SG4

Log SG5

Spare

Notice that three AMS1000–based storage modules are required. These reside, in this example, on separate trays depicted by the RKA number but connected to the same AMS1000. One storage module would be allocated to the production Exchange server volumes and the other two storage modules would be used for unique S-VOL copies. A fourth storage module would be required to maintain a third S-VOL if needed. An alternative method that can reduce cost entails constructing the S-VOL LUNs using one or more of the following techniques: • Using larger capacity physical disks • Increasing the number of physical spindles in the RAID group

However, realize that storage-based replication carries with it associated resource costs as well, so ensure that during recovery, when using larger or slower drives in a different RAID configuration, your design can maintain the RPO, RTO, and SLO solution goals without introducing response time or IOPS degradation to the production Exchange Server 2007 workloads. Hitachi Data Systems Global Solution Services offers a variety of professional services to assist with meeting these demands. Figure 6 illustrates an AMS1000–based storage module configured to use 300GB, 10,000 RPM Fibre Channel drives, in place of 146GB 15,000 RPM drives. This allows two S-VOLs to reside on a single storage module.

15

Figure 6. Single Storage Module for Replication Supporting Multiple S-VOLs Hitachi Adaptable Modular Storage AMS1000

Hitachi Adaptable Modular Storage AMS1000

Single Module (P-VOLs)

Single Module (Multiple S-VOL)

RKA 2

RKA 2

RKA 1

RKA 1

Secondary Copy #1 DB SG1

DB SG2

DB SG3

DB SG4

DB SG5

Log SG1 Log SG2

Log SG3 Log SG4

Log SG5

Spare

DB SG1

DB SG2

DB SG3

DB SG4

DB SG5

Log SG1 Log SG2

Log SG3 Log SG4

Log SG5

Spare

Secondary Copy #2 DB SG1

DB SG2

DB SG3

DB SG4

DB SG5

Log SG1 Log SG2

Log SG3 Log SG4

Log SG5

For the diagrams that follow, the primary volumes (P-VOLs) represent the primary database and log volumes for an entire storage group, and the secondary volumes (S-VOLs) represent image replicas that are maintained for rapid recovery on local storage, remote storage, or both. Figure 7 illustrates how the Exchange Server 2007, Windows Server 2003, ShadowImage and Protection Manager software, and AMS1000 map to one another. Figure 7. Exchange Server Components and Resources Mapped to Storage Volumes

1:00 AM

.edb file

2:00 AM

Mount Point PVOL PVOL

.log file

1:00 AM 2:00 AM

Mount Point .chk file Microsoft Exchange Server

Storage Group

PVOL PVOL

Mailbox Store Hitachi Adaptable Modular Storage AMS1000

Examples of Building-block-driven Hitachi Data Systems Protected Solution Architectures This section discusses the Hitachi Data Systems protected architectures available for Exchange Server 2007 on the AMS1000. As above, the protected architectures are applied in a building-block manner to produce a solution architecture that meets specific Exchange Server 2007 protection requirements. In addition, this section provides examples that illustrate how individual components can be combined to develop complete, protected solution architectures.

16

Basic Replication Architectures for Hitachi HiCommand Protection Manager software Figure 8 highlights the basic local replication architectures for Hitachi HiCommand Protection Manager when used with ShadowImage Replication software. Each P-VOL and S-VOL container in the following drawings represents a single Exchange Server 2007 storage group—not an entire Exchange instance or server. The P-VOL container corresponds to the two volumes—one for the mailbox store and the other for the logs and checkpoint file—and the S-VOL container represents the two volumes that contain the replicated recovery points of these storage group files. The S-VOL containers that are dashed in the drawing represent multiple copies, which are recommended to improve recoverability in the case of corruption or other data loss that is not discovered immediately. Each additional S-VOL provides an additional recovery point, which enhances the protection level of the overall Exchange Server 2007 solution. Figure 8. Basic Replication Capabilities of HiCommand Protection Manager Software Used with ShadowImage Software Single Generation Hitachi ShadowImage® Replication software

ShadowImage PSUS

Multiple Generation Hitachi ShadowImage® Replication software

ShadowImage PSUS

Figure 9 illustrates the basic local protection components for Exchange Server 2007 based on Protection Manager and ShadowImage software. The backup (import) server is used to confirm the integrity of the single secondary copy, via Eseutil (Extensible Storage Engine Utility). In most cases, the backup (import) server can be shared in multi-Exchange instance environments. This is outlined in the HiCommand Protection Manager documentation.

17

Figure 9. Hitachi Protection Manager Base Components LAN

Hitachi HiCommand® Protection Manager software

HiCommand Protection Manager software

CCI

CCI

Hitachi Resource Manager™ Shadow Copy Provider

Resource Manager Shadow Copy Provider

Hitachi Data Protection Suite Media Server Microsoft VSS Import Server

Production Server Microsoft Exchange or SQL

Backup Software’s Agent

Backup Software’s Master Agent

Switch

PVOLS

Hitachi Adaptable Modular Storage AMS500

SVOLS

Backup on Tape Library

The primary Exchange volumes are replicated locally using Protection Manager along with Volume ShadowCopy Service (VSS) and Virtual Disk Service (VDS) integration, to make image copies that can be used for rapid local recovery or that can be accessed for other protection alternatives. This elementary level of protection can allow recovery if corruption of the active Exchange instance is discovered before the copy is over-written by a subsequent backup. However, with only a single copy and no offsite backup, there is little true data protection. This point highlights the importance of maintaining multiple S-VOLs, providing multiple recovery points from which service can be rapidly restored. HiCommand Protection Manager software provides capabilities that can be used in conjunction with replication, tape backup and/or Hitachi Storage Cluster architectures. These capabilities are applicable to Exchange instances using the Single Copy Cluster and Majority Node Set cluster models available with Microsoft Cluster Services, in addition to the standalone model as illustrated in the previous drawing.

18

Multi-generation Protection Manager with Hitachi Data Protection Suite Figure 10 illustrates the components of a more robust solution architecture for Exchange Server 2007 that combines HiCommand Protection Manager, ShadowImage Replication and Hitachi Data Protection Suite software. Figure 10. Hitachi Protection Manager with Hitachi Data Protection Suite Protected Architecture Components LAN

Hitachi HiCommand® Protection Manager software

HiCommand Protection Manager software

CCI

CCI

Hitachi Resource Manager™ Shadow Copy Provider

Resource Manager Shadow Copy Provider

Hitachi Data Protection Suite Media Server Microsoft VSS Import Server

Production Server Microsoft Exchange or SQL

Backup Software’s Agent

Backup Software’s Master Agent

Switch

PVOLS

Hitachi Adaptable Modular Storage AMS500

SVOLS

Backup on Tape Library

In this example, the multiple generation replication architecture for ShadowImage software is applied to provide the replication component of HiCommand Protection Manager software, which was discussed earlier. This solution architecture combines the split-mirror backup, rapid recovery protected architecture of Protection Manager software with the preservation and retention protected architecture of Hitachi Data Protection Suite to create a resilient disk-to-disk-to tape backup solution. This is the most common deployment model for Protection Manager in medium-sized organizations.

Advanced Replication Architectures with HiCommand Protection Manager Software In addition to providing local copies for rapid recovery purposes, HiCommand Protection Manager software also has the ability to maintain a remote copy using Hitachi TrueCopy Synchronous software, which can be used for disaster recovery or for facilitating the replication leg of a solution designed to offload the tape backup process to the remote site.

19

Figure 11. Advanced Replication Model Combining HiCommand Protection Manager, ShadowImage Replication and TrueCopy Synchronous Software Single Generation Hitachi TrueCopy® Synchronous and Hitachi ShadowImage® Replication Software Generation(s) TrueCopy PSUS

ShadowImage PSUS

Combining HiCommand Protection Manager Software and Hitachi Storage Cluster Figure 11 highlights a more advanced replication architecture for Exchange Server 2007 using HiCommand Protection Manager software and Hitachi Storage Cluster combined with ShadowImage Replication and TrueCopy Synchronous software. Multiple S-VOLs provide additional recovery points, which enhance the protection level of the overall Exchange Server 2007 solution. Figure 12. Advanced Replication Model Combining Hitachi Storage Cluster and HiCommand Protection Manager, ShadowImage Replication and TrueCopy Synchronous Software Hitachi Storage Cluster Integration Hitachi TrueCopy® Synchronous Software Pair

Hitachi ShadowImage® Replication Software PSUS

ShadowImage PSUS

Hitachi Storage Cluster Integration (Normal Operation)

Hitachi Storage Cluster Integration (After Cluster Failover)

TrueCopy Pair

TrueCopy Pair

ShadowImage PSUS

ShadowImage PSUS

ShadowImage PSUS

SI PSUS

The grayed-out volumes in Figure 12 indicate when ShadowImage Replication software is inactive because a site does not house the active Exchange cluster node.

20

For the following advanced protection Reference Architecture, multiple-generation ShadowImage copies are combined with HiCommand Protection Manager and Hitachi TrueCopy Synchronous software and Hitachi Storage Cluster. Figure 13. Multiple-generation Protected Architecture Components for HiCommand Protection Manager Software and Hitachi Storage Cluster LAN/WAN

Hitachi HiCommand® Protection Manager Software CC (x32/x64)

Microsoft Windows Server 2003 x64 Enterprise Edition

Windows Server 2003 (x84/x64) Enterprise Edition

Windows Server 2003 x64 Enterprise Edition

Hitachi Resource Manager™ Shadow Copy Provider

Exchange 2007 (x64)

MSCS Majority Node Set

Exchange 2007 (x64)

CCI/Resource Manager Shadow Copy Provider

MSCS Majority Node Set

Exchange 2007 MC (x32/x64)

Microsoft Exchange 2007 MC (x32/x64)

Microsoft Cluster Server (MSCS) Majority Node Set

HSC Generic Script Resource

Hitachi Storage Cluster (HSC) Generic Script Resource

HiCommand Protection Manager for Exchange (x64)

HiCommand Protection Manager for Exchange (x64) HiCommand Protection Manager CC (x64) CCI/Resource Manager Shadow Copy Provider

HiCommand Protection Manager CC (x32/x64)

HiCommand Protection Manager CC (x64) Tie-breaker Site Majority Node Set File Share Witness

CCI/Resource Manager Shadow Copy Provider

Secondary Site Import Server

Primary Site Import Server

Primary Site Majority Node Set Exchange Mailbox Server Active Cluster Node

Secondary Site Majority Node Set Exchange Mailbox Server Passive Cluster Node

Hitachi TrueCopy® Synchronous Software

Hitachi Adaptable Modular Storage AMS1000

Hitachi Adaptable Modular Storage AMS1000

This solution architecture integrates the local, rapid backup and recovery functionality of Protection Manager software with the disaster recovery capabilities provided by Hitachi Storage Cluster to provide a superior level of data protection for Exchange Server 2007. Multiple local recovery points complement the synchronously connected remote Exchange cluster node to facilitate protection against a variety of outage scenarios, ranging from an isolated virus, to a server crash, to a complete site failure. The architecture of this solution provides the ability for organizations to realize extremely low RPO and RTO, and aggressive SLO.

21

Services Planning, design and implementation services play a key role in the successful deployments of protected environments. Hitachi Data Systems offers a range of services, many of which are crucial in constructing the solutions introduced. The following services leverage the hardware and software components of the various reference architectures to design, implement and validate highly protected, resilient Exchange Server 2007 infrastructures: • Implementation Services for Hitachi Data Protection Suite, powered by CommVault® • Hitachi SplitSecond™ Solutions • Remote Copy Planning and Design Service • Hitachi Storage Cluster for Microsoft® Environments Services • Risk Assessment Service

Summary As enhanced features and functionality drive broader adoption and migration to Exchange Server 2007, Hitachi Data Systems expects increased demand for mission-critical infrastructures, like the AMS1000, to support that deployment. As highlighted, a high-performing, resilient storage infrastructure is a critical success factor for supporting a diverse set of Exchange Server 2007 deployments. The ESRP submission for Exchange Server 2007 on the AMS1000 provides the foundation upon which solution architectures supporting varying levels of scalability, protection and performance can be built. The storage modules and reference architectures introduced in this paper provide the building-block components necessary to architect a high-performing and highly available Exchange Server 2007 data protection solution leveraging the AMS1000. These solutions are critical components in our customer’s overall efforts to meet their business needs while leveraging and maximizing investments in Microsoft and Hitachi Data Systems technology. The building-block framework introduced in this document can be applied to other Hitachi Data Systems ESRP submissions for Exchange Server 2007 to arrive at the optimal storage configuration for a given set of budget, performance, data protection and business requirements. However, be sure to follow the recommended best practices outlined in the ESRP for that particular family of Hitachi storage. Workloads and data protection requirements are unique for each organization. Accordingly, additional design and testing is always recommended, especially when data protection requirements are taken into account. Hitachi Data Systems will continue to develop Reference Architecture materials for Exchange Server 2007. These documents will provide technical insights, recommendations and configuration details used in this document to produce the data protection reference architectures. For additional information and contacts, please review www.hds.com and the Hitachi Data Systems Microsoft partner page at www.hds.com/microsoft.

22

Appendix: Supplemental Exchange Server 2007 Information The content in this paper applies the principles, guidance and best practices found in the publicly available information contained in the following Microsoft publications. Document Name

Hyperlink

Deploying a Simple Exchange Organization

http://technet.microsoft.com/en-us/library/bb201659.aspx

Deploying a Standard Exchange Organization

http://technet.microsoft.com/en-us/library/bb201727.aspx

Deploying a Large Exchange Organization

http://technet.microsoft.com/en-us/library/bb201708.aspx

Deploying a Complex Exchange Organization

http://technet.microsoft.com/en-us/library/bb201752.aspx

Planning for Edge Transport Servers

http://technet.microsoft.com/en-us/library/aa996562.aspx

Planning for Hub Transport Servers

http://technet.microsoft.com/en-us/library/aa998616.aspx

Planning for Client Access Servers

http://technet.microsoft.com/en-us/library/bb232184.aspx

Planning for Mailbox Servers

http://technet.microsoft.com/en-us/library/bb201699.aspx

Planning for Unified Messaging Servers

http://technet.microsoft.com/en-us/library/bb125141.aspx

Exchange Server 2007 Planning and Architecture

http://technet.microsoft.com/en-us/library/aa998636.aspx

Best Practices for Deploying a New Exchange Organization

http://technet.microsoft.com/en-us/library/bb123517.aspx

Best Practices for Transitioning an Exchange Organization

http://technet.microsoft.com/en-us/library/bb124920.aspx

Best Practices for Migrating to an Exchange Organization

http://technet.microsoft.com/en-us/library/b841fe7e-8961-4962-8f4b55d89dc9ef51.aspx

Features of Exchange Server 2007

http://www.microsoft.com/exchange/evaluation/features/default.mspx

Exchange Server 2007 Product Documentation

http://technet.microsoft.com/en-us/library/bb124558.aspx

Planning Processor and Memory Configurations

http://technet.microsoft.com/en-us/library/aa998874.aspx

Planning Disk Storage for Exchange Server 2007

http://technet.microsoft.com/en-us/library/bb124518.aspx

Exchange Server 2007 Security and Protection

http://technet.microsoft.com/en-us/library/aa996775.aspx

High Availability Strategies for Exchange Server 2007

http://technet.microsoft.com/en-us/library/bb123523.aspx

Local Continuous Replication

http://technet.microsoft.com/en-us/library/bb125195.aspx

Cluster Continuous Replication

http://technet.microsoft.com/en-us/library/bb124521.aspx

Single Copy Cluster

http://technet.microsoft.com/en-us/library/bb125217.aspx

Exchange Server 2007 High Availability

http://technet.microsoft.com/en-us/library/bb124721.aspx

Clustering Architecture Essentials

http://technet2.microsoft.com/windowsserver/en/library/42ba8090-cbb8-4e90ad7f-bc09d956bdac1033.mspx?mfr=true

Scheduled and Unscheduled Outages

http://go.microsoft.com/fwlink/?LinkId=65320

Optimizing Storage for Exchange Server

http://www.microsoft.com/technet/prodtechnol/exchange/2003/library/optimizestor age.mspx

Planning for Local Continuous Replication

http://technet.microsoft.com/en-us/library/bb124704.aspx

Planning for Cluster Continuous Replication

http://technet.microsoft.com/en-us/library/bb123996.aspx

23

Document Name

Hyperlink

Disaster Recovery Strategies

http://go.microsoft.com/fwlink/?LinkId=65320

Managing Storage Groups and Databases

http://go.microsoft.com/fwlink/?LinkId=65320

24

Hitachi Data Systems Corporation Corporate Headquarters 750 Central Expressway, Santa Clara, California 95050-2627 USA
Contact Information: + 1 408 970 1000 www.hds.com / [email protected]
Asia Pacific and Americas 750 Central Expressway, Santa Clara, California 95050-2627 USA
Contact Information: + 1 408 970 1000 www.hds.com / [email protected]
Europe Headquarters Sefton Park, Stoke Poges, Buckinghamshire SL2 4HD United Kingdom
Contact Information: + 44 (0) 1753 618000 www.hds.com / [email protected]

Hitachi is a registered trademark of Hitachi, Ltd. and/or its affiliates in the United States and other countries. Hitachi Data Systems is a registered trademark and service mark of Hitachi, Ltd., in the United States and other countries. HiCommand is a registered trademark of Hitachi, Ltd. TrueCopy and ShadowImage are registered trademarks and SplitSecond is a trademark of Hitachi Data Systems Corporation. IBM and BladeCenter are registered trademarks of International Business Machines Corporation. Microsoft is a registered trademark of Microsoft Corporation. CommVault is a registered trademark of CommVault Systems, Inc. All other product and company names are, or may be, trademarks or service marks of their respective owners. Notice: This document is for informational purposes only, and does not set forth any warranty, express or implied, concerning any equipment or service offered or to be offered by Hitachi Data Systems. This document describes some capabilities that are conditioned on a maintenance contract with Hitachi Data Systems being in effect, and that may be configuration-dependent, and features that may not be currently available. Contact your local Hitachi Data Systems sales office for information on feature and product availability. Hitachi Data Systems sells and licenses its products to certain terms and conditions, including limited warranties. To see a copy of these terms and conditions prior to purchase or license, please go to http://www.hds.com/products_services/support/warranty.html or call your local sales representative to obtain a printed copy. If you purchase or license the product, you are deemed to have accepted these terms and conditions. © Hitachi Data Systems Corporation 2007. All Rights Reserved WHP-263-00 September 2007