Exchange 2010 Tested Solutions: 9000

Mailboxes in Two Sites Running Hyper-V on

Dell M610 Servers, Dell EqualLogic Storage,
and F5 Load Balancing Solutions
Rob Simpson, Program Manager, Microsoft Exchange Server; Akshai Parthasarathy, Systems
Engineer, Dell; Casey Birch, Product Marketing Manager for Exchange Solutions, Dell
December 2010

In Exchange 2010 Tested Solutions, Microsoft and participating server, storage, and network
partners examine common customer scenarios and key design decision points facing customers
who plan to deploy Microsoft Exchange Server 2010. Through this series of white papers, we
provide examples of well-designed, cost-effective Exchange 2010 solutions deployed on
hardware offered by some of our server, storage, and network partners.
You can download this document from the Microsoft Download Center.

Applies to:
Microsoft Exchange Server 2010 release to manufacturing (RTM)
Microsoft Exchange Server 2010 with Service Pack 1 (SP1)
Windows Server 2008 R2
Windows Server 2008 R2 Hyper-V
Table of Contents

Solution Summary

Customer Requirements

Mailbox Profile Requirements

Geographic Location Requirements

Server and Data Protection Requirements

Design Assumptions

Server Configuration Assumptions

Storage Configuration Assumptions

Solution Design

Determine High Availability Strategy

Estimate Mailbox Storage Capacity Requirements

Estimate Mailbox I/O Requirements

Determine Storage Type

Choose Storage Solution

Determine Number of EqualLogic Arrays Required

Estimate Mailbox Memory Requirements

Estimate Mailbox CPU Requirements

Determine Whether Server Virtualization Will Be Used

Determine Whether Client Access and Hub Transport Server Roles Will Be Deployed in
Separate Virtual Machines

Determine Server Model for Hyper-V Root Server

Determine the CPU Capacity of the Virtual Machines

Determine Number of Mailbox Server Virtual Machines Required

Determine Number of Mailboxes per Mailbox Server

Determine Memory Required Per Mailbox Server

Determine Number of Client Access and Hub Transport Server Combo Virtual Machines
Required

Determine Memory Required per Combined Client Access and Hub Transport Virtual
Machines

Determine Virtual Machine Distribution

Determine Memory Required per Root Server

Determine Minimum Number of Databases Required

Identify Failure Domains Impacting Database Copy Layout

Design Database Copy Layout

Determine Storage Design

Determine Placement of the File Share Witness

Plan Namespaces

Determine Client Access Server Array and Load Balancing Strategy

Determine Hardware Load Balancing Solution

Determine Hardware Load Balancing Device Resiliency Strategy

Determine Hardware Load Balancing Methods

Solution Overview

Logical Solution Diagram

Physical Solution Diagram

Server Hardware Summary

Client Access and Hub Transport Server Configuration

Mailbox Server Configuration

Database Layout

Storage Hardware Summary

Storage Configuration

Network Switch Hardware Summary

Load Balancer Hardware Summary

Solution Validation Methodology

Storage Design Validation Methodology

Server Design Validation

Functional Validation Tests

Datacenter Switchover Validation

Primary Datacenter Service Restoration Validation

Storage Design Validation Results

Server Design Validation Results

This document provides an example of how to design, test, and validate an Exchange Server
2010 solution for environments with 9,000 mailboxes deployed on Dell server and storage
solutions and F5 load balancing solutions. One of the key challenges with designing Exchange
2010 environments is examining the current server and storage options available and making the
right hardware choices that provide the best value over the anticipated life of the solution.
Following the step-by-step methodology in this document, we walk through the important design
decision points that help address these key challenges while ensuring that the customer's core
business requirements are met. After we have determined the optimal solution for this customer,
the solution undergoes a standard validation process to ensure that it holds up under simulated
production workloads for normal operating, maintenance, and failure scenarios.
Return to top

Solution Summary
The following tables summarize the key Exchange and hardware components of this solution.
Exchange components
Exchange component

Value or description

Target mailbox count

9000

Target mailbox size

750 megabytes (MB)

Target message profile

103 messages per day

Database copy count

Volume Shadow Copy Service (VSS) backup

None

Exchange component

Value or description

Site resiliency

Yes

Virtualization

Hyper-V

Exchange server count

18 virtual machines (VMs)

Physical server count

Hardware components
Hardware component

Value or description

Server partner

Dell

Server model

PowerEdge M610

Server type

Blade

Processor

Intel Xeon X5550

Storage partner

Dell EqualLogic

Storage type

Internet SCSI (iSCSI) storage area network

(SAN)

Disk type

1 terabyte 7.2 kilobyte (KB) Serial ATA (SATA)

3.5"

Return to top

Customer Requirements
One of the most important first steps in Exchange solution design is to accurately summarize the
business and technical requirements that are critical to making the correct design decisions. The
following sections outline the customer requirements for this solution.
Return to top

Mailbox Profile Requirements

Determine mailbox profile requirements as accurately as possible because these requirements
may impact all other components of the design. If Exchange is new to you, you may have to
make some educated guesses. If you have an existing Exchange environment, you can use the
Microsoft Exchange Server Profile Analyzer tool to assist with gathering most of this information.
The following tables summarize the mailbox profile requirements for this solution.

Mailbox count requirements

Mailbox count requirements

Value

Mailbox count (total number of mailboxes

including resource mailboxes)

9000

Projected growth percent (%) in mailbox count

(projected increase in mailbox count over the
life of the solution)

0%

Expected mailbox concurrency % (maximum

number of active mailboxes at any time)

100%

Mailbox size requirements

Mailbox size requirements

Value

Average mailbox size in MB

750 MB (742)

Tiered mailbox size

Yes
450 @ 4 gigabytes (GB)
900 @ 1 GB
7650 @ 512 MB

Average mailbox archive size in MB

Projected growth (%) in mailbox size in MB

(projected increase in mailbox size over the life
of the solution)

included

Target average mailbox size in MB

750 MB

Mailbox profile requirements

Mailbox profile requirements

Value

Target message profile (average total number

of messages sent plus received per user per
day)

103 messages per day

Tiered message profile

Yes
450 @ 150 messages per day
8550 @ 100 messages per day

Target average message size in KB

75

% in MAPI cached mode

100

% in MAPI online mode

Mailbox profile requirements

Value

% in Outlook Anywhere cached mode

% in Microsoft Office Outlook Web App

(Outlook Web Access in Exchange 2007 and
previous versions)

% in Exchange ActiveSync

Return to top

Geographic Location Requirements

Understanding the distribution of mailbox users and datacenters is important when making design
decisions about high availability and site resiliency.
The following table outlines the geographic distribution of people who will be using the Exchange
system.
Geographic distribution of people
Mailbox user site requirements

Value

Number of major sites containing mailbox users

Number of mailbox users in site 1

9000

Number of mailbox users in site 2

The following table outlines the geographic distribution of datacenters that could potentially
support the Exchange e-mail infrastructure.
Geographic distribution of datacenters
Datacenter site requirements

Value or description

Total number of datacenters

Number of active mailboxes in proximity to

datacenter 1

9000

Number of active mailboxes in proximity to

datacenter 2

Requirement for Exchange to reside in more

than one datacenter

Yes

Return to top

Server and Data Protection Requirements

It's also important to define server and data protection requirements for the environment because
these requirements will support design decisions about high availability and site resiliency.
The following table identifies server protection requirements.
Server protection requirements
Server protection requirements

Value

Number of simultaneous server or VM failures

within site

Number of simultaneous server or VM failures

during site failure

The following table identifies data protection requirements.

Data protection requirements
Data protection requirement

Value or description

Requirement to maintain a backup of the

Exchange databases outside of the Exchange
environment (for example, third-party backup
solution)

No

Requirement to maintain copies of the

Exchange databases within the Exchange
environment (for example, Exchange native
data protection)

Yes

Requirement to maintain multiple copies of

mailbox data in the primary datacenter

Yes

Requirement to maintain multiple copies of

mailbox data in a secondary datacenter

Yes

Requirement to maintain a lagged copy of any

Exchange databases

No

Lagged copy period in days

Not applicable

Target number of database copies

Deleted Items folder retention window in days

14 days

Return to top

Design Assumptions
This section includes information that isn't typically collected as part of customer requirements,
but is critical to both the design and the approach to validating the design.
Return to top

Server Configuration Assumptions

The following table describes the peak CPU utilization targets for normal operating conditions,
and for site server failure or server maintenance conditions.
Server utilization targets
Target server CPU utilization design assumption

Value

Normal operating for Mailbox servers

<70%

Normal operating for Client Access servers

<70%

Normal operating for Hub Transport servers

<70%

Normal operating for multiple server roles

(Client Access, Hub Transport, and Mailbox
servers)

<70%

Normal operating for multiple server roles

(Client Access and Hub Transport servers)

<70%

Node failure for Mailbox servers

<80%

Node failure for Client Access servers

<80%

Node failure for Hub Transport servers

<80%

Node failure for multiple server roles (Client

Access, Hub Transport, and Mailbox servers)

<80%

Node failure for multiple server roles (Client

Access and Hub Transport servers)

<80%

Return to top

Storage Configuration Assumptions

The following tables summarize some data configuration and input/output (I/O) assumptions
made when designing the storage configuration.
Data configuration assumptions
Data configuration assumption

Value or description

Data overhead factor

20%

Data configuration assumption

Value or description

Mailbox moves per week

1%

Dedicated maintenance or restore logical unit

number (LUN)

No

LUN free space

20%

Log shipping compression enabled

Yes

Log shipping encryption enabled

Yes

I/O configuration assumptions

I/O configuration assumption

Value or description

I/O overhead factor

20%

Additional I/O requirements

None

Return to top

Solution Design
The following section provides a step-by-step methodology used to design this solution. This
methodology takes customer requirements and design assumptions and walks through the key
design decision points that need to be made when designing an Exchange 2010 environment.
Return to top

Determine High Availability Strategy

When designing an Exchange 2010 environment, many design decision points for high availability
strategies impact other design components. We recommend that you determine your high
availability strategy as the first step in the design process. We highly recommend that you review
the following information prior to starting this step:

Understanding High Availability Factors

Planning for High Availability and Site Resilience

Understanding Backup, Restore and Disaster Recovery

Step 1: Determine whether site resiliency is required

If you have more than one datacenter, you must decide whether to deploy Exchange
infrastructure in a single datacenter or distribute it across two or more datacenters. The
organization's recovery service level agreements (SLAs) should define what level of service is
required following a primary datacenter failure. This information should form the basis for this
decision.

*Design Decision Point*

In this example, there is a service level agreement that requires the ability to restore the
messaging service within four hours in the event of a primary datacenter failure. Therefore the
customer must deploy Exchange infrastructure in a secondary datacenter for disaster recovery
purposes.

Step 2: Determine relationship between mailbox user locations and

datacenter locations
In this step, we look at whether all mailbox users are located primarily in one site or if they're
distributed across many sites and whether those sites are associated with datacenters. If they're
distributed across many sites and there are datacenters associated with those sites, you need to
determine if there's a requirement to maintain affinity between mailbox users and the datacenter
associated with that site.
*Design Decision Point*
In this example, all of the active users are located in one primary location. The primary location is
in geographic proximity to the primary datacenter and therefore there's a desire for all active
mailboxes to reside in the primary datacenter during normal operating conditions.

Step 3: Determine database distribution model

Because the customer has decided to deploy Exchange infrastructure in more than one physical
location, the customer needs to determine which database distribution model best meets the
needs of the organization. There are three database distribution models:

Active/Passive distribution Active mailbox database copies are deployed in the primary
datacenter and only passive database copies are deployed in a secondary datacenter. The
secondary datacenter serves as a standby datacenter and no active mailboxes are hosted in
the datacenter under normal operating conditions. In the event of an outage impacting the
primary datacenter, a manual switchover to the secondary datacenter is performed and active
databases are hosted there until the primary datacenter returns online.
Active/Passive distribution

10

Active/Active distribution (single DAG) Active mailbox databases are deployed in the
primary and secondary datacenters. A corresponding passive copy is located in the alternate
datacenter. All Mailbox servers are members of a single database availability group (DAG). In
this model, the wide area network (WAN) connection between two datacenters is potentially a
single point of failure. Loss of the WAN connection results in Mailbox servers in one of the
datacenters going into a failed state due to loss of quorum.
Active/Active distribution (single DAG)

Active/Active distribution (multiple DAGs) This model leverages multiple DAGs to

remove WAN connectivity as a single point of failure. One DAG has active database copies in
the first datacenter and its corresponding passive database copies in the second datacenter.
The second DAG has active database copies in the second datacenter and its corresponding
passive database copies in the first datacenter. In the event of loss of WAN connectivity, the
active copies in each site continue to provide database availability to local mailbox users.

11

Active/Active distribution (multiple DAGs)

*Design Decision Point*

In this example, active mailbox users are only in a single location and only the secondary
datacenter will be used in the event that the primary datacenter fails. Therefore, an
Active/Passive distribution model is the obvious choice.

Step 4: Determine backup and database resiliency strategy

Exchange 2010 includes several new features and core changes that, when deployed and
configured correctly, can provide native data protection that eliminates the need to make
traditional data backups. Backups are traditionally used for disaster recovery, recovery of
accidentally deleted items, long term data storage, and point-in-time database recovery.
Exchange 2010 can address all of these scenarios without the need for traditional backups:

Disaster recovery In the event of a hardware or software failure, multiple database copies
in a DAG enable high availability with fast failover and no data loss. DAGs can be extended
to multiple sites and can provide resilience against datacenter failures.

Recovery of accidentally deleted items With the new Recoverable Items folder in
Exchange 2010 and the hold policy that can be applied to it, it's possible to retain all deleted
and modified data for a specified period of time, so recovery of these items is easier and
faster. For more information, see Messaging Policy and Compliance, Understanding
Recoverable Items, and Understanding Retention Tags and Retention Policies.

12

Long-term data storage Sometimes, backups also serve an archival purpose. Typically,
tape is used to preserve point-in-time snapshots of data for extended periods of time as
governed by compliance requirements. The new archiving, multiple-mailbox search, and
message retention features in Exchange 2010 provide a mechanism to efficiently preserve
data in an end-user accessible manner for extended periods of time. For more information,
see Understanding Personal Archives, Understanding Multi-Mailbox Search, and
Understanding Retention Tags and Retention Policies.

Point-in-time database snapshot If a past point-in-time copy of mailbox data is a

requirement for your organization, Exchange provides the ability to create a lagged copy in a
DAG environment. This can be useful in the rare event that there's a logical corruption that
replicates across the databases in the DAG, resulting in a need to return to a previous point
in time. It may also be useful if an administrator accidentally deletes mailboxes or user data.

There are technical reasons and several issues that you should consider before using the
features built into Exchange 2010 as a replacement for traditional backups. Prior to making this
decision, see Understanding Backup, Restore and Disaster Recovery.
*Design Decision Point*
In this example, maintaining tape backups has been difficult, and testing and validating restore
procedures hasn't occurred on a regular basis. Therefore, using Exchange native data protection
in place of traditional backups as the database resiliency strategy is preferred.

Step 5: Determine number of database copies required

There are a number of factors to consider when determining the number of database copies that
you'll deploy. The first is whether you're using a third-party backup solution. In the previous step,
this decision was made. We strongly recommend deploying a minimum of three copies of a
mailbox database before eliminating traditional forms of protection for the database, such as
Redundant Array of Independent Disks (RAID) or traditional VSS-based backups.
Prior to making this decision, see Understanding Mailbox Database Copies.
*Design Decision Point*
In the previous step, it was decided not to deploy a third-party backup solution. As a result, the
design should have a minimum of three copies of each database. This ensures that both the
recovery time objective and recovery point objective requirements are met.

Step 6: Determine database copy type

There are two types of database copies:

High availability database copy This database copy is configured with a replay lag time of
zero. As the name implies, high availability database copies are kept up-to-date by the
system, can be automatically activated by the system, and are used to provide high
availability for mailbox service and data.

Lagged database copy This database copy is configured to delay transaction log replay for
a period of time. Lagged database copies are designed to provide point-in-time protection,
which can be used to recover from store logical corruptions, administrative errors (for

13

example, deleting or purging a disconnected mailbox), and automation errors (for example,
bulk purging of disconnected mailboxes).
*Design Decision Point*
In this example, all three mailbox database copies will be deployed as high availability database
copies. The primary need for a lagged copy is to provide the ability to recover single deleted
items. This requirement can be met using the deleted items retention feature.

Step 7: Determine number of database availability groups

A DAG is the base component of the high availability and site resilience framework built into
Exchange 2010. A DAG is a group of up to 16 Mailbox servers that hosts a set of replicated
databases and provides automatic database-level recovery from failures that affect individual
servers or databases.
A DAG is a boundary for mailbox database replication, database and server switchovers and
failovers, and for an internal component called Active Manager. Active Manager is an
Exchange 2010 component, which manages switchovers and failovers. Active Manager runs on
every server in a DAG.
From a planning perspective, you should try to minimize the number of DAGs deployed. You
should consider going with more than one DAG if:

You deploy more than 16 Mailbox servers.

You have active mailbox users in multiple sites (active/active site configuration).

You require separate DAG-level administrative boundaries.

You have Mailbox servers in separate domains. (DAG is domain bound.)

*Design Decision Point*

In a previous step, it was decided that the database distribution model was going to be
active/passive. This model doesn't require multiple DAGs to be deployed. This example isn't likely
to require more than 16 Mailboxes servers for 10,000 mailboxes, and there is no requirement for
separate DAG-level administrative boundaries. Therefore, a single DAG will be used in this
design.

Step 8: Determine Mailbox server resiliency strategy

Exchange 2010 has been re-engineered for mailbox resiliency. Automatic failover protection is
now provided at the mailbox database level instead of at the server level. You can strategically
distribute active and passive database copies to Mailbox servers within a DAG. Determining how
many database copies you plan to activate on a per-server basis is a key aspect to Exchange
2010 capacity planning. There are different database distribution models that you can deploy, but
generally we recommend one of the following:

Design for all copies activated In this model, the Mailbox server role is sized to
accommodate the activation of all database copies on the server. For example, a Mailbox
server may host four database copies. During normal operating conditions, the server may
have two active database copies and two passive database copies. During a failure or
maintenance event, all four database copies would become active on the Mailbox server.

14

This solution is usually deployed in pairs. For example, if deploying four servers, the first pair
is servers MBX1 and MBX2, and the second pair is servers MBX3 and MBX4. In addition,
when designing for this model, you will size each Mailbox server for no more than 40 percent
of available resources during normal operating conditions. In a site resilient deployment with
three database copies and six servers, this model can be deployed in sets of three servers,
with the third server residing in the secondary datacenter. This model provides a three-server
building block for solutions using an active/passive site resiliency model.
This model can be used in the following scenarios:

Active/Passive multisite configuration where failure domains (for example, racks, blade
enclosures, and storage arrays) require easy isolation of database copies in the primary
datacenter

Active/Passive multisite configuration where anticipated growth may warrant easy

addition of logical units of scale

Configurations that aren't required to survive the simultaneous loss of any two Mailbox
servers in the DAG

This model requires servers to be deployed in pairs for single site deployments and sets of
three for multisite deployments. The following table illustrates a sample database layout for
this model.
Design for all copies activated

In the preceding table, the following applies:

C1 = active copy (activation preference value of 1) during normal operations

C2 = passive copy (activation preference value of 2) during normal operations

C3 = passive copy (activation preference value of 3) during site failure event

Design for targeted failure scenarios In this model, the Mailbox server role is designed to
accommodate the activation of a subset of the database copies on the server. The number of

15

database copies in the subset will depend on the specific failure scenario that you're
designing for. The main goal of this design is to evenly distribute active database load across
the remaining Mailbox servers in the DAG.
This model should be used in the following scenarios:

All single site configurations with three or more database copies

Configurations required to survive the simultaneous loss of any two Mailbox servers in
the DAG

The DAG design for this model requires between 3 and 16 Mailbox servers. The following
table illustrates a sample database layout for this model.
Design for targeted failure scenarios

In the preceding table, the following applies:

C1 = active copy (activation preference value of 1) during normal operations

C2 = passive copy (activation preference value of 2) during normal operations

C3 = passive copy (activation preference value of 3) during normal operations

*Design Decision Point*

In a previous step, it was decided to deploy an Active/Passive database distribution model with
two high availability database copies in the primary datacenter and one high availability copy in
the secondary datacenter. Because the two high availability copies in the primary datacenter are
usually deployed in separate hardware failure domains, this model usually results in a Mailbox
server resiliency strategy that designs for all copies being activated.

16

Step 9: Determine number of Mailbox servers and DAGs

The number of Mailbox servers required to support the workload and the minimum number of
Mailbox servers required to support the DAG design may be different. In this step, a preliminary
result is obtained. The final number of Mailbox servers will be determined in a later step.
*Design Decision Point*
This example uses three high availability database copies. To support three copies, a minimum of
three Mailbox servers in the DAG is required. In an active/passive configuration, two of the
servers will reside in the primary datacenter, and the third server will reside in the secondary
datacenter. In this model, the number of servers in the DAG should be deployed in multiples of
three. The following table outlines the possible configurations.
Number of Mailbox servers and DAGs
Primary datacenter

Secondary datacenter

Total Mailbox server count

12

Return to top

Estimate Mailbox Storage Capacity Requirements

Many factors influence the storage capacity requirements for the Mailbox server role. For
additional information, we recommend that you review Understanding Mailbox Database and Log
Capacity Factors.
The following steps outline how to calculate mailbox capacity requirements. These requirements
will then be used to make decisions about which storage solution options meet the capacity
requirements. A later section covers additional calculations required to properly design the
storage layout on the chosen storage platform.
Microsoft has created a Mailbox Server Role Requirements Calculator that will do most of this
work for you. To download the calculator, see E2010 Mailbox Server Role Requirements
Calculator. For additional information about using the calculator, see Exchange 2010 Mailbox
Server Role Requirements Calculator.

Step 1: Calculate mailbox size on disk

Before attempting to determine what your total storage requirements are, you should know what
the mailbox size on disk will be. A full mailbox with a 1-GB quota requires more than 1 GB of disk
space because you have to account for the prohibit send/receive limit, the number of messages
the user sends or receives per day, the Deleted Items folder retention window (with or without
calendar version logging and single item recovery enabled), and the average database daily

17

variations per mailbox. The Mailbox Server Role Requirements Calculator does these
calculations for you. You can also use the following information to do the calculations manually.
The following calculations are used to determine the mailbox size on disk for the three mailbox
tiers in this solution:

Tier 1 (512 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Whitespace = 100 messages per day 75 1024 MB = 7.3 MB

Dumpster = (100 messages per day 75 1024 MB 14 days) + (512 MB 0.012) +

(512 MB 0.058) = 138 MB

Mailbox size on disk = mailbox limit + whitespace + dumpster

= 512 MB + 7.3 MB + 138 MB
= 657 MB

Tier 2 (1024 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Whitespace = 100 messages per day 75 1024 MB = 7.3 MB

Dumpster = (100 messages per day 75 1024 MB 14 days) + (1024 MB 0.012) +

(1024 MB 0.058) = 174 MB

Mailbox size on disk = mailbox limit + whitespace + dumpster

= 1024 MB + 7.3 MB + 174 MB
= 1205 MB

Tier 3 (4096 MB mailbox quota, 150 messages per day message profile, 75 KB average
message size)

Whitespace = 150 messages per day 75 1024 MB = 11 MB

Dumpster = (150 messages per day 75 1024 MB 14 days) + (4096 MB 0.012) +

(4096 MB 0.058) = 441 MB

Mailbox size on disk = mailbox limit + whitespace + dumpster

= 4096 MB + 11 MB + 441 MB
= 4548 MB

Average size on disk = [(657 7650) + (1205 900) + (4548 450)] 9000
= 907 MB

Step 2: Calculate database storage capacity requirements

In this step, the high level storage capacity required for all mailbox databases is determined. The
calculated capacity includes database size, catalog index size, and 20 percent free space.
To determine the storage capacity required for all databases, use the following formulas:

Tier 1 (512 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

18

Database size = (number of mailboxes mailbox size on disk database overhead

growth factor) (20% data overhead)
= (7650 657 1) 1.2
= 6031260 MB
= 5890 GB

Database index size = 10% of database size

= 589 GB

Total database capacity = (database size + index size) 0.80 to add 20% volume free
space
= (5890 + 589) 0.8
= 8099 GB

Tier 2 (1024 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Database size= (number of mailboxes mailbox size on disk database overhead

growth factor) x (20% data overhead)
= (900 1205 1) x 1.2
= 1301400 MB
=1271 GB

Database index size = 10% of database size

= 127 GB

Total database capacity = (database size + index size) 0.80 to add 20% volume free
space
= (1271 + 127) 0.8
= 1747 GB

Tier 3 (4096 MB mailbox quota, 150 messages per day message profile, 75 KB average
message size)

Database size = (number of mailboxes mailbox size on disk database overhead

growth factor) x (20% data overhead)
= (450 4548 1) x 1.2
= 2455920 MB
= 2400 GB

Database index size = 10% of database size

= 240 GB

Total database capacity = (database size + index size) 0.80 to add 20% volume free
space
= (2400+ 240) 0.8
= 3301 GB

Total database capacity (all tiers) = 8099 + 1747 + 3301

19

= 13147 GB
= 12.3 terabytes

Step 3: Calculate transaction log storage capacity requirements

To ensure that the Mailbox server doesn't sustain any outages as a result of space allocation
issues, the transaction logs also need to be sized to accommodate all of the logs that will be
generated during the backup set. Provided that this architecture is leveraging the mailbox
resiliency and single item recovery features as the backup architecture, the log capacity should
allocate for three times the daily log generation rate in the event that a failed copy isn't repaired
for three days. (Any failed copy prevents log truncation from occurring.) In the event that the
server isn't back online within three days, you would want to temporarily remove the copy to allow
truncation to occur.
To determine the storage capacity required for all transaction logs, use the following formulas:

Tier 1 (512 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Log files size = (log file size number of logs per mailbox per day number of days
required to replace failed infrastructure number of mailbox users) + (1% mailbox move
overhead)
= (1 MB 20 3 7650) + (7650 0.01 512)
= 498168 MB
= 487 GB

Total log capacity = log files size 0.80 to add 20% volume free space
= (487) 0.80
= 608 GB

Tier 2 (1024 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Log files size = (log file size number of logs per mailbox per day number of days
required to replace failed infrastructure number of mailbox users) + (1% mailbox move
overhead)
= (1 MB 20 3 900) + (900 0.01 1024)
= 63216 MB
= 62 GB

Total log capacity = log files size 0.80 to add 20% volume free space
= (62) 0.80
= 77 GB

Tier 3 (4096 MB mailbox quota, 150 messages per day message profile, 75 KB average
message size)

20

Log files size = (log file size number of logs per mailbox per day number of days
required to replace failed infrastructure number of mailbox users) + (1% mailbox move
overhead) = (1 MB 30 3 450) + (450 0.01 4096)
= 58932 MB
= 58 GB

Total log capacity = log files size 0.80 to add 20% volume free space
= (58) 0.80
= 72 GB

Total log capacity (all tiers) = 608 + 77 + 72

= 757 GB

Step 4: Determine total storage capacity requirements

The following table summarizes the high level storage capacity requirements for this solution. In a
later step, you will use this information to make decisions about which storage solution to deploy.
You will then take a closer look at specific storage requirements in later steps.
Summary of storage capacity requirements
Disk space requirements

Value

Average mailbox size on disk (MB)

907

Database space required (GB)

13147

Log space required (GB)

757

Total space required (GB)

13904

Total space required for three database copies

(GB)

41712

Total space required for three database copies

(terabytes)

41

Return to top

Estimate Mailbox I/O Requirements

When designing an Exchange environment, you need an understanding of database and log
performance factors. We recommend that you review Understanding Database and Log
Performance Factors.

Calculate total mailbox I/O requirements

Because it's one of the key transactional I/O metrics needed for adequately sizing storage, you
should understand the amount of database I/O per second (IOPS) consumed by each mailbox
user. Pure sequential I/O operations aren't factored in the IOPS per Mailbox server calculation

21

because storage subsystems can handle sequential I/O much more efficiently than random I/O.
These operations include background database maintenance, log transactional I/O, and log
replication I/O. In this step, you calculate the total IOPS required to support all mailbox users,
using the following:
Note:
To determine the IOPS profile for a different message profile, see the table "Database
cache and estimated IOPS per mailbox based on message activity" in Understanding
Database and Log Performance Factors.
Total required IOPS = IOPS per mailbox user number of mailboxes I/O overhead factor
= 0.15 450 1.2
= 81
Total required IOPS (all tiers) = 1107
Average IOPS per mailbox = 1107 9000 = 0.123
The high level storage IOPS requirements are approximately 1107. When choosing a storage
solution, ensure that the solution meets this requirement.
Return to top

Determine Storage Type

Exchange 2010 includes improvements in performance, reliability, and high availability that
enable organizations to run Exchange on a wide range of storage options.
When examining the storage options available, being able to balance the performance, capacity,
manageability, and cost requirements is essential to achieving a successful storage solution for
Exchange.
For more information about choosing a storage solution for Exchange 2010, see Mailbox Server
Storage Design.

Determine whether you prefer an internal or external storage solution

A number of server models on the market today support from 8 through 16 internal disks. These
servers are a fit for some Exchange deployments and provide a solid solution at a low price point.
If your storage capacity and I/O requirements are met with internal storage and you don't have a
specific requirement to use external storage, you should consider using server models with an
internal disk for Exchange deployments. If your storage and I/O requirements are higher or your
organization has an existing investment in SANs, you should examine larger external directattached storage (DAS) or SAN solutions.
*Design Decision Point*
In this example, the external storage solution is selected.
Return to top

22

Choose Storage Solution

Use the following steps to choose a storage solution.

Step 1: Identify preferred storage vendor

In this solution, the preferred storage vendor is Dell.
Dell, Inc. is a leading IT infrastructure and services company with a broad portfolio of servers,
storage, networking products, and comprehensive service offerings. Dell also provides testing,
best practices, and architecture guidance specifically for Exchange 2010 and other Microsoftbased solutions in the unified communications and collaboration stack such as Microsoft Office
SharePoint Server and Office Communications Server.
Dell offers a wide variety of storage solutions from Dell EqualLogic, Dell PowerVault, and
Dell/EMC. Dell storage technologies help you minimize cost and complexity, increase
performance and reliability, simplify storage management, and plan for future growth.

Step 2: Review available options from preferred vendor

There are a number of storage options that would be a good fit for this solution. The following
options were considered:
Option 1:Dell EqualLogic PS 6000 Series iSCSI SAN Array
The Dell EqualLogic PS Series is fundamentally changing the way enterprises think about
purchasing and managing storage. Built on breakthrough virtualized peer storage architecture,
the EqualLogic PS Series simplifies the deployment and administration of consolidated storage
environments. Its all-inclusive, intelligent feature set streamlines purchasing and delivers rapid
SAN deployment, easy storage management, comprehensive data protection, enterprise-class
performance and reliability, and seamless pay-as-you grow expansion. The PS6000 is a 3u
chassis that contains sixteen 3.5 inch hard disk drives with two iSCSI controllers and four 1 GBEthernet ports per controller. Up to 16 arrays can be included in a single managed unit known as
a group.
Option 2:Dell EqualLogic PS 6500 Series iSCSI SAN Array
The Dell EqualLogic PS Series 6500 arrays also provide the same ease of use and intelligence
features. However, this array was built with maximum density in mind. This 4u chassis holds up to
48 3.5 inch hard disk drives, making it incredibly space efficient. It also contains four 1 GBEthernet ports per controller. The PS6500 can be mixed with other PS series arrays in the same
group.
Dell EqualLogic PS Series 6500 array
Components

Dell EqualLogic PS6000E, X,

Dell EqualLogic PS6500E and X

XV, and XVS

Storage controllers:

Dual controllers with a total of

4 GB-battery-backed memory.
Battery-backed memory
provides up to 72 hours of

Dual controllers with a total of

4 GB-battery-backed memory.
Battery-backed memory
provides up to 72 hours of

23

Components

Dell EqualLogic PS6000E, X,

Dell EqualLogic PS6500E and X

XV, and XVS

data protection.

data protection.

Hard disk drives:

16x SATA, SAS, or SSD.

48x SATA or SAS.

Volumes

Up to 1024.

Up to 1024.

RAID support

RAID-5, RAID-6, RAID-10, and RAID-5, RAID-6, RAID-10, and

RAID-50.
RAID-50.

Network interfaces

4 copper per controller.

4 copper per controller.

Reliability

Redundant, hot-swappable
controllers, power supplies,
cooling fans, and disks.

Redundant, hot-swappable
controllers, power supplies,
cooling fans, and disks.

Individual disk drive slot power

control.

Individual disk drive slot power

control.

Option 3: Dell PowerVault MD3200i iSCSI SAN Array

The PowerVault MD3200i is a high performance iSCSI SAN designed to deliver storage
consolidation and data management capabilities in an easy to use, cost effective solution. Shared
storage is required to enable VM mobility, which is the key benefit of a virtual environment. The
PowerVault MD3000i is a networked shared storage solution, providing the high availability,
expandability, and ease of management desired in virtual environments. The PowerVault
MD3000i leverages existing IP networks and offers small and medium businesses an easy to use
iSCSI SANs without the need for extensive training or new expensive infrastructures.

Step 3: Select an array

The listed arrays were the PS 6000E and the PS 6500E. PS6500E enclosures can accommodate
a total of 46 + 2 (hot spare) drives and are the most dense storage solution offered. Therefore,
the cost per gigabyte of deploying a PS6500E solution would be lower than that for a PS6000E
solution. The PS6500E array is also an intelligent solution that offers SAN configuration and
monitoring features, auto-build of RAID sets, network sensing mechanisms, and continuous
health monitoring. The MD3200i is a less expensive solution but lacks some of the management
and deployment features in the PS series arrays.
In this example, the PS6500 series is selected because this storage enclosure offers a
comprehensive datacenter consolidation solution spread across multiple sites as opposed to a
Server Message Block (SMB) or branch-office storage need.

Step 4: Select a disk type

The Exchange 2010 solution is optimized to use more sequential I/O and less random I/O with
larger mailboxes. This implies less disk intensive activity, even during peak usage hours when
compared to Exchange 2007. Therefore, high capacity SATA disks are used to save cost.

24

For a list of supported disk types, see "Physical Disk Types" in Understanding Storage
Configuration.
To help determine which disk type to choose, see "Factors to Consider When Choosing Disk
Types" in Understanding Storage Configuration.
Return to top

Determine Number of EqualLogic Arrays Required

In a previous step, it was determined to deploy three copies of each database. One of the three
copies will be located in the secondary datacenter. Therefore, to meet the site resiliency
requirements, a minimum of one PS6500E in the primary datacenter and one PS6500E in the
secondary datacenter is needed.
Consider IOPS requirements. In a previous step it was determined that 1,107 IOPS were required
to support the 9,000 mailboxes. For a RAID-10 configuration of SATA disks, this IOPS
requirement can be met in a single PS 6500 array. In a failure event, a single PS6500E would
have to support 100 percent of the IOPS requirement. Therefore, to meet the IOPS requirements,
a minimum of one PS6500E in the primary datacenter and one PS6500E in the secondary
datacenter is needed.
Consider storage capacity requirements. In a previous step, it was determined that approximately
26 terabytes were required to support two copies of each database in the primary datacenter and
approximately 13 terabytes to support one copy of each database in the secondary datacenter. A
single PS6500E configured with two spares and the remaining 46 disks in a RAID-10 disk group
provides approximately 20 terabytes. Therefore, two PS6500E's in the primary datacenter and
one PS6500E in the secondary datacenter are required to support the capacity requirements.
Three PS6500E's will be deployed to support the capacity requirements of this solution.
Return to top

Estimate Mailbox Memory Requirements

Sizing memory correctly is an important step in designing a healthy Exchange environment. We
recommend that you review Understanding Memory Configurations and Exchange Performance
and Understanding the Mailbox Database Cache.

Calculate required database cache

The Extensible Storage Engine (ESE) uses database cache to reduce I/O operations. In general,
the more database cache available, the less I/O generated on an Exchange 2010 Mailbox server.
However, there's a point where adding additional database cache no longer results in a
significant reduction in IOPS. Therefore, adding large amounts of physical memory to your
Exchange server without determining the optimal amount of database cache required may result
in higher costs with minimal performance benefit.
The IOPS estimates that you completed in a previous step assume a minimum amount of
database cache per mailbox. These minimum amounts are summarized in the table "Estimated

25

IOPS per mailbox based on message activity and mailbox database cache" in Understanding the
Mailbox Database Cache.
The following table outlines the database cache per user for various message profiles.
Database cache per user
Messages sent or received per mailbox per day

Database cache per user (MB)

(about 75 KB average message size)

50

3 MB

100

6 MB

150

9 MB

200

12 MB

In this step, you determine high level memory requirements for the entire environment. In a later
step, you use this result to determine the amount of physical memory needed for each Mailbox
server. Use the following information:

Tier 1 (512 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Database cache = profile specific database cache number of mailbox users

= 6 MB 7650
= 45900 MB
= 45 GB

Tier 2 (1024 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Database cache = profile specific database cache number of mailbox users

= 6 MB 900
= 5400 MB
= 6 GB

Tier 3 (4096 MB mailbox quota, 150 messages per day message profile, 75 KB average
message size)

Database cache = profile specific database cache number of mailbox users

= 9 MB 450
= 4050 MB
= 4 GB

Total database cache (all tiers) = 55 GB

Average per active mailbox = 55 GB 9000 1024 = 6.2 MB
The total database cache requirements for the environment are 55 GB or 6.2 MB per mailbox
user.

26

Return to top

Estimate Mailbox CPU Requirements

Mailbox server capacity planning has changed significantly from previous versions of Exchange
due to the new mailbox database resiliency model provided in Exchange 2010. For additional
information, see Mailbox Server Processor Capacity Planning.
In the following steps, you calculate the high level megacycle requirements for active and passive
database copies. These requirements will be used in a later step to determine the number of
Mailbox servers needed to support the workload. Note that the number of Mailbox servers
required also depends on the Mailbox server resiliency model and database copy layout.
Using megacycle requirements to determine the number of mailbox users that an Exchange
Mailbox server can support isn't an exact science. A number of factors can result in unexpected
megacycle results in test and production environments. Megacycles should only be used to
approximate the number of mailbox users that an Exchange Mailbox server can support. It's
always better to be conservative rather than aggressive during the capacity planning portion of
the design process.
The following calculations are based on published megacycle estimates as summarized in the
following table.
Megacycle estimates
Messages sent or

Megacycles per

Megacycles per

Megacycles per

received per mailbox

mailbox for active

mailbox for remote

mailbox for local

per day

mailbox database

passive mailbox

passive mailbox

database

50

0.1

0.15

100

0.2

0.3

150

0.3

0.45

200

0.4

0.6

Step 1: Calculate active mailbox CPU requirements

In this step, you calculate the megacycles required to support the active database copies, using
the following:

Tier 1 (512 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Active mailbox megacycles required = profile specific megacycles number of mailbox

users
= 2 7650
= 15300

27

Tier 2 (1024 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Active mailbox megacycles required = profile specific megacycles number of mailbox

users
= 2 900
= 1800

Tier 3 (4096 MB mailbox quota, 150 messages per day message profile, 75 KB average
message size)

Active mailbox megacycles required = profile specific megacycles number of mailbox

users
= 3 450
= 1350

Total active mailbox megacycles required (all tiers) = 18450 megacycles

Step 2: Calculate active mailbox remote database copy CPU requirements

In a design with three copies of each database, there is processor overhead associated with
shipping logs required to maintain database copies on the remote servers. This overhead is
typically 10 percent of the active mailbox megacycles for each remote copy being serviced.
Calculate the requirements, using the following:

Tier 1 (512 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Remote copy megacycles required = profile specific megacycles number of mailbox

users number of remote copies
= (0.2) (7650) 2
= 3060

Tier 2 (1024 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Remote copy megacycles required = profile specific megacycles number of mailbox

users number of remote copies
= (0.2) (900) 2
= 360

Tier 3 (4096 MB mailbox quota, 150 messages per day message profile, 75 KB average
message size)

Remote copy megacycles required = profile specific megacycles number of mailbox

users number of remote copies
= (0.3) (450) 2
= 270

Total remote copy megacycles required (all tiers) = 3690

28

Step 3: Calculate local passive mailbox CPU requirements

In a design with three copies of each database, there is processor overhead associated with
maintaining the local passive copies of each database. In this step, the high level megacycles
required to support local passive database copies will be calculated. These numbers will be
refined in a later step so that they match the server resiliency strategy and database copy layout.
Calculate the requirements, using the following:

Tier 1 (512 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Passive mailbox megacycles required = profile specific megacycles number of mailbox

users number of passive copies
= 0.3 7650 2
= 4590

Tier 2 (1024 MB mailbox quota, 100 messages per day message profile, 75 KB average
message size)

Passive mailbox megacycles required = profile specific megacycles number of mailbox

users number of passive copies
= 0.3 900 2
= 540

Tier 3 (4096 MB mailbox quota, 150 messages per day message profile, 75 KB average
message size)

Passive mailbox megacycles required = profile specific megacycles number of mailbox

users number of passive copies
= 0.45 450 2
= 405

Total passive mailbox megacycles required (all tiers) = 5535

Step 4: Calculate total CPU requirements

Calculate the total requirements, using the following:
Total megacycles required = active mailbox + remote passive copies + local passive copies
= 18450 + 3690 + 5535
= 27676
Total megacycles per mailbox = 3.08
Return to top

Determine Whether Server Virtualization Will Be Used

Several factors are important when considering server virtualization for Exchange. For more
information about supported configurations for virtualization, see Exchange 2010 System
Requirements.
The main reasons customers use virtualization with Exchange are as follows:

29

If you expect server capacity to be underutilized and anticipate better utilization, you may
purchase fewer servers as a result of virtualization.

You may want to use Windows Network Load Balancing when deploying Client Access, Hub
Transport, and Mailbox server roles on the same physical server.

If your organization is using virtualization in all server infrastructure, you may want to use
virtualization with Exchange, to be in alignment with corporate standard policy.

*Design Decision Point*

In this solution, deploying additional physical hardware for Client Access servers and Hub
Transport servers isn't wanted. Active/passive site resiliency design would require several
Mailbox servers to support the DAG design and database copy layout, which may result in
unused capacity on the Mailbox servers. Virtualization will be used to better utilize capacity
across server roles.
Return to top

Determine Whether Client Access and Hub Transport Server

Roles Will Be Deployed in Separate Virtual Machines
When using virtualization for the Client Access and Hub Transport server roles, you may consider
deploying both roles on the same VM. This approach reduces the number of VMs to manage, the
number of server operating systems to update, and the number of Windows and Exchange
licenses you need to purchase. Another benefit to combining the Client Access and Hub
Transport server roles is to simplify the design process. When deploying roles in isolation, we
recommend that you deploy one Hub Transport server logical processor for every four Mailbox
server logical processors, and that you deploy three Client Access server logical processors for
every four Mailbox server logical processors. This can be confusing, especially when you have to
provide sufficient Client Access and Hub Transport servers during multiple VM or physical server
failures or maintenance scenarios. When deploying Client Access, Hub Transport, and Mailbox
servers on like physical servers or like VMs, you can deploy one server with the Client Access
and Hub Transport server roles for every one Mailbox server in the site.
*Design Decision Point*
In this solution, co-locating the Hub Transport and Client Access server roles in the same VM is
wanted. The Mailbox server role is deployed separately in a second VM. This will reduce the
number of VMs and operating systems to manage as well as simplify planning for server
resiliency.
Return to top

Determine Server Model for Hyper-V Root Server

Step 1: Identify preferred server vendor
In this solution, the preferred server vendor is Dell.

30

The Dell eleventh generation PowerEdge servers offer industry leading performance and
efficiency. Innovations include increased memory capacity and faster I/O rates, which help deliver
the performance required by today's most demanding applications.

Step 2: Review available options from preferred vendor

Dell's server portfolio includes several models that were considered for this implementation.
Option 1: Dell PowerEdge M610 Blade Server
The decision to use iSCSI attached storage provides the potential for taking advantage of Dell
blades, based on the M1000e chassis. The M610 combines two sockets and twelve DIMMs in a
half-height blade for a dense and power efficient server.
Dell PowerEdge M1000e blade chassis
Components

Description

Chassis\enclosure

Form factor: 10U modular enclosure holds up

to sixteen half-height blade servers
44.0 cm (17.3") height 44.7 cm (17.6") width
75.4 cm (29.7") depth
Weight:

Empty chassis98 pounds

Chassis with all rear modules (IOMs,

PSUs, CMCs, KVM)176lbs

Max fully loaded with blades and rear

modules394lbs

Power supplies

3 (non-redundant) or 6 (redundant) 2,360 watt

hot-plug power supplies

Cooling fans

M1000e chassis comes standard with 9 hotpluggable, redundant fan modules

Input device

Front control panel with interactive graphical

LCD:

Supports initial configuration wizard

Local server blade, enclosure, and module

information and troubleshooting

Two USB keyboard/mouse connections and

one video connection (requires the optional
Avocent iKVM switch to enable these ports) for
local front crash cart console connections that
can be switched between blades
Enclosure I/O modules

Up to six total I/O modules for three fully

redundant fabrics, featuring Ethernet FlexIO

31

Components

Description

technology providing on-demand stacking and

uplink scalability. Dell FlexIO technology
delivers a level of I/O flexibility, bandwidth,
investment protection, and capabilities
unrivaled in the blade server market.
FlexIO technologies include:

Management

Completely passive, highly available

midplane that can deliver greater than
5 terabytes per second (TBps) of total I/O
bandwidth

Support for up to two ports of up to

40 gigabits per second (Gbps) from each
I/O mezzanine card on the blade server

1 (standard) or optional second (redundant)

Chassis Management Controller (CMC)
Optional integrated Avocent keyboard, video
and mouse (iKVM) switch
Dell OpenManage systems management

External storage options

Dell EqualLogic PS series, Dell/EMC AX series,

Dell/EMC CX series, Dell/EMC NS series, Dell
PowerVault MD series, Dell PowerVault NX
series

Dell PowerEdge M610 server

Components

Description

Processors (x2)

Latest quad-core or six-core Intel Xeon

processors 5500 and 5600 series

Form factor

Blade/modular half-height slot in an M1000e

blade chassis

Memory

12 DIMM slots
1 GB/2 GB/4 GB/8 GB/16 GB ECC DDR3
Support for up to 192 GB using 12 16 GB
DIMMs

Drives

Internal hot-swappable drives:

2.5" SAS (10,000 rpm): 36 GB, 73 GB,

146 GB, 300 GB, 600 GB

32

Components

Description

2.5" SAS (15,000 rpm): 36 GB, 73 GB

146 GB

Solid-state drives (SSD):

I/O slots

25 GB, 50 GB, 100 GB, 150 GB

Maximum internal storage:

Up to 1.2 terabyte via 4 300 GB SAS

hard disk drives

For external storage options, see previous

M1000e blade chassis information

For details, see previous M1000e blade chassis

information

Option 2: Dell PowerEdge M710 blade server

The M710 provides two sockets in a blade form factor but extends the number of DIMMs to
eighteen, greatly expanding memory capacity. However, the M710 also is a full height blade. The
extra RAM can make the R710 an attractive virtualization server.
Dell PowerEdge T710 server
Components

Description

Processors (x2)

Latest quad-core or six-core Intel Xeon

processors 5500 and 5600 series

Form factor

Blade/modularfull-height slot in an M1000e

blade chassis

Memory

18 DIMM slots
1 GB/2 GB/4 GB/8 GB/16 GB ECC DDR3
Support for up to 192 GB using 12 16 GB
DIMMs

Drives

Internal hot-swappable drives:

2.5" SAS (10,000 rpm): 36 GB, 73 GB,

146 GB, 300 GB, 600 GB

2.5" SAS (15,000 rpm): 36 GB, 73 GB

146 GB

SSD:

25 GB, 50 GB, 100 GB, 150 GB

Maximum Internal Storage:

Up to 1.2 terabyte via 4 300 GB SAS

33

Components

Description

Hard Drives

I/O slots

For external storage options, see previous

M1000e blade chassis information

For details, see previous M1000e blade chassis

information

Option 3: Dell PowerEdge R710 rack mounted server

Another choice for this implementation could be the Dell PowerEdge R710. This Intel-based
platform is a 2u rack mounted server containing two sockets, eighteen DIMM slots, and the option
of either eight 2.5", or six 3.5" internal hard disk drives. Although limited in internal disk capacity
compared to the other server models presented, it scales beyond the R510 in memory (eighteen
DIMMS compared to eight) and provides more I/O options. Storage capabilities may be expanded
by using Dell PowerVault MD1200 or MD1220 direct attached storage arrays. The MD1200
provides twelve 3.5" hard disk drives in a 2u rack mounted form factor, while the MD1220
provides twenty-five 2.5" hard disk drives in the same 2u rack mounted form factor. These
6 Gbps SAS connected arrays can be daisy chained, up to four arrays per RAID controller, and
also support redundant connections from the server. This storage option satisfies requirements
for lower cost storage and simplicity while giving each node the ability to scale in the number of
supported mailboxes.
Dell PowerEdge R710 server
Components

Description

Processors (x2)

Latest quad-core or six-core Intel Xeon

processors 5500 and 5600 series

Form factor

2U rack

Memory

Up to 192 GB (18 DIMM slots*):

1 GB/2 GB/4 GB/8 GB/16 GB DDR3,
800 megahertz (MHz), 1066 MHz, or 1333 MHz

Drives

Eight 2.5" hard disk drive option or six 3.5" hard

disk drive option with optional flex bay
expansion to support half-height TBU
Up to six 3.5" drives with optional flex bay or up
to eight 2.5" SAS or SATA drives with optional
flex bay
Peripheral bay options include slim optical drive
bay with choice of DVD-ROM, combo CDRW/DVD-ROM, or DVD + RW

I/O slots

2 PCIe x8 + 2 PCIe x4 G2 or 1 x16 + 2 x4 G2

34

Option 4: Dell PowerEdge R810 rack mounted server

The R810 is a two or four socket platform in a 2u form factor. It contains Dell patented FlexMem
bridge technology, which allows the server to take advantage of all thirty-two DIMM slots even
with only two processors installed. This enables the R810 to be a virtualization platform
providing great compute power in a dense package.
Dell PowerEdge R810 server
Components

Description

Processors (x4)

Up to Eight-Core Intel Xeon 7500 and 6500

series processors

Form factor

2U rack

Memory

Up to 512 GB (32 DIMM slots)

1 GB/2 GB/4 GB/8 GB/16 GB DDR3 1066 MHz

Drives

Hot-swap option available with up to six 2.5"

SAS or SATA drives, including SATA SSD

I/O slots

6 PCIe G2 slots:

Five x8 slot

One x4 slot

One storage x4 slot

Step 3: Select a server model

For this solution, Dell PowerEdge M610 blades is selected. To standardize on blades in the
datacenter to take advantage of the density and power efficiencies is desired. Although the M710
may be able to support more VMs per server than the M610, there is still more capacity to be
saved with the M610 in this deployment due to it being half-height versus the M710 full-height
form factor.
In previous steps, megacycles required to support the number of active mailbox users were
calculated. In the following steps, the number of available megacycles the selected server model
and processor can support will be determined so that the number of active mailboxes each server
can support can then be determined.

Step 4: Determine benchmark value for server and processor

Because the megacycle requirements are based on a baseline server and processor model, you
need to adjust the available megacycles for the server against the baseline. To do this,
independent performance benchmarks maintained by Standard Performance Evaluation
Corporation (SPEC) are used. SPEC is a non-profit corporation formed to establish, maintain,
and endorse a standardized set of relevant benchmarks that can be applied to the newest
generation of high-performance computers.

35

To help simplify the process of obtaining the benchmark value for your server and processor, we
recommend you use the Exchange Processor Query tool. This tool automates the manual steps
to determine your planned processor's SPECInt 2006 rate value. To run this tool, your computer
must be connected to the Internet. The tool uses your planned processor model as input, and
then runs a query against the Standard Performance Evaluation Corporation Web site returning
all test result data for that specific processor model. The tool also calculates an average SPECint
2006 rate value based on the number of processors planned to be used in each Mailbox server
Use the following calculations:

Processor and server platform = Intel X5550 2.6 gigahertz (GHz) in a Dell M610

SPECint_rate2006 value = 234

SPECint_rate2006 value per processor core = 234 8

= 29.25

Step 5: Calculate adjusted megacycles

In previous steps, you calculated the required megacycles for the entire environment based on
megacycle per mailbox estimates. Those estimates were measured on a baseline system (HP
DL380 G5 x5470 3.33 GHz, 8 cores) that has a SPECint_rate2006 value of 150 (for an 8 core
server), or 18.75 per core.
In this step, you need to adjust the available megacycles for the chosen server and processor
against the baseline processor so that the required megacycles can be used for capacity
planning.
To determine the megacycles of the Dell M610 Intel X5550 2.6 GHz platform, use the following
formula:

Adjusted megacycles per core = (new platform per core value) (hertz per core of baseline
platform) (baseline per core value)
= (29.25 3330) 18.75
= 5195

Adjusted megacycles per server = adjusted megacycles per core number of cores
= 5195 8
= 41558

Step 6: Adjust available megacycles for virtualization overhead

When deploying VMs on the root server, megacycles required to support the hypervisor and
virtualization stack must be accounted for. This overhead varies from server to server and under
different workloads. A conservative estimate of 10 percent of available megacycles will be used.
Use the following calculation:

Adjusted available megacycles = usable megacycles 0.90

= 41558 0.90
= 37403

So each server has a usable capacity for VMs of 37403 megacycles.

36

The usable capacity per logical processor is 4675 megacycles.

Return to top

Determine the CPU Capacity of the Virtual Machines

Now that we know the megacycles of the root server we can calculate the megacycles of each
VM. These values will be used to determine how many VMs are required and how many
mailboxes will be hosted by each VM.

Step 1: Calculate available megacycles per virtual machine

In this step, you determine how many megacycles are available for each VM deployed on the root
server. Because the server has eight logical processors, plan to deploy two VMs per server, each
with four virtual processors. Use the following calculation:

Available megacycles per VM = adjusted available megacycles per server number of VMs
= 37403 2
= 18701

Step 2: Determine the target available megacycles per virtual machine

Because the design assumptions state not to exceed 80 percent processor utilization, in this step,
you adjust the available megacycles to reflect the 80 percent target. Use the following calculation:
Because the design assumptions state not to exceed 70 percent processor utilization, in this step,
you adjust the available megacycles to reflect the 70 percent target. Use the following calculation:

Target available megacycles = available megacycles target max processor utilization

= 18701 0.70
= 13091

Return to top

Determine Number of Mailbox Server Virtual Machines Required

You can use the following steps to determine the number of Mailbox server VMs required.

Step 1: Determine the maximum number of mailboxes supported by the

MBX virtual machine
To determine the maximum number of mailboxes supported by the MBX VM, use the following
calculation:

Number of active mailboxes = available megacycles megacycles per mailbox

= 13091 3.08
= 4250

37

Step 2: Determine the minimum number of mailbox virtual machines

required in the primary site
To determine the minimum number of mailbox VMs required in the primary site, use the following
calculation:

Number of VMs required = total mailbox count in site active mailboxes per VM
= 9000 4250
= 2.2

Based on processor capacity, minimum of three Mailbox server VMs to support the anticipated
peak work load during normal operating conditions is required.

Step 3: Determine number of Mailbox server virtual machines required to

support the mailbox resiliency strategy
In the previous step, you determined that a minimum of three Mailbox server VMs to support the
target workload are needed. In an active/passive database distribution model, you need a
minimum of three Mailbox server VMs in the secondary datacenter to support the workload during
a site failure event. The DAG design will have nine Mailbox server VMs with six in the primary site
and three in the secondary site.
Datacenter vs. Mailbox server count
Primary datacenter

Secondary datacenter

Total Mailbox server count

12

Return to top

Determine Number of Mailboxes per Mailbox Server

You can use the following steps to determine the number of mailboxes per Mailbox server.

Step 1: Determine number of active mailboxes per server during normal

operation
To determine the number of active mailboxes per server during normal operation, use the
following calculation:

Number of active mailboxes per server = total mailbox count server count
= 9000 6
= 1500

38

Step 2: Determine number of active mailboxes per server worst case failure
event
To determine the number of active mailboxes per server worst case failure event, use the
following calculation:

Number of active mailboxes per server = total mailbox count server count
= 9000 3
= 3000

Return to top

Determine Memory Required Per Mailbox Server

You can use the following steps to determine the memory required per Mailbox server.

Step 1: Determine database cache requirements per server for the worst
case failure scenario
In a previous step, you determined that the database cache requirements for all mailboxes was
55 GB and the average cache required per active mailbox was 6.2 MB.
To design for the worst case failure scenario, you calculate based on active mailboxes residing
on three of six Mailbox servers. Use the following calculation:

Memory required for database cache = number of active mailboxes average cache per
mailbox
= 3000 6.2 MB
= 18600 MB
= 18.2 GB

Step 2: Determine total memory requirements per mailbox virtual machine

server for the worst case failure scenario
In this step, reference the following table to determine the recommended memory configuration.
Memory requirements
Server physical memory (RAM)

Database cache size (Mailbox role only)

24 GB

17.6 GB

32 GB

24.4 GB

48 GB

39.2 GB

The recommended memory configuration to support 18.2 GB of database cache for a mailbox
role server is 32 GB.
Return to top

39

Determine Number of Client Access and Hub Transport Server

Combo Virtual Machines Required
In a previous step, it was determined that nine Mailbox server VMs are required. We recommend
that you deploy one Client Access and Hub Transport server combo VM for every MBX VM.
Therefore, the design will have nine Client Access and Hub Transport server combo VMs.
Number of Client Access and Hub Transport server combo VMs required
Server role configuration

Recommended processor core ratio

Mailbox server role: Client Access and Hub

Transport combined server role

1:1

Determine Memory Required per Combined Client Access and

Hub Transport Virtual Machines
To determine the memory configuration for the combined Client Access and Hub Transport server
role VM, reference the following table.
Memory configurations for Exchange 2010 servers based on installed server roles
Exchange 2010 server role

Minimum supported

Recommended maximum

Hub Transport server role

4 GB

1 GB per core

Client Access server role

4 GB

2 GB per core

Client Access and Hub

4 GB
Transport combined server
role (Client Access and Hub
Transport server roles running
on the same physical server)

2 GB per core

Based on the preceding table, each combination Client Access and Hub Transport server VM
requires a minimum of 8 GB of memory.
Return to top

Determine Virtual Machine Distribution

When deciding which VMs to host on which root server, your main goal should be to eliminate
single points of failure. Don't locate both Client Access and Hub Transport server role VMs on the
same root server, and don't locate both Mailbox server role VMs on the same root server.

40

Virtual machine distribution (incorrect)

The correct distribution is one Client Access and Hub Transport server role VM on each of the
physical host servers and one Mailbox server role VM on each of the physical host servers. So in
this solution there will be nine Hyper-V root servers each supporting one Client Access and Hub
Transport server role VM and one Mailbox server role VM.
Virtual machine distribution (correct)

Return to top

Determine Memory Required per Root Server

To determine the memory required for each root server, use the following calculation:

Root server memory = Client Access and Hub Transport server role VM memory + Mailbox
server role VM memory
= 8 GB + 32 GB
= 40 GB

The Hyper-V root server will require a minimum of 40 GB.

Return to top

Determine Minimum Number of Databases Required

To determine the optimal number of Exchange databases to deploy, use the Exchange 2010
Mailbox Role Calculator. Enter the appropriate information on the input tab and select Yes for
Automatically Calculate Number of Unique Databases / DAG.

41

Database configuration

On the Role Requirements tab, the recommended number of databases appears.

Recommended number of databases

In this solution, a minimum of 12 databases will be used. The exact number of databases may be
adjusted in future steps to accommodate the database copy layout.
Return to top

Identify Failure Domains Impacting Database Copy Layout

Use the following steps to identify failure domains impacting database copy layout.

Step 1: Identify failure domains associated with storage

In a previous step, it was decided to deploy three Dell EqualLogic PS6500E arrays and to deploy
three copies of each database. To provide maximum protection for each of those database
copies, we recommend that no more than one copy of a single database be located on the same
physical array. In this scenario, each PS6500E represents a failure domain that will impact the
layout of database copies in the DAG.
Dell EqualLogic PS6500E arrays

Step 2: Identify failure domains associated with servers

In a previous step, it was determined that nine physical blade servers will be deployed. Six of
those servers will be deployed in the primary datacenter and three in the secondary datacenter.
Blades are associated with blade enclosures. So to support the site resiliency requirements, a
minimum of 2 blade enclosures are required.

42

Failure domains associated with servers

In the previous step, it was determined that PS6500E represents three failure domains. Consider
when all six blades in the first enclosure to the two PS6500Es in the primary datacenter are
connected. In the event that there is an issue impacting the enclosure, there are no other servers
in the primary datacenter and you're forced to conduct a manual site switchover to the secondary
datacenter. A better design is to deploy three blade enclosures, each with three of the nine server
blades. Pair the servers in the first enclosure with the first PS6500E, the servers in the second
enclosure with the second PS6500E, and the three servers in the secondary site with the
PS6500E in the secondary site. By aligning the server and storage failure domains, the database
copies are set in a manner that protects against issues with either the storage array or an entire
blade enclosure.
Failure domains associated with servers in two sites

Return to top

Design Database Copy Layout

Use the following steps to design database copy layout.

Step 1: Determine number of database copies per Mailbox server

In a previous step, it was determined that the minimum number of unique databases that should
be deployed is 12. In an active/passive configuration with three copies, we recommend that the
number of databases equal the total number of Mailbox servers in the primary site multiplied by
the number of Mailbox servers in a single failure domain and be greater than the minimum
number of recommended databases. Use the following calculation:

Unique database count = total number of Mailbox servers in primary datacenter number of
Mailbox servers in failure domain
=63

43

=18

Step 2: Determine database layout during normal operating conditions

Consider equally distributing the C1 database copies (or the copies with an activation preference
value of 1) to the servers in the primary datacenter. These are the copies that will be active during
normal operating conditions.
Database copy layout during normal operating conditions
DB

MBX1

DB1

C1

DB2

C1

DB3

C1

MBX2

DB4

C1

DB5

C1

DB6

C1

MBX3

DB7

C1

DB8

C1

DB9

C1

MBX4

DB10

C1

DB11

C1

DB12

C1

MBX5

DB13

C1

DB14

C1

DB15

C1

MBX6

DB16

C1

DB17

C1

DB18

C1

In the preceding table, the following applies:

C1 = active copy (activation preference value of 1) during normal operations

Next distribute the C2 database copies (or the copies with an activation preference value of 2) to
the servers in the second failure domain. During the distribution, you distribute the C2 copies
across as many servers in the alternate failure domain as possible to ensure that a single server
failure has a minimal impact on the servers in the alternate failure domain.

44

Database copy layout with C2 database copies distributed

DB

MBX1

DB1

C1

DB2

C1

DB3

C1

MBX2

MBX3

MBX4

MBX5

MBX6

C2
C2
C2

DB4

C1

DB5

C1

DB6

C1

C2
C2
C2

DB7

C1

DB8

C1

DB9

C1

C2
C2
C2

In the preceding table, the following applies:

C1 = active copy (activation preference value of 1) during normal operations

C2 = passive copy (activation preference value of 2) during normal operations

Consider the opposite configuration for the other failure domain. Again, you distribute the C2
copies across as many servers in the alternate failure domain as possible to ensure that a single
server failure has a minimal impact on the servers in the alternate failure domain.
Database copy layout with C2 database copies distributed in the opposite configuration
DB

MBX1

DB10

C2

DB11

MBX2

C2

C2

DB17
DB18

MBX6

C1
C1

C2

DB15
DB16

MBX5

C1
C2

DB14

MBX4

C1

DB12
DB13

MBX3

C1
C2

C1

C2

C1
C2

C1
C2

C1

In the preceding table, the following applies:

C1 = active copy (activation preference value of 1) during normal operations

45

C2 = passive copy (activation preference value of 2) during normal operations

Step 3: Determine database layout during server failure and maintenance

conditions
Before the secondary datacenter and distribute the C3 copies are considered, examine the
following server failure scenario. In the following example, if server MBX1 fails, the active
database copies will automatically move to servers MBX4, MBX5, and MBX6. Notice that each of
the three servers in the alternate failure domain are now running with four active databases and
the active databases are equally distributed across all three servers.
Database copy layout during server maintenance or failure

In the preceding table, the following applies:

46

C1 = active copy (activation preference value of 1) during normal operations

C2 = passive copy (activation preference value of 2) during normal operations

In a maintenance scenario, you could move the active mailbox databases from the servers in the
first failure domain (MBX1, MBX2, MBX3) to the servers in the second failure domain (MBX4,
MBX5, MBX6), complete maintenance activities, and then move the active database copies back
to the C1 copies on the servers in the first failure domain. You can conduct maintenance activities
on all servers in the primary datacenter in two passes.
Database copy layout during server maintenance

In the preceding table, the following applies:

C1 = active copy (activation preference value of 1) during normal operations

C2 = passive copy (activation preference value of 2) during normal operations

47

Step 4: Add database copies to secondary datacenter to support site

resiliency
The last step in the database copy layout is to add the C3 copies (or copies with an activation
preference value of 3) to the servers in the secondary datacenter to provide site resiliency. As
performed with the C2 copies, distribute the C3 copies across as many servers in the alternate
failure domain as possible to ensure that any issues impacting multiple Mailbox servers in the
primary datacenter has a minimal impact on the servers in the alternate failure domain in the
secondary datacenter. In a full site failure scenario, activate all C3 copies in the secondary
datacenter using Datacenter Activation Coordination (DAC) and the distribution of database
copies in relation to servers in the primary datacenter is less important.
Database copy layout to support site resiliency
DB

MBX1

DB1

C1

DB2

C1

DB3

C1

MBX2

MBX3

C1

DB6

C1

C1

DB9

C1
C2
C2

DB18

C3

C2

C3
C3

C2

C3
C2

C3
C3

C1
C2

C3

C1

C2

C3
C1

C2

DB15

DB17

C3

C1

DB12

C3

C2

DB8

C3

C1
C2

C2

C3

C1

C3
C1

C2

C1
C2

MBX9

C3

C2

C1

DB16

MBX8

C3

C2

DB7

DB14

MBX7

C2

DB5

DB13

MBX6

C2

C1

DB11

MBX5

C2

DB4

DB10

MBX4

C3
C3

C1

C3

In the preceding table, the following applies:

C1 = active copy (activation preference value of 1) during normal operations

48

C2 = passive copy (activation preference value of 2) during normal operations

C3 = remote passive copy (activation preference value of 3) during normal operations

Return to top

Determine Storage Design

A well designed storage solution is a critical aspect of a successful Exchange 2010 Mailbox
server role deployment. For more information, see Mailbox Server Storage Design.

Step 1: Summarize storage requirements

The following table summarizes the storage requirements that have been calculated or
determined in a previous design step.
Summary of disk space requirements
Disk space requirements

Value

Average mailbox size on disk (MB)

907

Database space required (GB)

13147

Log space required (GB)

757

Total space required (GB)

13904

Total space required for three database copies

(GB)

41712

Total space required for three database copies

(terabytes)

41

Step 2: Determine whether logs and databases will be co-located on the

same LUN
In previous Exchange releases, it was a recommended best practice to separate database files
and log files from the same mailbox database to different volumes backed by different physical
disks for recoverability purposes. This is still a recommended best practice for stand-alone
architectures and architectures using VSS-based backups. If you're using Exchange native data
protection and have deployed a minimum of three database copies, isolation of logs and
databases isn't necessary.
*Design Decision Point*
With the EqualLogic array, the RAID-10 set spans across all 46 disks. Because this architecture
doesn't offer spindle isolation, there is no reason to create separate LUNs for database and log
files, therefore subsequent design decisions will be based on a single LUN for each database and
log set.

49

Step 3: Determine number of LUNs required per array

In a previous step, it was identified that each primary Mailbox server would support three active
databases, three passive database copies, and three lagged database copies. Therefore there
will be a total of nine LUNs for each primary datacenter Mailbox server.
Number of LUNs required per array
Databases

LUNs per server

LUNs per array

Active databases

Passive databases

Lagged databases

Total LUNs

27

Step 4: Determine required LUN size

This step determines the size of the LUN required to support both the database and log capacity
requirements. Use the following calculations:

Database capacity = [(number of mailbox users average mailbox size on disk) + (20% data
overhead factor)] + (10% content indexing overhead)
= [(500 907) + (90700)] + 54420
= 598620 MB
= 585 GB

Log capacity = (log size number of logs per mailbox per day number of days required to
replace hardware number of mailbox users) + (mailbox move percent overhead)
= (1 MB 20.5 3 500) + (500 0.01 907 MB)
=35285 MB
=35 GB

LUN size = [(database capacity) + (log capacity)] +20% volume free space
= [(585) + (35)] .8
= 775 GB

The required LUN size is 775 GB.

Step 5: Calculate actual LUN size

In a previous step, it was determined that the EqualLogic PS6500E has a usable capacity of
20.8 terabytes or 21299 GB when using RAID-10 and having two spares configured. Each array
needs to have 27 LUNs.

21299 27 = 789 GB

The actual LUN size will be 789 GB, which will support the required LUN size of 775 GB.

50

Actual LUN size

Description

Value

Usable capacity

21299 GB

Number of LUNs required

27

Required LUN Size

775 GB

Actual LUN Size

789 GB

Step 6: Determine volume layout on PS6500Es

The following table illustrates how the database copies are positioned on the XIV Storage
Systems.
Volume layout on PS6500Es
Database

Array1

Database

Array2

Database

Array3

DB1

C1

DB1

C2

DB1

C3

DB2

C1

DB2

C2

DB2

C3

DB3

C1

DB3

C2

DB3

C3

DB4

C1

DB4

C2

DB4

C3

DB5

C1

DB5

C2

DB5

C3

DB6

C1

DB6

C2

DB6

C3

DB7

C1

DB7

C2

DB7

C3

DB8

C1

DB8

C2

DB8

C3

DB9

C1

DB9

C2

DB9

C3

DB10

C2

DB10

C1

DB10

C3

DB11

C2

DB11

C1

DB11

C3

DB12

C2

DB12

C1

DB12

C3

DB13

C2

DB13

C1

DB13

C3

DB14

C2

DB14

C1

DB14

C3

DB15

C2

DB15

C1

DB15

C3

DB16

C2

DB16

C1

DB16

C3

DB17

C2

DB17

C1

DB17

C3

DB18

C2

DB18

C1

DB18

C3

51

Return to top

Determine Placement of the File Share Witness

In Exchange 2010, the DAG uses a minimal set of components from Windows failover clustering.
One of those components is the quorum resource, which provides a means for arbitration when
determining cluster state and making membership decisions. It's critical that each DAG member
have a consistent view of how the DAGs underlying cluster is configured. The quorum acts as the
definitive repository for all configuration information relating to the cluster. The quorum is also
used as a tiebreaker to avoid split brain syndrome. Split brain syndrome is a condition that occurs
when DAG members can't communicate with each other but are available and running. Split brain
syndrome is prevented by always requiring a majority of the DAG members (and in the case of
DAGs with an even number of members, the DAG witness server) to be available and interacting
for the DAG to be operational.
A witness server is a server outside of a DAG that hosts the file share witness, which is used to
achieve and maintain quorum when the DAG has an even number of members. DAGs with an
odd number of members don't use a witness server. Upon creation of a DAG, the file share
witness is added by default to a Hub Transport server (that doesn't have the Mailbox server role
installed) in the same site as the first member of the DAG. If your Hub Transport server is running
in a VM that resides on the same root server as VMs running the Mailbox server role, we
recommend that you move the location of the file share witness to another highly available server.
You can move the file share witness to a domain controller, but because of security implications,
do this only as a last resort.
*Design Decision Point*
The file and print server is reasonably stable and is managed by the same administrator who
supports the Exchange servers, so it's a good choice for the location of the file share witness.
Return to top

Plan Namespaces
When you plan your Exchange 2010 organization, one of the most important decisions that you
must make is how to arrange your organization's external namespace. A namespace is a logical
structure usually represented by a domain name in Domain Name System (DNS). When you
define your namespace, you must consider the different locations of your clients and the servers
that house their mailboxes. In addition to the physical locations of clients, you must evaluate how
they connect to Exchange 2010. The answers to these questions will determine how many
namespaces you must have. Your namespaces will typically align with your DNS configuration.
We recommend that each Active Directory site in a region that has one or more Internet-facing
Client Access servers have a unique namespace. This is usually represented in DNS by an A
record, for example, mail.contoso.com or mail.europe.contoso.com.
For more information, see Understanding Client Access Server Namespaces.
There are a number of different ways to arrange your external namespaces, but usually your
requirements can be met with one of the following namespace models:

52

Consolidated datacenter model This model consists of a single physical site. All servers
are located within the site, and there is a single namespace, for example, mail.contoso.com.

Single namespace with proxy sites This model consists of multiple physical sites. Only
one site contains an Internet-facing Client Access server. The other sites aren't exposed to
the Internet. There is only one namespace for the sites in this model, for example,
mail.contoso.com.

Single namespace and multiple sites This model consists of multiple physical sites. Each
site can have an Internet-facing Client Access server. Alternatively, there may be only a
single site that contains Internet-facing Client Access servers. There is only one namespace
for the sites in this model, for example, mail.contoso.com.

Regional namespaces This model consists of multiple physical sites and multiple
namespaces. For example, a site located in New York City would have the namespace
mail.usa.contoso.com, a site located in Toronto would have the namespace
mail.canada.contoso.com, and a site located in London would have the namespace
mail.europe.contoso.com.

Multiple forests This model consists of multiple forests that have multiple namespaces. An
organization that uses this model could be made up of two partner companies, for example,
Contoso and Fabrikam. Namespaces might include mail.usa.contoso.com,
mail.europe.contoso.com, mail.asia.fabrikam.com, and mail.europe.fabrikam.com.

*Design Decision Point*

Because this solution is deploying an active/passive site resiliency model and doesn't have any
active mailbox users in the secondary site, the best option is the single namespace with multiple
sites model.
Return to top

Determine Client Access Server Array and Load Balancing

Strategy
In Exchange 2010, the RPC Client Access service and the Exchange Address Book service were
introduced on the Client Access server role to improve the mailbox users experience when the
active mailbox database copy is moved to another Mailbox server (for example, during mailbox
database failures and maintenance events). The connection endpoints for mailbox access from
Microsoft Outlook and other MAPI clients have been moved from the Mailbox server role to the
Client Access server role. Therefore, both internal and external Outlook connections must now be
load balanced across all Client Access servers in the site to achieve fault tolerance. To associate
the MAPI endpoint with a group of Client Access servers rather than a specific Client Access
server, you can define a Client Access server array. You can only configure one array per Active
Directory site, and an array can't span more than one Active Directory site. For more information,
see Understanding RPC Client Access and Understanding Load Balancing in Exchange.
*Design Decision Point*
In a previous step, it was determined that Client Access servers would be deployed in two
physical locations in two Active Directory sites. Therefore, you need to deploy two Client Access

53

server arrays. A single namespace will be load balanced across the Client Access servers in the
primary active Client Access server array using redundant hardware load balancers. In a site
failure, the namespace will be load balanced across the Client Access servers in the secondary
Client Access server array.
Return to top

Determine Hardware Load Balancing Solution

Use the following steps to determine a hardware load balancing solution.

Step 1: Identify preferred server vendor

The preferred vendor for application load balancing is F5.
The F5 comprehensive Application Ready infrastructure for Exchange Server allows
organizations to easily provide additional performance, security, and availability to ensure
maximum return on investment for Exchange deployments.

Step 2: Review available options from preferred vendor

F5 offers a suite of appliance-based networking technologies designed to optimize networks for
applications such as Exchange 2010:

BIG-IP Local Traffic Manager (LTM) BIG-IP LTM is designed to monitor and manage traffic
to Client Access, Hub Transport, Edge Transport, and Unified Messaging servers, while
ensuring that users are always sent to the best performing resource. Whether your users are
connecting via MAPI, Outlook Web App, ActiveSync, or Outlook Anywhere, BIG-IP LTM will
load balance the connections appropriately, allowing you to seamlessly scale to any size
deployment. BIG-IP LTM now offers several modules that also provide significant value in an
Exchange environment, which include:

Access Policy Manager (APM) Designed to secure access to Exchange resources, APM
can authenticate users before they attach to your Exchange Client Access servers,
providing a strong perimeter security.

Big-IP WebAccelerator Targeting customers with large Outlook Web App

constituencies, WebAccelerator can drive down bandwidth usage and server utilization
while accelerating content to end users.

WAN Optimization Module (WOM) Focused on network optimization for WANs, WOM
has proven capable in accelerating DAG replication by over five times between
datacenters.

BIG-IP Global Traffic Manager (GTM) BIG-IP GTM can provide wide area resiliency,
providing disaster recovery and load balancing for those with multiple datacenter Exchange
deployments.

BIG-IP Application Security Manager (ASM) A fully featured Layer 7 firewall, ASM thwarts
HTTP, XML, and SMTP based attacks. By combining a negative and positive security model,
ASM provides protection against all L7 attacks, both known and unknown.

For more information about these technologies, see F5 Solutions for Exchange Server.

54

Sizing the appropriate F5 hardware model for your Exchange 2010 deployment is an exercise
best done with the guidance of your local F5 team. F5 offers production hardware-based and
software-based BIG-IP platforms that range from supporting up to 200 megabits per second
(Mbps) all the way up to 80 Gbps. To learn more about the specifications for each of the F5 BIGIP LTM hardware platforms, see BIG-IP System Hardware Datasheet.
Option 1: BIG-IP 1600 series
The BIG-IP 1600 offers all the functionality of TMOS in a cost-effective, entry-level platform for
intelligent application delivery.
BIG-IP 1600 appliance-based networking technologies
Components

Value or description

Traffic throughput

1 Gbps

Hardware Secure Sockets Layer (SSL)

Included: 500 transactions per second

Maximum: 5,000 transactions per second
1 Gbps bulk encryption

Software compression

Included: 50 Mbps
Maximum: 1 Gbps

Processor

Dual core CPU

Memory

4 GB

Gigabit Ethernet CU ports

Gigabit fiber ports (small form-factor pluggable

transceiver)

2 optional LX, SX, or copper

Power supply

One 300 watt included with a dual power option

Typical consumption

150 watt (110 volt input)

Option 2: BIG-IP 3900 series

With a quad-core processor that enables support for multiple BIG-IP modules, the BIG-IP 3900
unifies application delivery in a 1U, cost-effective platform.
BIG-IP 3900 appliance-based networking technologies
Components

Value or description

Traffic throughput

4 Gbps

Hardware SSL

Included: 500 transactions per second

Maximum: 15,000 transactions per second
2.4 Gbps bulk encryption

Software compression

Included: 50 Mbps

55

Components

Value or description

Maximum: 3.8 Gbps

Processor

Dual core CPU

Memory

8 GB

Gigabit Ethernet CU ports

Gigabit fiber ports (small form-factor pluggable

transceiver)

4 optional LX, SX, or copper

Power supply

One 300 watt included with a dual power option

Typical consumption

175 watt (110 volt input)

Option 3: BIG-IP 6900 series

With two dual-core processors as well as hardware, SSL, and compression, the BIG-IP 6900 has
the performance to provide an integrated platform for application delivery. The BIG-IP 6900 can
process up to 6 Gbps of throughput to handle the most demanding applications.
BIG-IP 6900 appliance-based networking technologies
Components

Value or description

Traffic throughput

6 Gbps

Hardware SSL

Included: 500 transactions per second

Maximum: 25,000 transactions per second
4 Gbps bulk encryption

FIPS SSL

FIPS 140-2 Level 2 (option)

20,000 transactions per second

Software compression

Included: 50 Mbps
Maximum: 5 Gbps

Processor

Dual core CPU (4 processors)

Memory

8 GB

Gigabit Ethernet CU ports

16

Gigabit fiber ports (small form-factor pluggable

transceiver)

8 optional LX, SX, or copper

Power supply

Dual 850 watt included

Typical consumption

300 watt (110 volt input)

56

Step 3: Select a hardware load balancing solution model

When it comes time to determine which application delivery controller is suitable, consider the
following:

Purpose of the application delivery controllers:

Simple load balancing

Security

Acceleration

How users are connecting:

IMAP

Outlook Web App

Outlook Anywhere

Hardware benefits of an appliance-based BIG-IP vs. flexibility of a software-based BIG-IP

Desired scale and number of concurrent users

Percentage of local users vs. remote users

Average user expectations, such as number of messages per day average e-mail message
size

This information can be used to ensure the right BIG-IP LTM platform is selected.
*Design Decision Point*
The BIG-IP 3900 is selected for this solution. The 3900 4 GB capacity and connection count limits
are enough to cover normal usage as well as unexpected traffic spikes for 15,000 active
mailboxes with a 50 message per day profile. The quad core CPU is also capable enough to
handle the processing associated with connection and persistence handling.
Return to top

Determine Hardware Load Balancing Device Resiliency Strategy

Whenever deploying BIG-IP LTM, it's important that all efforts are made to implement with fault
tolerance in mind. BIG-IP LTM is designed to ensure that application server outages never affect
end users, and that the technology helps ensure BIG-IP LTM failures are recovered from in a
controlled and seamless manner.
Customers typically deploy BIG-IP LTM in redundant pairs. Connected by a dedicated network
and serial channel, the two BIG-IP LTMs coordinate network responsibilities, ensuring failure of
one device is automatically detected and recovered from by its peer. BIG-IP LTM excels in this
area by offering unique functionality such as:

Connection mirroring This ensures the connection table in each BIG-IP LTM is mirrored to
its peer. This means that in case of a BIG-IP LTM failure, no connections are dropped
because the BIG-IP LTM failover partner is already aware of the previously established
connections and it assumes responsibilities for the network.

57

Network-based outage detection This ensures that a network outage is just as critical as a
server outage for the BIG-IP LTM, and that proper remediation steps need to be taken to
attempt to remedy the situation.

Software-based and hardware-based watchdog functionality This ensures proper

failover when a BIG-IP LTM isn't functioning properly.

Besides deploying BIG-IP LTMs in redundant pairs, customers often build redundancy into the
architecture by building a multiple datacenter environment. BIG-IP GTM is designed to add
datacenter load balancing so that wide area resiliency is also achieved. For more information
about GTM, see Global Load Balancing Solutions.
Return to top

Determine Hardware Load Balancing Methods

Exchange protocols and client access services have different load balancing requirements. Some
Exchange protocols and client access services require client to Client Access server affinity.
Others work without it, but display performance improvements from such affinity. Other Exchange
protocols don't require client to Client Access server affinity, and performance doesn't decrease
without affinity. For additional information, see Load Balancing Requirements of Exchange
Protocols and Understanding Load Balancing in Exchange 2010.
For more information about configuring F5 BIG-IP LTMs, see Deploying F5 with Microsoft
Exchange Server 2010.
Return to top

Solution Overview
The previous section provided information about the design decisions that were made when
considering an Exchange 2010 solution. The following section provides an overview of the
solution.
Return to top

Logical Solution Diagram

This solution consists of a total of 18 Exchange 2010 servers deployed in a multisite topology.
Nine of the 18 servers are running both the Client Access and Hub Transport server roles. The
other nine servers are running the Mailbox server role. The primary namespace is load balanced
across six Client Access and Hub Transport servers in a Client Access server array in the primary
site. There are three Client Access and Hub Transport servers in a second Client Access server
array located in the secondary site. All nine Mailbox servers are members of a single DAG. There
are six Mailbox servers located in the primary site and three Mailbox servers in the secondary
site. The site resiliency model is active/passive.

58

Logical solution

Return to top

Physical Solution Diagram

This solution consists of nine Dell PowerEdge M610 blade servers in three PowerEdge M1000e
modular blade enclosures attached to three EqualLogic PS6500E iSCSI storage arrays via four
redundant modular PowerConnect M6220 switches. The hardware in this solution has been
provisioned such that there are three failure domains. A failure domain represents a single point
of failure and is used to ensure that database copy layouts in the DAG protect against loss of any
component in a failure domain. Each failure domain consists of one blade enclosure holding three
blade servers and two modular switches connected to a single PS6500E storage array.

59

Physical solution

Return to top

Server Hardware Summary

The following table summarizes the physical server hardware used in this solution.

60

Server hardware summary

Component

Description

Server vendor

Dell

Server model

PowerEdge M610 blade server

Processor

2 x Intel Xeon CPU X5550 2.66 GHz

Chipset

Intel 5520/5500/X58

Memory

48 GB

Operating system

Windows Server 2008 R2

Virtualization

Microsoft Hyper-V

Internal disk

2 x 300 GB SAS 15k

Operating system disk configuration

RAID-1

RAID controller

Dell SAS 6/iR integrated blades controller

Network interface

Broadcom NetXtreme II C-NIC GigE

For more information, see PowerEdge M610 Blade Server.

Return to top

Client Access and Hub Transport Server Configuration

The following table summarizes the Client Access and Hub Transport server configuration used in
this solution.
Client Access and Hub Transport server configuration
Component

Description

Physical or virtual

Hyper-V VM

Virtual processors

Memory

8 GB

Storage

Virtual hard disk on root server operating

system volume

Operating system

Windows Server 2008 R2

Exchange version

Exchange Server 2010 Standard Edition

Exchange patch level

Exchange 2010 Update Rollup 3

Return to top

61

Mailbox Server Configuration

The following table summarizes the Mailbox server configuration used in this solution.
Mailbox server configuration
Component

Description

Physical or virtual

Hyper-V VM

Virtual processors

Memory

32 GB

Storage

Virtual hard disk on root server operating

system volume

Pass-through storage

9 volumes 789 GB

Operating system

Windows Server 2008 R2

Exchange version

Exchange Server 2010 Enterprise Edition

Exchange patch level

Exchange 2010 Update Rollup 2

Third-party software

None

Return to top

Database Layout
The following diagram illustrates the database layout across the primary and secondary
datacenters.

62

Database layout

Return to top

Storage Hardware Summary

The following table summarizes the storage hardware used in this solution.
Storage hardware summary
Component

Description

Storage vendor

Dell

Storage model

EqualLogic PS6500E

63

Component

Description

Category

iSCSI

Disks

48 1 terabyte 7200 rpm SATA

Active disks

46

Spares

RAID level

10

Usable capacity

20.8 terabytes

For more information, see Dell EqualLogic PS6500E iSCSI SAN.

Return to top

Storage Configuration
Each of the Dell EqualLogic PS6500E storage arrays used in the solution were configured as
illustrated in the following table.
Storage configuration
Component

Description

Storage enclosures

LUNs per enclosure

27

LUNs per server

LUN size

798 GB

RAID level

RAID-10

The following table illustrates how the available storage was designed and allocated between the
three PS6500E storage arrays.
PS6500 storage array design and allocation
Database

Array1

Database

Array2

Database

Array3

DB1

C1

DB1

C2

DB1

C3

DB2

C1

DB2

C2

DB2

C3

DB3

C1

DB3

C2

DB3

C3

DB4

C1

DB4

C2

DB4

C3

DB5

C1

DB5

C2

DB5

C3

DB6

C1

DB6

C2

DB6

C3

64

Database

Array1

Database

Array2

Database

Array3

DB7

C1

DB7

C2

DB7

C3

DB8

C1

DB8

C2

DB8

C3

DB9

C1

DB9

C2

DB9

C3

DB10

C2

DB10

C1

DB10

C3

DB11

C2

DB11

C1

DB11

C3

DB12

C2

DB12

C1

DB12

C3

DB13

C2

DB13

C1

DB13

C3

DB14

C2

DB14

C1

DB14

C3

DB15

C2

DB15

C1

DB15

C3

DB16

C2

DB16

C1

DB16

C3

DB17

C2

DB17

C1

DB17

C3

DB18

C2

DB18

C1

DB18

C3

Return to top

Network Switch Hardware Summary

The following table summarizes the network switch hardware used in this solution.
Network switch hardware summary
Component

Description

Vendor

Dell

Model

PowerConnect M6220 Ethernet switch

Ports

20 (16 internal, 4 external)

Port bandwidth

10/100/1000 BASE-T auto-sensing

Switch fabric capacity

128 Gbps

Number per blade enclosures

For more information, download a .pdf file about the PowerConnect M6220 Ethernet Switch.
Return to top

Load Balancer Hardware Summary

The following table summarizes the storage hardware used in this solution.

65

Load balancer hardware summary

Component

Description

Vendor

F5

Model

BIG-IP 3900

Traffic throughput

4 Gbps

Hardware SSL

Included: 500 transactions per second

Maximum: 15,000 transactions per second
2.4 Gbps bulk encryption

Software compression

Included: 50 Mbps
Maximum: 3.8 Gbps

Processor

Dual core CPU

Memory

8 GB

Gigabit Ethernet CU ports

Gigabit fiber ports (small form-factor pluggable

transceiver)

4 optional LX, SX, or copper

Power supply

One 300 watt included with a dual power option

Typical consumption

175 watt (110 volt input)

Return to top

Solution Validation Methodology

Prior to deploying an Exchange solution in a production environment, validate that the solution
was designed, sized, and configured properly. This validation must include functional testing to
ensure that the system is operating as desired as well as performance testing to ensure that the
system can handle the desired user load. This section describes the approach and test
methodology used to validate server and storage design for this solution. In particular, the
following tests will be defined in detail:

Performance tests

Storage performance validation (Jetstress)

Server performance validation (Loadgen)

Functional tests

Database switchover validation

Server switchover validation

Server failover validation

Datacenter switchover validation

66

Return to top

Storage Design Validation Methodology

The level of performance and reliability of the storage subsystem connected to the Exchange
Mailbox server role has a significant impact on the overall health of the Exchange deployment.
Additionally, poor storage performance will result in high transaction latency, primarily reflected in
poor client experience when accessing the Exchange system. To ensure the best possible client
experience, validate storage sizing and configuration via the method described in this section.

Tool Set
For validating Exchange storage sizing and configuration, we recommend the Microsoft
Exchange Server Jetstress tool. The Jetstress tool is designed to simulate an Exchange I/O
workload at the database level by interacting directly with the ESE, which is also known as Jet.
The ESE is the database technology that Exchange uses to store messaging data on the Mailbox
server role. Jetstress can be configured to test the maximum I/O throughput available to your
storage subsystem within the required performance constraints of Exchange. Or, Jetstress can
accept a target profile of user count and per-user IOPS, and validate that the storage subsystem
is capable of maintaining an acceptable level of performance with the target profile. Test duration
is adjustable and can be run for a minimal period of time to validate adequate performance or for
an extended period of time to additionally validate storage subsystem reliability.
The Jetstress tool can be obtained from the Microsoft Download Center at the following locations:

Microsoft Exchange Server Jetstress 2010 (64 bit)

Microsoft Exchange Server Jetstress 2010 (32 bit)

The documentation included with the Jetstress installer describes how to configure and execute a
Jetstress validation test on your server hardware.
Approach to Storage Validation
There are two main types of storage configurations:

Direct-attached storage (DAS) or internal disk scenarios

Storage area network (SAN) scenarios

With DAS or internal disk scenarios, there's only one server accessing the disk subsystem, so the
performance capabilities of the storage subsystem can be validated in isolation.
In SAN scenarios, the storage utilized by the solution may be shared by many servers and the
infrastructure that connects the servers to the storage may also be a shared dependency. This
requires additional testing, as the impact of other servers on the shared infrastructure must be
adequately simulated to validate performance and functionality.

Test Cases for Storage Validation

The following storage validation test cases were executed against the solution and should be
considered as a starting point for storage validation. Specific deployments may have other

67

validation requirements that can be met with additional testing, so this list isn't intended to be
exhaustive:

Validation of worst case database switchover scenario In this test case, the level of I/O
is expected to be serviced by the storage subsystem in a worst case switchover scenario
(largest possible number of active copies on fewest servers). Depending on whether the
storage subsystem is DAS or SAN, this test may be required to run on multiple hosts to
ensure that the end-to-end solution load on the storage subsystem can be sustained.

Validation of storage performance under storage failure and recovery scenario (for
example, failed disk replacement and rebuild) In this test case, the performance of the
storage subsystem during a failure and rebuild scenario is evaluated to ensure that the
necessary level of performance is maintained for optimal Exchange client experience. The
same caveat applies for a DAS vs. SAN deployment: If multiple hosts are dependent on a
shared storage subsystem, the test must include load from these hosts to simulate the entire
effect of the failure and rebuild.

Analyzing the Results

The Jetstress tool produces a report file after each test is completed. To help you analyze the
report, use the guidelines in Reading Jetstress 2010 Test Reports.
Specifically, you should use the guidelines in the following table when you examine data in the
Test Results table of the report.
Jetstress results analysis
Performance counter instance

Guidelines for performance test

I/O Database Reads Average Latency (msec)

The average value should be less than 20

milliseconds (msec) (0.020 seconds), and the
maximum values should be less than 50 msec.

I/O Log Writes Average Latency (msec)

Log disk writes are sequential, so average write

latencies should be less than 10 msec, with a
maximum of no more than 50 msec.

%Processor Time

Average should be less than 80%, and the

maximum should be less than 90%.

Transition Pages Repurposed/sec (Windows

Server 2003, Windows Server 2008, Windows
Server 2008 R2)

Average should be less than 100.

The report file shows various categories of I/O performed by the Exchange system:

Transactional I/O Performance This table reports I/O that represents user activity against
the database (for example, Outlook generated I/O). This data is generated by subtracting
background maintenance I/O and log replication I/O from the total I/O measured during the
test. This data provides the actual database IOPS generated along with I/O latency
measurements required to determine whether a Jetstress performance test passed or failed.

68

Background Database Maintenance I/O Performance This table reports the I/O
generated due to ongoing ESE database background maintenance.

Log Replication I/O Performance This table reports the I/O generated from simulated log
replication.

Total I/O Performance This table reports the total I/O generated during the Jetstress test.

Return to top

Server Design Validation

After the performance and reliability of the storage subsystem is validated, ensure that all of the
components in the messaging system are validated together for functionality, performance, and
scalability. This means moving up in the stack to validate client software interaction with the
Exchange product as well as any server-side products that interact with Exchange. To ensure
that the end-to-end client experience is acceptable and that the entire solution can sustain the
desired user load, the method described in this section can be applied for server design
validation.

Tool Set
For validation of end-to-end solution performance and scalability, we recommend the Microsoft
Exchange Server Load Generator tool (Loadgen). Loadgen is designed to produce a simulated
client workload against an Exchange deployment. This workload can be used to evaluate the
performance of the Exchange system, and can also be used to evaluate the effect of various
configuration changes on the overall solution while the system is under load. Loadgen is capable
of simulating Microsoft Office Outlook 2007 (online and cached), Office Outlook 2003 (online and
cached), POP3, IMAP4, SMTP, ActiveSync, and Outlook Web App (known in Exchange 2007
and earlier versions as Outlook Web Access) client activity. It can be used to generate a single
protocol workload, or these client protocols can be combined to generate a multiple protocol
workload.
You can get the Loadgen tool from the Microsoft Download Center at the following locations:

Exchange Load Generator 2010 (64 bit)

Exchange Load Generator 2010 (32 bit)

The documentation included with the Loadgen installer describes how to configure and execute a
Loadgen test against an Exchange deployment.
Approach to Server Validation
When validating your server design, test the worst case scenario under anticipated peak
workload. Based on a number of data sets from Microsoft IT and other customers, peak load is
generally equal to 2x the average workload throughout the remainder of the work day. This is
referred to as the peak-to-average workload ratio.

69

Peak load

In this Performance Monitor snapshot, which displays various counters that represent the amount
of Exchange work being performed over time on a production Mailbox server, the average value
for RPC operations per second (the highlighted line) is about 2,386 when averaged across the
entire day. The average for this counter during the peak period from 10:00 through 11:00 is about
4,971, giving a peak-to-average ratio of 2.08.
To ensure that the Exchange solution is capable of sustaining the workload generated during the
peak average, modify Loadgen settings to generate a constant amount of load at the peak
average level, rather than spreading out the workload over the entire simulated work day.
Loadgen task-based simulation modules (like the Outlook simulation modules) utilize a task
profile that defines the number of times each task will occur for an average user within a
simulated day.
The total number of tasks that need to run during a simulated day is calculated as the number of
users multiplied by the sum of task counts in the configured task profile. Loadgen then
determines the rate at which it should run tasks for the configured set of users by dividing the
total number of tasks to run in the simulated day by the simulated day length. For example, if
Loadgen needs to run 1,000,000 tasks in a simulated day, and a simulated day is equal to 8
hours (28,800 seconds), Loadgen must run 1,000,000 28,800 = 34.72 tasks per second to meet
the required workload definition. To increase the amount of load to the desired peak average,
divide the default simulated day length (8 hours) by the peak-to-average ratio (2) and use this as
the new simulated day length.
Using the task rate example again, 1,000,000 14,400 = 69.44 tasks per second. This reduces
the simulated day length by half, which results in doubling the actual workload run against the
server and achieving our goal of a peak average workload. You don't adjust the run length
duration of the test in the Loadgen configuration. The run length duration specifies the duration of
the test and doesn't affect the rate at which tasks will be run against the Exchange server.

70

Test Cases for Server Design Validation

The following server design validation test cases were executed against the solution and should
be considered as a starting point for server design validation. Specific deployments may have
other validation requirements that can be met with additional testing, so this list isn't intended to
be exhaustive:

Normal operating conditions In this test case, the basic design of the solution is validated
with all components in their normal operating state (no failures simulated). The desired
workload is generated against the solution, and the overall performance of the solution is
validated against the metrics that follow.

Single server failure or single server maintenance (in site) In this test case, a single
server is taken down to simulate either an unexpected failure of the server or a planned
maintenance operation for the server. The workload that would normally be handled by the
unavailable server is now handled by other servers in the solution topology, and the overall
performance of the solution is validated.

Test Execution and Data Collection

Exchange performance data has some natural variation within test runs and among test runs. We
recommend that you take the average of multiple runs to smooth out this variation. For Exchange
tested solutions, a minimum of three separate test runs with durations of eight hours was
completed. Performance data was collected for the full eight-hour duration of the test.
Performance summary data was taken from a three to four hour stable period (excluding the first
two hours of the test and the last hour of the test). For each Exchange server role, performance
summary data was averaged between servers for each test run, providing a single average value
for each data point. The values for each run were then averaged, providing a single data point for
all servers of a like server role across all test runs.

Validation of Expected Load

Before you look at any performance counters or start your performance validation analysis, verify
that the workload you expected to run matched the workload that you actually ran. Although there
are many ways to determine whether the simulated workload matched the expected workload,
the easiest and most consistent way is to look at the message delivery rate.
Calculating Expected Peak Message Delivery Rate
Every message profile consists of the sum of the average number of messages sent per day and
the average number of messages received per day. To calculate the message delivery rate,
select the average number of messages received per day from the following table.
Peak message delivery rate
Message profile

Messages sent per day

Messages received per day

50

10

40

100

20

80

71

Message profile

Messages sent per day

Messages received per day

150

30

120

200

40

160

The following example assumes that each Mailbox server has 5,000 active mailboxes with a 150
messages per day profile (30 messages sent and 120 messages received per day).
Peak message delivery rate for 5,000 active mailboxes
Description

Calculation

Value

Message profile

Number of messages received

per day

120

Number of active mailboxes per Not applicable

Mailbox server

5000

Total messages received per

day per Mailbox server

5000 120

600000

Total messages received per

second per Mailbox server

600000 28800

20.83

Total messages adjusted for

peak load

20.83 2

41.67

You expect 41.67 messages per second delivered on each Mailbox server running 5,000 active
mailboxes with a message profile of 150 messages per day during peak load.
Measuring Actual Message Delivery Rate
The actual message delivery rate can be measured using the following counter on each Mailbox
server: MSExchangeIS Mailbox(_Total)\Messages Delivered/sec. If the measured message
delivery rate is within one or two messages per second of the target message delivery rate, you
can be confident that the desired load profile was run successfully.

Server Validation: Performance and Health Criteria

This section describes the Performance Monitor counters and thresholds used to determine
whether the Exchange environment was sized properly and is able to run in a healthy state during
extended periods of peak workload. For more information about counters relevant to Exchange
performance, see Performance and Scalability Counters and Thresholds.
Hyper-V Root Servers
To validate the performance and health criteria of a Hyper-V root server and the applications
running within VMs, you should have a basic understanding of the Hyper-V architecture and how
that impacts performance monitoring.

72

Hyper-V has three main components: the virtualization stack, the hypervisor, and devices. The
virtualization stack handles emulated devices, manages VMs, and services I/O. The hypervisor
schedules virtual processors, manages interrupts, services timers, and controls other chip-level
functions. The hypervisor doesn't handle devices or I/O (for example, there are no hypervisor
drivers). The devices are part of the root server or installed in guest servers as part of integration
services. Because the root server has a full view of the system and controls the VMs, it also
provides monitoring information via Windows Management Instrumentation (WMI) and
performance counters.
Processor
When validating physical processor utilization on the root server (or within the guest VM), the
standard Processor\% Processor Time counter isn't very useful.
Instead, you can examine the Hyper-V Hypervisor Logical Processor\% Total Run Time counter.
This counter shows the percentage of processor time spent in guest and hypervisor runtime and
should be used to measure the total processor utilization for the hypervisor and all VMs running
on the root server. This counter shouldn't exceed 80 percent or whatever the maximum utilization
target you have designed for.
Counter

Target

Hyper-V Hypervisor Logical Processor\% Total

Run Time

<80%

If you're interested in what percentage of processor time is spent servicing the guest VMs, you
can examine the Hyper-V Hypervisor Logical Processor\% Guest Run Time counter. If you're
interested in what percentage of processor time is spent in hypervisor, you can look at the HyperV Hypervisor Logical Processor\% Hypervisor Run Time counter. This counter should be below
5 percent. The Hyper-V Hypervisor Root Virtual Processor\% Guest Run Time counter shows the
percentage of processor time spent in the virtualization stack. This counter should also be below
5 percent. These two counters can be used to determine what percentage of your available
physical processor time is being used to support virtualization.
Counter

Target

Hyper-V Hypervisor Logical Processor\% Guest

Run Time

<80%

Hyper-V Hypervisor Logical Processor\%

Hypervisor Run Time

<5%

Hyper-V Hypervisor Root Virtual Processor\%

Guest Run Time

<5%

Memory

73

You need to ensure that your Hyper-V root server has enough memory to support the memory
allocated to VMs. Hyper-V automatically reserves 512 MB (this may vary with different Hyper-V
releases) for the root operating system. If you don't have enough memory, Hyper-V will prevent
the last VM from starting. In general, don't worry about validating the memory on a Hyper-V root
server. Be more concerned with ensuring that sufficient memory is allocated to the VMs to
support the Exchange roles.
Application Health
An easy way to determine whether all the VMs are in a healthy state is to look at the Hyper-V
Virtual Machine Health Summary counters.
Counter

Target

Hyper-V Virtual Machine Health

Summary\Health OK

Hyper-V Virtual Machine Health

Summary\Health Critical

Mailbox Servers
When validating whether a Mailbox server was properly sized, focus on processor, memory,
storage, and Exchange application health. This section describes the approach to validating each
of these components.
Processor
During the design process, you calculated the adjusted megacycle capacity of the server or
processor platform. You then determined the maximum number of active mailboxes that could be
supported by the server without exceeding 80 percent of the available megacycle capacity. You
also determined what the projected CPU utilization should be during normal operating conditions
and during various server maintenance or failure scenarios.
During the validation process, verify that the worst case scenario workload doesn't exceed
80 percent of the available megacycles. Also, verify that actual CPU utilization is close to the
expected CPU utilization during normal operating conditions and during various server
maintenance or failure scenarios.
For physical Exchange deployments, use the Processor(_Total)\% Processor Time counter and
verify that this counter is less than 80 percent on average.
Counter

Target

Processor(_Total)\% Processor Time

<80%

For virtual Exchange deployments, the Processor(_Total)\% Processor Time counter is measured
within the VM. In this case, the counter isn't measuring the physical CPU utilization. It's
measuring the utilization of the virtual CPU provided by the hypervisor. Therefore, it doesn't
provide an accurate reading of the physical processor and shouldn't be used for design validation

74

purposes. For more information, see Hyper-V: Clocks lie... which performance counters can you
trust.
For validating Exchange deployments running on Microsoft Hyper-V, use the Hyper-V Hypervisor
Virtual Processor\% Guest Run Time counter. This provides a more accurate value for the
amount of physical CPU being utilized by the guest operating system. This counter should be less
than 80 percent on average.
Counter

Target

Hyper-V Hypervisor Virtual Processor\% Guest

Run Time

<80%

Memory
During the design process, you calculated the amount of database cache required to support the
maximum number of active databases on each Mailbox server. You then determined the optimal
physical memory configuration to support the database cache and system memory requirements.
Validating whether an Exchange Mailbox server has sufficient memory to support the target
workload isn't a simple task. Using available memory counters to view how much physical
memory is remaining isn't helpful because the memory manager in Exchange is designed to use
almost all of the available physical memory. The information store (store.exe) reserves a large
portion of physical memory for database cache. The database cache is used to store database
pages in memory. When a page is accessed in memory, the information doesn't have to be
retrieved from disk, reducing read I/O. The database cache is also used to optimize write I/O.
When a database page is modified (known as a dirty page), the page stays in cache for a period
of time. The longer it stays in cache, the better the chance that the page will be modified multiple
times before those changes are written to the disk. Keeping dirty pages in cache also causes
multiple pages to be written to the disk in the same operation (known as write coalescing).
Exchange uses as much of the available memory in the system as possible, which is why there
aren't large amounts of available memory on an Exchange Mailbox server.
It may not be easy to know whether the memory configuration on your Exchange Mailbox server
is undersized. For the most part, the Mailbox server will still function, but your I/O profile may be
much higher than expected. Higher I/O can lead to higher disk read and write latencies, which
may impact application health and client user experience. In the results section, there isn't any
reference to memory counters. Potential memory issues will be identified in the storage validation
and application health result sections, where memory-related issues are more easily detected.
Storage
If you have performance issues with your Exchange Mailbox server, those issues may be
storage-related issues. Storage issues may be caused by having an insufficient number of disks
to support the target I/O requirements, having overloaded or poorly designed storage connectivity
infrastructure, or by factors that change the target I/O profile like insufficient memory, as
discussed previously.

75

The first step in storage validation is to verify that the database latencies are below the target
thresholds. In previous releases, logical disk counters determined disk read and write latency. In
Exchange 2010, the Exchange Mailbox server that you are monitoring is likely to have a mix of
active and passive mailbox database copies. The I/O characteristics of active and passive
database copies are different. Because the size of the I/O is much larger on passive copies, there
are typically much higher latencies on passive copies. Latency targets for passive databases are
200 msec, which is 10 times higher than targets on active database copies. This isn't much of a
concern because high latencies on passive databases have no impact on client experience. But if
you are using the traditional logical disk counters to measure latencies, you must review the
individual volumes and separate volumes containing active and passive databases. Instead, we
recommend that you use the new MSExchange Database counters in Exchange 2010.
When validating latencies on Exchange 2010 Mailbox servers, we recommend you use the
counters in the following table for active databases.
Counter

Target

MSExchange Database\I/O Database Reads

(Attached) Average Latency

<20 msec

MSExchange Database\I/O Database Writes

(Attached) Average Latency

<20 msec

MSExchange Database\IO Log Writes Average

Latency

<1 msec

We recommend that you use the counters in the following table for passive databases
Counter

Target

MSExchange Database\I/O Database Reads

(Recovery) Average Latency

<200 msec

MSExchange Database\I/O Database Writes

(Recovery) Average Latency

<200 msec

MSExchange Database\IO Log Read Average

Latency

<200 msec

Note:
To view these counters in Performance Monitor, you must enable the advanced database
counters. For more information, see How to Enable Extended ESE Performance
Counters.
When you're validating disk latencies for Exchange deployments running on Microsoft Hyper-V,
be aware that the I/O Database Average Latency counters (as with many time-based counters)
may not be accurate because the concept of time within the VM is different than on the physical

76

server. The following example shows that the I/O Database Reads (Attached) Average Latency is
22.8 in the VM and 17.3 on a physical server for the same simulated workload. If the values of
time-based counters are over the target thresholds, your server may be running correctly. Review
all health criteria to make a decision regarding server health when your Mailbox server role is
deployed within a VM.
Values of disk latency counters for virtual and physical Mailbox servers
Counter

Virtual Mailbox server

Physical Mailbox server

I/O Database Reads (Attached) /

Average Latency

22.792

17.250

I/O Database Reads (Attached) /

sec

17.693

18.131

I/O Database Reads (Recovery) / 34.215

Average Latency

27.758

MSExchange Database/

I/O Database Writes (Recovery) /

sec

10.829

8.483

I/O Database Writes (Attached) /

Average Latency

0.944

0.411

I/O Database Writes (Attached) /

sec

10.184

10.963

1.966

1.695

RPC Operations / sec

334.371

341.139

RPC Packets / sec

180.656

183.360

Messages Delivered / sec

2.062

2.065

Messages Sent / sec

0.511

0.514

MSExchangeIS
RPC Averaged Latency

MSExchangeIS Mailbox

In addition to disk latencies, review the Database\Database Page Fault Stalls/sec counter. This
counter indicates the rate of page faults that can't be serviced because there are no pages
available for allocation from the database cache. This counter should be 0 on a healthy server.
Counter

Target

Database\Database Page Fault Stalls/sec

<1

77

Also, review the Database\Log Record Stalls/sec counter, which indicates the number of log
records that can't be added to the log buffers per second because the log buffers are full. This
counter should average less than 10.
Counter

Target

Database\Log Record Stalls/sec

<10

Exchange Application Health

Even if there are no obvious issues with processor, memory, and disk, we recommend that you
monitor the standard application health counters to ensure that the Exchange Mailbox server is in
a healthy state.
The MSExchangeIS\RPC Averaged Latency counter provides the best indication of whether other
counters with high database latencies are actually impacting Exchange health and client
experience. Often, high RPC averaged latencies are associated with a high number of RPC
requests, which should be less than 70 at all times.
Counter

Target

MSExchangeIS\RPC Averaged Latency

<10 msec on average

MSExchangeIS\RPC Requests

<70 at all times

Next, make sure that the transport layer is healthy. Any issues in transport or issues downstream
of transport affecting the transport layer can be detected with the MSExchangeIS
Mailbox(_Total)\Messages Queued for Submission counter. This counter should be less than 50
at all times. There may be temporary increases in this counter, but the counter value shouldn't
grow over time and shouldn't be sustained for more than 15 minutes.
Counter

Target

MSExchangeIS Mailbox(_Total)\Messages
Queued for Submission

<50 at all times

Next, ensure that maintenance of the database copies is in a healthy state. Any issues with log
shipping or log replay can be identified using the MSExchange Replication(*)\CopyQueueLength
and MSExchange Replication(*)\ReplayQueueLength counters. The copy queue length shows
the number of transaction log files waiting to be copied to the passive copy log file folder and
should be less than 1 at all times. The replay queue length shows the number of transaction log
files waiting to be replayed into the passive copy and should be less than 5. Higher values don't
impact client experience, but result in longer store mount times when a handoff, failover, or
activation is performed.

78

Counter

Target

MSExchange Replication(*)\CopyQueueLength

<1

MSExchange
Replication(*)\ReplayQueueLength

<5

Client Access Servers

To determine whether a Client Access server is healthy, review processor, memory, and
application health. For an extended list of important counters, see Client Access Server Counters.
Processor
For physical Exchange deployments, use the Processor(_Total)\% Processor Time counter. This
counter should be less than 80 percent on average.
Counter

Target

Processor(_Total)\% Processor Time

<80%

For validating Exchange deployments running on Microsoft Hyper-V, use the Hyper-V Hypervisor
Virtual Processor\% Guest Run Time counter. This provides an accurate value for the amount of
physical CPU being utilized by the guest operating system. This counter should be less than
80 percent on average.
Counter

Target

Hyper-V Hypervisor Virtual Processor\% Guest

Run Time

<80%

Application Health
To determine whether the MAPI client experience is acceptable, use the MSExchange
RpcClientAccess\RPC Averaged Latency counter. This counter should be below 250 msec. High
latencies can be associated with a large number of RPC requests. The MSExchange
RpcClientAccess\RPC Requests counter should be below 40 on average.
Counter

Target

MSExchange RpcClientAccess\RPC Averaged

Latency

<250 msec

MSExchange RpcClientAccess\RPC Requests

<40

79

Transport Servers
To determine whether a transport server is healthy, review processor, disk, and application
health. For an extended list of important counters, see Transport Server Counters.
Processor
For physical Exchange deployments, use the Processor(_Total)\% Processor Time counter. This
counter should be less than 80 percent on average.
Counter

Target

Processor(_Total)\% Processor Time

<80%

For validating Exchange deployments running on Microsoft Hyper-V, use the Hyper-V Hypervisor
Virtual Processor\% Guest Run Time counter. This provides an accurate value for the amount of
physical CPU being utilized by the guest operating system. This counter should be less than
80 percent on average.
Counter

Target

Hyper-V Hypervisor Virtual Processor\% Guest

Run Time

<80%

Disk
To determine whether disk performance is acceptable, use the Logical Disk(*)\Avg. Disk
sec/Read and Write counters for the volumes containing the transport logs and database. Both of
these counters should be less than 20 msec.
Counter

Target

Logical Disk(*)\Avg. Disk sec/Read

<20 msec

Logical Disk(*)\Avg. Disk sec/Write

<20 msec

Application Health
To determine whether a Hub Transport server is sized properly and running in a healthy state,
examine the MSExchangeTransport Queues counters outlined in the following table. All of these
queues will have messages at various times. You want to ensure that the queue length isn't
sustained and growing over a period of time. If larger queue lengths occur, this could indicate an
overloaded Hub Transport server. Or, there may be network issues or an overloaded Mailbox
server that's unable to receive new messages. You will need to check other components of the
Exchange environment to verify.
Counter

Target

MSExchangeTransport

<3000

80

Counter

Target

Queues(_total)\Aggregate Delivery
MSExchangeTransport Queues(_total)\Active
Remote Delivery Queue Length

<250

MSExchangeTransport Queues(_total)\Active
Mailbox Delivery Queue Length

<250

MSExchangeTransport Queues(_total)\Retry
Mailbox Delivery Queue Length

<100

MSExchangeTransport
Queues(_total)\Submission Queue Length

<100

Return to top

Functional Validation Tests

You can use the information in the following sections for functional validation tests.

Database Switchover Validation

A database switchover is the process by which an individual active database is switched over to
another database copy (a passive copy), and that database copy is made the new active
database copy. Database switchovers can happen both within and across datacenters. A
database switchover can be performed by using the Exchange Management Console (EMC) or
the Exchange Management Shell.
To validate that a passive copy of a database can be successfully activated on another server,
run the following command.
Move-ActiveMailboxDatabase <DatabaseName> -ActivateOnServer
<TargetServer>
Success criteria: The active mailbox database is mounted on the specified target server. This
result can be confirmed by running the following command.
Get-MailboxDatabaseCopyStatus <DatabaseName>

Server Switchover Validation

A server switchover is the process by which all active databases on a DAG member are activated
on one or more other DAG members. Like database switchovers, a server switchover can occur
both within a datacenter and across datacenters, and it can be initiated by using both the EMC
and the Shell.

To validate that all passive copies of databases on a server can be successfully activated on
other servers hosting a passive copy, run the following command.

81

Get-MailboxDatabase -Server <ActiveMailboxServer> | MoveActiveMailboxDatabase -ActivateOnServer <TargetServer>

Success criteria: The active mailbox databases are mounted on the specified target server.
This can be confirmed by running the following command.
Get-MailboxDatabaseCopyStatus <DatabaseName>

To validate that one copy of each of the active databases will be successfully activated on
another Mailbox server hosting passive copies of the databases, shut down the server by
performing the following action.
Turn off the current active server.
Success criteria: The active mailbox databases are mounted on another Mailbox server in the
DAG. This can be confirmed by running the following command.
Get-MailboxDatabaseCopyStatus <DatabaseName>

Server Failover Validation

A server failover occurs when the DAG member can no longer service the MAPI network, or
when the Cluster service on a DAG member can no longer contact the remaining DAG members.
To validate that one copy of each of the active databases will be successfully activated on
another Mailbox server hosting passive copies of the databases, turn off the server by performing
one of the following actions:

Press and hold the power button on the server until the server turns off.

Pull the power cables from the server, which results in the server turning off.

Success criteria: The active mailbox databases are mounted on another Mailbox server in the
DAG. This can be confirmed by running the following command.
Get-MailboxDatabase -Server <MailboxServer> | GetMailboxDatabaseCopyStatus
Return to top

Datacenter Switchover Validation

A datacenter or site failure is managed differently from the types of failures that can cause a
server or database failover. In a high availability configuration, automatic recovery is initiated by
the system, and the failure typically leaves the messaging system in a fully functional state. By
contrast, a datacenter failure is considered to be a disaster recovery event, and as such, recovery
must be manually performed and completed for the client service to be restored and for the
outage to end. The process you perform is called a datacenter switchover. As with many disaster
recovery scenarios, prior planning and preparation for a datacenter switchover can simplify your
recovery process and reduce the duration of your outage.
For more information, including detailed steps for performing a datacenter switchover, see
Datacenter Switchovers.
There are four basic steps that you complete to perform a datacenter switchover, after making the
initial decision to activate the second datacenter:

82

1. Terminate a partially running datacenter.

2. Validate and confirm the prerequisites for the second datacenter.
3. Activate the Mailbox servers.
4. Activate the Client Access servers.
The following describes the steps used to validate a datacenter switchover.

Terminate Partially Failed Datacenter (Assuming DAG Is in DAC Mode)

When the DAG is in DAC mode, the specific actions to terminate any surviving DAG members in
the primary datacenter depend on the state of the failed datacenter. Perform one of the following:

If the Mailbox servers in the failed datacenter are still accessible (usually not the case), run
the following command on each Mailbox server.
Stop-DatabaseAvailabilityGroup -ActiveDirectorySite <insertsitename>

If the Mailbox servers in the failed datacenter are unavailable but Active Directory is operating
in the primary datacenter, run the following command on a domain controller.
Stop-DatabaseAvailabilityGroup -ActiveDirectorySite <insertsitename>
-ConfigurationOnly
Note:
Failure to either turn off the Mailbox servers in the failed datacenter or to successfully
perform the Stop-DatabaseAvailabilityGroup command against the servers will create
the potential for split brain syndrome to occur across the two datacenters. You may need
to individually turn off computers through power management devices to satisfy this
requirement.

Success criteria: All Mailbox servers in the failed site are in a stopped state. You can verify this by
running the following command from a server in the failed datacenter.
Get-DatabaseAvailabilityGroup | Format-List

Validate and Confirm Prerequisites for Secondary Datacenter

The second datacenter must be updated to represent which primary datacenter servers are
stopped. From a server in the secondary datacenter, run the following command.
Stop-DatabaseAvailabilityGroup -ActiveDirectorySite <insertsitename> ConfigurationOnly
The purpose of this step is to inform the servers in the secondary datacenter about which Mailbox
servers are available to use when restoring service.
Success criteria: All Mailbox servers in the failed datacenter are in a stopped state. To verify this,
run the following command from a server in the secondary datacenter.
Get-DatabaseAvailabilityGroup | Format-List

83

Activate Mailbox Servers (Assuming DAG Is in DAC Mode)

Before activating the DAG members in the secondary datacenter, we recommend that you verify
that the infrastructure services in the secondary datacenter are ready for messaging service
activation.
When the DAG is in DAC mode, the steps to complete activation of the Mailbox servers in the
second datacenter are as follows:
1. Stop the cluster service on each DAG member in the secondary datacenter. You can use the
Stop-Service cmdlet to stop the service (for example, Stop-Service ClusSvc), or use net
stop clussvc from an elevated command prompt.
2. To activate the Mailbox servers in the secondary datacenter, run the following command.
Restore-DatabaseAvailabilityGroup -Identity <DAGname> ActiveDirectorySite <insertsitename>
If this command succeeds, the quorum criteria are shrunk to the servers in the secondary
datacenter. If the number of servers in that datacenter is an even number, the DAG will
switch to using the alternate witness server as identified by the setting on the DAG object.
3. To activate the databases, run one of the following commands.
Get-MailboxDatabase <insertcriteriatoselectDBs> | MoveActiveMailboxDatabase -ActivateOnServer <DAGMemberinPrimarySite>
or
Move-ActiveMailboxDatabase -Server <DAGMemberInSecondarySite> ActivateOnServer <DAGMemberinPrimarySite>
4. Check the event logs and review all error and warning messages to ensure that the
secondary site is healthy. Any indicated issues should be followed up and corrected prior to
mounting the databases.
5. To mount the databases, run the following command.
Get-MailboxDatabase <DAGMemberInSecondarySite> | Mount-Database
Success criteria: The active mailbox databases are mounted on Mailbox servers in the secondary
site. To confirm, run the following command.
Get-MailboxDatabaseCopyStatus <DatabaseName>

Activate Client Access Servers

Clients connect to service endpoints to access Exchange services and data. Activating Internetfacing Client Access servers therefore involves changing DNS records to point to the new IP
addresses that will be configured for the new service endpoints. Clients will then automatically
connect to the new service endpoints in one of two ways:

Clients will continue to try to connect, and should automatically connect after Time to Live
(TTL) has expired for the original DNS entry, and after the entry is expired from the client's
DNS cache. Users can also run the ipconfig /flushdns command from a command prompt
to manually clear their DNS cache. If using Outlook Web App, the Web browser may need to
be closed and restarted to clear the DNS cache used by the browser. In Exchange 2010 SP1,

84

this browser caching issue can be mitigated by configuring the FailbackURL parameter on
the Outlook Web App virtual directory owa.

Clients starting or restarting will perform a DNS lookup on startup and will get the new IP
address for the service endpoint, which will be a Client Access server or array in the second
datacenter.

To validate the scenario with Loadgen, perform the following actions:

1. Change the DNS entry for the Client Access server array to point to the virtual IP address of
the hardware load balancing in the secondary site.
2. Run the ipconfig /flushdns command on all Loadgen servers.
3. Restart the Loadgen test.
4. Verify that the Client Access servers in the secondary site are now servicing the load.
To validate the scenario with an Outlook 2007 client, perform the following:
1. Change the DNS entry for the Client Access server array to point to the VIP of the hardware
load balancing in the secondary site.
2. Run the ipconfig /flushdns command on the client or wait until TTL expires.
3. Wait for the Outlook client to reconnect.
Return to top

Primary Datacenter Service Restoration Validation

The process of restoring service to a previously failed datacenter is referred to as a failback. The
steps used to perform a datacenter failback are similar to the steps used to perform a datacenter
switchover. A significant distinction is that datacenter failbacks are scheduled, and the duration of
the outage is often much shorter.
It's important that failback not be performed until the infrastructure dependencies for Exchange
have been reactivated, are functioning and stable, and have been validated. If these
dependencies aren't available or healthy, it's likely that the failback process will cause a longer
than necessary outage, and it's possible the process could fail altogether.

Mailbox Server Role Failback (Assuming DAG Is in DAC Mode)

The Mailbox server role should be the first role that's failed back to the primary datacenter. The
following steps detail the Mailbox server role failback process.
1. To reincorporate the DAG members in the primary site, run the following command.
Start-DatabaseAvailabilityGroup -Identity
<DatabaseAvailabilityGroupIdParameter> -ActiveDirectorySite
<insertsitename>
2. To verify the state of the database copies in the primary datacenter, run the following
command.
Get-MailboxDatabaseCopyStatus

85

After the Mailbox servers in the primary datacenter have been incorporated into the DAG, they
will need some time to synchronize their database copies. Depending on the nature of the failure,
the length of the outage, and actions taken by an administrator during the outage, this may
require reseeding the database copies. For example, if during the outage, you remove the
database copies from the failed primary datacenter to allow log file truncation to occur for the
surviving active copies in the secondary datacenter, reseeding will be required. At this time, each
database can be synchronized individually. After a replicated database copy in the primary
datacenter is healthy, you can proceed to the next step.
1. During the datacenter switchover process, the DAG was configured to use an alternate
witness server. To reconfigure the DAG to use a witness server in the primary datacenter, run
the following command.
Set-DatabaseAvailabilityGroup -Identity <DAGName> -WitnessServer
<PrimaryDatacenterWitnessServer>
2. The databases being reactivated in the primary datacenter should now be dismounted in the
secondary datacenter. Run the following command.
Get-MailboxDatabase | Dismount-Database
3. After the databases have been dismounted, the Client Access server URLs should be moved
from the secondary datacenter to the primary datacenter. To do this, change the DNS record
for the URLs to point to the Client Access server or array in the primary datacenter.
Important:
Don't proceed to the next step until the Client Access server URLs have been moved
and the DNS TTL and cache entries have expired. Activating the databases in the
primary datacenter prior to moving the Client Access server URLs to the primary
datacenter will result in an invalid configuration (for example, a mounted database
that has no Client Access servers in its Active Directory site).
4. To activate the databases, run one of the following commands.
Get-MailboxDatabase <insertcriteriatoselectDBs> | MoveActiveMailboxDatabase -ActivateOnServer <DAGMemberinSecondSite>
or
Move-ActiveMailboxDatabase -Server <DAGMemberinPrimarySite> ActivateOnServer <DAGMemberinSecondSite>
5. To mount the databases, run the following command.
Get-MailboxDatabase <insertcriteriatoselectDBs> | Mount-Database
Success criteria: The active mailbox databases are successfully mounted on Mailbox servers in
the primary site. To confirm, run the following command.
Get-MailboxDatabaseCopyStatus <DatabaseName>
Return to top

86

Storage Design Validation Results

The following tables summarize the Jetstress storage validation results. This solution achieved
higher than target transactional I/O while maintaining database latencies well under the 20 msec
target.
Overall test result

Pass

Overall throughput
Overall throughput

Result

Target transactional I/O per second

383

Achieved transactional I/O per second

540

Transactional I/O performance: database reads

Database

I/O database reads per second

I/O database reads average

latency (msec)

Instance1

42.2

18.9

Instance2

42.7

17.9

Instance3

42.9

17.4

Instance4

42.0

17.9

Instance5

42.0

18.0

Instance6

41.8

17.0

Instance7

42.8

17.7

Instance8

42.6

17.4

Transactional I/O performance: database writes

Database

I/O database writes per second

I/O database writes average

latency (msec)

Instance1

25.9

25.9

Instance2

26.4

25.1

Instance3

26.4

21.7

Instance4

26.1

22.6

Instance5

25.9

23.8

87

Database

I/O database writes per second

I/O database writes average

latency (msec)

Instance6

25.5

19.8

Instance7

26.3

21.2

Instance8

26.5

18.5

Transactional I/O performance: log writes

Database

I/O log writes per second

I/O database writes average

latency (msec)

Instance1

23.8

3.8

Instance2

23.7

3.7

Instance3

24.0

3.3

Instance4

23.5

3.8

Instance5

23.7

3.8

Instance6

23.7

3.5

Instance7

23.7

3.7

Instance8

24.3

3.3

Return to top

Server Design Validation Results

The following sections summarize the server design validation results for the test cases.

Test Case: Normal Operating Conditions

The first test case represents peak workload during normal operating conditions. Normal
operating conditions refer to a state where all of the active and passive databases reside on the
servers they were planned to run on. Because this test case doesn't represent the worst case
workload, it isn't the key performance validation test. It provides a good indication of how this
environment should run outside of a server failure or maintenance event. In this case each
Mailbox server is running four active and four passive databases.
Validation of Expected Load
The message delivery rate verifies that tested workload matched the target workload. The actual
message delivery rate is slightly higher than target.

88

Counter

Target

Message Delivery Rate / Server 8.54

Tested result

8.63

Validation of Mailbox Servers

Processor
Processor utilization is low as expected.
Counter

Target

Tested result

Hyper-V Hypervisor Virtual

Processor\% Guest Run Time

<70%

54

Storage
The storage results look good. The average read latency for the active databases is 19.3 when
measured in the VM and 15.9 when measured on the EqualLogic storage array. As discussed in
"Server Validation: Performance and Health Criteria" earlier in this white paper, time-based
counters measured in a VM may not be accurate because the VM has a different concept of time
than the physical server. The difference between these counters is likely the result of a
combination of iSCSI network latency (generally <1 msec) and inaccurate counter values in the
VM.
Counter

Target

Tested result

MSExchange Database\I/O
Database Reads (Attached)
Average Latency

<20 msec

19.3

EqualLogic Average Disk Read

Latency

<20 msec

15.9

MSExchange Database\I/O
Database Writes (Attached)
Average Latency

<20 msec

6.8

EqualLogic Average Disk Write

Latency

<20 msec

2.5

Database\Database Page Fault

Stalls/sec

MSExchange Database\IO Log

Writes Average Latency

<20 msec

5.2

Database\Log Record Stalls/sec

MSExchange Database\I/O

<200 msec

23.7

<Reads average

89

Counter

Target

Tested result

MSExchange Database\I/O
Database Writes (Recovery)
Average Latency

<200 msec

7.6

MSExchange Database\IO Log

Read Average Latency

<20 0msec

7.5

Database Reads (Recovery)

Average Latency

Application Health
Exchange is healthy, and all of the counters used to determine application health are well under
target values.
Counter

Target

Tested result

MSExchangeIS\RPC Requests

<70

2.7

MSExchangeIS\RPC Averaged Latency

<10 msec

2.4

MSExchangeIS
Mailbox(_Total)\Messages Queued for
Submission

1.5

MSExchange
Replication(*)\CopyQueueLength

<1

0.1

MSExchange
Replication(*)\ReplayQueueLength

<5

2.1

Validation of Client Access and Hub Transport Servers

Processor
Processor utilization is low, as expected.
Counter

Target

Tested result

Hyper-V Hypervisor Virtual

Processor\% Guest Run Time

<70%

19

Storage
The storage results look good. The very low latencies should have no impact on message
transport.

90

Counter

Target

Tested result

Logical/Physical Disk(*)\Avg.
Disk sec/Read

<20 msec

0.012

Logical/Physical Disk(*)\Avg.
Disk sec/Write

<20 msec

0.012

Application Health
The low RPC Averaged Latency values confirm a healthy Client Access server with no impact on
client experience.
Counter

Target

Tested result

MSExchange
RpcClientAccess\RPC Averaged
Latency

<250 msec

MSExchange
RpcClientAccess\RPC Requests

<40

Hub Transport Server Health

The Transport Queue counters are all well under target, confirming that the Hub Transport server
is healthy and able to process and deliver the required messages.
Counter

Target

Tested result

\MSExchangeTransport
Queues(_total)\Aggregate Delivery
Queue Length (All Queues)

<3000

1.5

\MSExchangeTransport
Queues(_total)\Active Remote
Delivery Queue Length

<250

\MSExchangeTransport
Queues(_total)\Active Mailbox
Delivery Queue Length

<250

1.1

\MSExchangeTransport
Queues(_total)\Submission Queue
Length

<100

\MSExchangeTransport
Queues(_total)\Retry Mailbox
Delivery Queue Length

<100

0.4

91

Validation of Root Server Health

Processor
As expected, the processor utilization is very low and well under target thresholds.
Counter

Target

Tested result

Hyper-V Hypervisor Logical

Processor(_total)\% Guest Run
Time

<75%

42

Hyper-V Hypervisor Logical

Processor(_total)\% Hypervisor
Run Time

<5%

Hyper-V Hypervisor Logical

Processor(_total)\% Total Run
Time

<80%

44

Hyper-V Hypervisor Root Virtual

Processor(_total)\% Guest Run
Time

<5%

Application Health
The VM health summary counters indicate that all VMs are in a healthy state.
Counter

Target

Tested result

Hyper-V Virtual Machine Health

Summary\Health Critical

Test Case: Single Server Failure or Single Server Maintenance (In Site)
Validation of Expected Load
The message delivery rate verifies that tested workload matched the target workload. The actual
message delivery rate is slightly higher than target.
Counter

Target

Tested result

Message Delivery Rate /

Mailbox

17.08

17.3

Validation of Mailbox Servers

Processor
Processor utilization is low, as expected.

92

Counter

Target

Tested result

Hyper-V Hypervisor Virtual

Processor\% Guest Run Time

<70%

69

Storage
In this test case, the average read latency for the active databases is 26.2 when measured in the
VM and 16.2 when measured on the EqualLogic storage array. As discussed in "Server
Validation: Performance and Health Criteria" earlier in this white paper, time-based counters
measured in a VM may not be accurate because the VM has a different concept of time than the
physical server. The difference between these counters is likely the result of a combination of
iSCSI network latency (generally <1 msec) and inaccurate counter values in the VM. Because the
read latency measured on the EqualLogic array is less than 20, there's no concern about the
counter measured in the VM being over target.
Counter

Target

Tested result

MSExchange Database\I/O
Database Reads (Attached)
Average Latency

<20 msec

26.2

EqualLogic Average Disk Read

Latency

<20 msec

16.2

MSExchange Database\I/O
Database Writes (Attached)
Average Latency

<20 msec

7.4

EqualLogic Average Disk Write

Latency

<20 msec

2.1

Database\Database Page Fault

Stalls/sec

MSExchange Database\IO Log

Writes Average Latency

<20 msec

5.2

Database\Log Record Stalls/sec

MSExchange Database\I/O
Database Reads (Recovery)
Average Latency

<200 msec

Not applicable

MSExchange Database\I/O
Database Writes (Recovery)
Average Latency

<200 msec

Not applicable

MSExchange Database\IO Log

<200 msec

Not applicable

<Reads average

93

Counter

Target

Tested result

Read Average Latency

Application Health
Exchange is very healthy, and all of the counters used to determine application health are well
under target values.
Counter

Target

Tested result

MSExchangeIS\RPC Requests

<70

8.0

MSExchangeIS\RPC Averaged Latency

<10 msec

3.7

MSExchangeIS
Mailbox(_Total)\Messages Queued for
Submission

3.3

MSExchange
Replication(*)\CopyQueueLength

<1

Not applicable

MSExchange
Replication(*)\ReplayQueueLength

<5

Not applicable

Validation of Client Access and Hub Transport Servers

Processor
Processor utilization is low as expected.
Counter

Target

Tested result

Hyper-V Hypervisor Virtual

Processor\% Guest Run Time

<70%

26.3

Storage
The storage results look good. The very low latencies should have no impact on message
transport.
Counter

Target

Tested result

Logical/Physical Disk(*)\Avg.
Disk sec/Read

<20 msec

0.0041

Logical/Physical Disk(*)\Avg.
Disk sec/Write

<20 msec

0.0005

Application Health

94

The low RPC Averaged Latency values confirm a healthy Client Access server with no impact on
client experience.
Counter

Target

Tested result

MSExchange
RpcClientAccess\RPC Averaged
Latency

<250 msec

13.2

MSExchange
RpcClientAccess\RPC Requests

<40

6.1

The Transport Queue counters are all well under target, confirming that the Hub Transport server
is healthy and able to process and deliver the required messages.
Counter

Target

Tested result

\MSExchangeTransport
Queues(_total)\Aggregate Delivery
Queue Length (All Queues)

<3000

4.7

\MSExchangeTransport
Queues(_total)\Active Remote
Delivery Queue Length

<250

\MSExchangeTransport
Queues(_total)\Active Mailbox
Delivery Queue Length

<250

3.6

\MSExchangeTransport
Queues(_total)\Submission Queue
Length

<100

\MSExchangeTransport
Queues(_total)\Retry Mailbox
Delivery Queue Length

<100

1.1

Validation of Root Server Health

Processor
As expected the processor utilization is very low and well under target thresholds.
Counter

Target

Tested result

Hyper-V Hypervisor Logical

Processor(_total)\% Guest Run
Time

<75%

49.9

95

Counter

Target

Tested result

Hyper-V Hypervisor Logical

Processor(_total)\% Hypervisor
Run Time

<5%

1.3

Hyper-V Hypervisor Logical

Processor(_total)\% Total Run
Time

<80%

51.2

Hyper-V Hypervisor Root Virtual

Processor(_total)\% Guest Run
Time

<5%

3.6

Application Health
The VM health summary counters indicate that all VMs are in a healthy state.
Counter

Target

Tested result

Hyper-V Virtual Machine Health

Summary\Health Critical

Test Case: Site Failure

The message delivery rate verifies that tested workload matched the target workload. The actual
message delivery rate is slightly higher than target.
Validation of Expected Load
Counter

Target

Tested result

Message Delivery Rate /

Mailbox

17.08

17.4

Validation of Mailbox Servers

Processor
Processor utilization is low as expected.
Counter

Target

Tested result

Hyper-V Hypervisor Virtual

Processor\% Guest Run Time

<70%

64%

Storage

96

In this test case, the average read latency for the active databases is 24.0 when measured in the
VM and 15.9 when measured on the EqualLogic storage array. As discussed in "Server
Validation: Performance and Health Criteria" earlier in this white paper, time-based counters
measured in a VM may not be accurate because the VM has a different concept of time than the
physical server. The difference between these counters is likely the result of a combination of
iSCSI network latency (generally <1 msec) and inaccurate counter values in the VM. Because the
read latency measured on the EqualLogic array is less than 20, there's no concern about the
counter measured in the VM being over target.
Counter

Target

Tested result

MSExchange Database\I/O
Database Reads (Attached)
Average Latency

<20 msec

24.0

EqualLogic Average Disk Read

Latency

<20 msec

15.9

MSExchange Database\I/O
Database Writes (Attached)
Average Latency

<20 msec

7.2

EqualLogic Average Disk Write

Latency

<20 msec

2.0

Database\Database Page Fault

Stalls/sec

MSExchange Database\IO Log

Writes Average Latency

<20 msec

5.0

Database\Log Record Stalls/sec

MSExchange Database\I/O
Database Reads (Recovery)
Average Latency

<200 msec

Not applicable

MSExchange Database\I/O
Database Writes (Recovery)
Average Latency

<200 msec

Not applicable

MSExchange Database\IO Log

Read Average Latency

<200 msec

Not applicable

<Reads average

Application Health
Exchange is healthy, and all of the counters used to determine application health are well under
target values.

97

Counter

Target

Tested result

MSExchangeIS\RPC Requests

<70

7.8

MSExchangeIS\RPC Averaged Latency

<10 msec

3.5

MSExchangeIS
Mailbox(_Total)\Messages Queued for
Submission

3.0

MSExchange
Replication(*)\CopyQueueLength

<1

Not applicable

MSExchange
Replication(*)\ReplayQueueLength

<5

Not applicable

Validation of Client Access and Hub Transport Servers

Processor
Processor utilization is low as expected.
Counter

Target

Tested result

Hyper-V Hypervisor Virtual

Processor\% Guest Run Time

<70%

25

Storage
The storage results look good. The very low latencies should have no impact on message
transport.
Counter

Target

Tested result

Logical/Physical Disk(*)\Avg.
Disk sec/Read

<20 msec

0.003

Logical/Physical Disk(*)\Avg.
Disk sec/Write

<20 msec

0.001

Application Health
The low RPC Averaged Latency values confirm a healthy Client Access server with no impact on
client experience.
Counter

Target

Tested result

MSExchange
RpcClientAccess\RPC Averaged
Latency

<250 msec

13.0

98

Counter

Target

Tested result

MSExchange
RpcClientAccess\RPC Requests

<40

5.9

The Transport Queue counters are all well under target, confirming that the Hub Transport server
is healthy and able to process and deliver the required messages.
Counter

Target

Tested result

\MSExchangeTransport
Queues(_total)\Aggregate Delivery
Queue Length (All Queues)

<3000

4.2

\MSExchangeTransport
Queues(_total)\Active Remote
Delivery Queue Length

<250

\MSExchangeTransport
Queues(_total)\Active Mailbox
Delivery Queue Length

<250

3.4

\MSExchangeTransport
Queues(_total)\Submission Queue
Length

<100

\MSExchangeTransport
Queues(_total)\Retry Mailbox
Delivery Queue Length

<100

0.6

Validation of Root Server Health

Processor
As expected, the processor utilization is very low and well under target thresholds.
Counter

Target

Tested result

Hyper-V Hypervisor Logical

Processor(_total)\% Guest Run
Time

<75%

47.5

Hyper-V Hypervisor Logical

Processor(_total)\% Hypervisor
Run Time

<5%

1.2

Hyper-V Hypervisor Logical

Processor(_total)\% Total Run

<80%

48.7

99

Counter

Target

Tested result

<5%

3.5

Time
Hyper-V Hypervisor Root Virtual
Processor(_total)\% Guest Run
Time
Application Health
The VM health summary counters indicate that all VMs are in a healthy state.
Counter

Target

Tested result

Hyper-V Virtual Machine Health

Summary\Health Critical

Return to top

Conclusion

This white paper provides an example of how to design, test, and validate an Exchange 2010
solution for customer environments with 9,000 mailboxes deployed on Dell server and storage
solutions. The step-by-step methodology in this document walks through the important design
decision points that help address key challenges while ensuring that core business requirements
are met.
Return to top
For the complete Exchange 2010 documentation, see Exchange Server 2010.
For more information from Dell, see the following resources:

Dell and Exchange Server 2010

Dell Exchange 2010 Advisor Tool

PowerEdge Servers

Dell Data Storage & Backup

Dell Power Solutions Article: Optimizing Microsoft Exchange Server 2010 Deployments on
Dell Servers and Storage

Dell Exchange 2010 ROI Study:

http://www.dell.com/downloads/global/products/pedge/en/ROI_exchange.pdf

Dell Enterprise Technology Center

For more information from F5, see the following resources:

Con

100

F5 Exchange Solution Home Page

F5 Exchange 2010 Deployment Guide: http://www.f5.com/pdf/deployment-guides/f5exchange-2010-dg.pdf

F5 Exchange Solution Brief: http://www.f5.com/pdf/application-ready-networkguides/microsoft-exchange-2010-arng.pdf

If you need more information, contact the F5 Microsoft Partnership Team at

[email protected].
This document is provided "as-is." Information and views expressed in this document, including
URL and other Internet Web site references, may change without notice. You bear the risk of
using it.
This document does not provide you with any legal rights to any intellectual property in any
Microsoft product. You may copy and use this document for your internal, reference purposes.
Return to top

101