Applies to: Exchange Server 2010 SP3, Exchange Server 2010 SP2
Topic Last Modified: 2011-02-10
To provide continuous access to e-mail and voice mail, you must correctly plan and implement a high performance, highly available, and scalable solution for Microsoft Exchange Server 2010. To implement a high performance voice mail solution that's highly available and scalable, you must understand the system requirements for Exchange Unified Messaging (UM) in Exchange 2010. This will help you to select the UM servers and other voice mail components that will have the greatest effect on system performance, scalability, and availability for your deployment.
As you plan your deployment, you can use this topic and other topics in this section to help design your Unified Messaging environment. For details, see Exchange 2010 System Requirements.
Unified Messaging System Hardware Overview
Choosing the correct system hardware is one of the most important steps when in planning and implementing voice mail solution. You must ensure that the hardware you choose will meet or even exceed the system requirements. You must consider the following when you plan and deploy Unified Messaging 2010:
- How to tune your high-performing UM servers for optimum
- How to make your Unified Messaging deployment scalable
- How to make your Unified Messaging deployment highly
Tuning a Unified Messaging server for optimum performance is an ongoing process. You must understand all the variables that affect your system, including user profile, architecture, and the hardware that's used on a Unified Messaging server. When you know your requirements for a Unified Messaging server, you can establish baseline metrics for the server and make adjustments to improve overall system performance.
Generally, the maximum level of performance for a Unified Messaging server is determined by the component that has the lowest level of performance, which can cause a bottleneck in the system. The key to improving performance is learning to identify bottlenecks in your Unified Messaging system, determine the cause of the system bottlenecks, and then take the appropriate action or actions.
There are several tools you can use to measure performance of Exchange 2010 Unified Messaging, including Jetstress and Load Generator (LoadGen). For details, see Tools for Performance and Scalability Evaluation. The Windows Server 2008 operating system also includes some general performance tools, including Windows Performance Monitor.
Also, you should analyze your current loads to establish minimum server requirements. One of the biggest challenges in determining minimum server requirements is understanding how your users use your voice mail system. After you determine your hardware requirements, you should conduct a pilot test to make sure the Unified Messaging server performance levels are acceptable.
To provide a highly available UM system, you must protect against hardware failures for the following:
- UM servers Unified Messaging runs as a
service and a worker process. This means that if the service is
using lots of system resources or has become unresponsive, the
worker process can be recycled to bring the Unified Messaging
service back online. Having the UM worker process recycle allows
for a more highly available system and prevents the Unified
Messaging service from being down for extended periods of time. The
UM worker process is responsible for dealing with outages in
Mailbox servers, Hub Transport servers, and domain controllers. For
example, if a Mailbox server for a user is unavailable, the UM
server will continue to accept calls on behalf of the user.
However, the user's custom greeting won't be played. Instead, a
standard greeting will be used for calls to that user.
Additionally, if the Hub Transport server is unavailable, the UM
server will continue to accept calls and queue the calls, depending
on how you've set up the queuing limit, until the Hub Transport
server is available. In a situation where all domain controllers
are unavailable, the UM server will be unable to accept calls.
- IP gateways If you want to create
redundancy to help ensure that IP gateways are always available,
add multiple IP gateways in an N+1 configuration. For example, if
you need two IP gateways, install and set up an additional IP
gateway so that you have an IP gateway to take the place of an IP
gateway that fails or must be taken offline.
For details, see Understanding Unified Messaging Availability.
Minimum requirements for uptime may vary among organizations. But every organization wants to achieve a high level of uptime, especially for their telephony system. An organization's telephony system is frequently business-critical and must be highly available to users. One of the factors you need to consider when you deploy a UM system is the ability of the system to provide service to users when a key component, for example, a UM server or an IP gateway, becomes unavailable.
Scalability is defined as the capability to increase resources to increase the capacity of a given service. There are two types of scalability that can be used to increase the capacity of UM servers in your organization: horizontal and vertical. In Unified Messaging, when you scale vertically, you add hardware resources to a single UM server or multiple UM servers by, for example:
- Adding more hard disk space for message storage.
- Increasing the speed or number of processors.
- Increasing the amount or speed of RAM.
- Increasing the number of network adapters or the number of
local area network ports in a single network adapter.
For details, see Understanding Unified Messaging Performance and Scalability.
When selecting hardware for your Unified Messaging servers, you must consider several components, including disk storage, memory, processor, and network hardware. The hardware that you select for your Exchange Unified Messaging deployment will have an effect on performance. Because many variables affect performance, including the hardware, it's difficult to predict the effect that any particular hardware component will have.
When you're selecting the hardware components for a Unified Messaging server, consider the following definitions:
- Minimum This is the minimum processor
and memory configuration that's suitable for specific Exchange 2010
server roles. Minimum hardware requirements must be met before you
can receive help from Microsoft Customer Service and Support.
- Maximum This is the maximum recommended
processor and memory configuration for specific Exchange 2010
server roles. Maximum is defined as the upper limit of the
processor and memory configurations for Exchange 2010, based on the
price and performance. Maximum is a guideline and not a support
criterion. It doesn't take into account the resource requirements
of third-party applications. The recommended maximum may change
over time, based on price changes and technology advancements.
- Recommended This is the recommended
processor and memory configuration for specific Exchange 2010
server roles. Recommended can be defined as the best configuration
based on price and performance. The recommended configuration also
provides a balance between processor and memory capacity. The goal
is to match the memory configuration to the processor configuration
so that the system will use the processors effectively without
causing a bottleneck in memory or vice versa.
Disk Storage Requirements
Here are a few helpful guidelines for selecting a storage configuration that provides good performance and capacity for Exchange 2010. Considerations of capacity and performance are often at odds with each other when it comes to selecting a storage solution, and you must consider both before you make a purchase. Generally, you should make sure that:
- There will be enough space to store all the data. Determining
your capacity needs is a relatively straightforward process.
- The solution provides acceptable disk latency and a responsive
user experience. You determine this by measuring or predicting
transactional input/output (I/O) delivered by the solution.
- Non-transactional I/O has both enough time to complete and
enough disk throughput to meet your service-level agreements.
The goal is to balance these factors so that you can design the hardware solution for your servers.
UM doesn’t really require much storage because each message is converted and forwarded as soon as a Hub Transport server is available. However, for Unified Messaging, each UM server requires storage for the following:
- Any UM language packs that are installed along with the en-US
UM language pack
- Custom audio prompts for UM dial plans and auto attendants that
are stored in a system mailbox
- UM messages that are queued when Hub Transport servers are
unavailable, such as:
- Missed call notifications
- E-mail messages when a user has been UM-enabled
- Voice mail messages with an audio attachment
- Fax messages with an attachment
Note: If no Hub Transport servers are available, a Unified Messaging server will queue 100 voice messages before shutting down the Microsoft Exchange Unified Messaging service.
- Missed call notifications
To determine the maximum amount of storage required to accommodate these language packs, audio prompts, and queued messages, take the maximum message size of a voice message, based on the audio codec that's used and the message length, and then multiply it by 1,000 bytes. For example, if an average voice message is approximately 30 seconds, which is usually under 100 KB in size, take that number times 100 voice mails.
When you select hardware for Exchange 2010, we recommend that you consider the maximum memory limits of the server. Different server architectures have different memory limits. We recommend that you check the following technical specifications of the server to determine the most cost-efficient maximum memory limits for your servers:
- Memory speed Some server architectures
require slower memory modules to scale to the maximum supported
amount of memory. For example, maximum server memory could be
limited to 32 GB with PC3 10666 (DDR3 1333) or 128 GB
using PC2 6400 (DDR2 800). Check with the manufacturer to ensure
that the memory configuration target for Exchange 2010 is
compatible in terms of speed.
- Memory module size Consider the largest
memory module size that the server will support. Generally, the
larger the memory module, the more expensive it is. For example,
two 2 GB DDR SDRAM memory modules generally cost much less
than one 4 GB DDR SDRAM memory module, and two 4 GB DDR
SDRAM memory modules generally cost much less than one 8 GB
DDR SDRAM memory module. Make sure the size of the maximum memory
module allows you to meet your target memory requirements for
- Total number of memory slots Consider
how many memory modules a specific server will support. The total
number of slots multiplied by the maximum memory module size
provides the maximum memory for the server. Remember that memory
modules must sometimes be installed in pairs.
Thus, the recommended minimum of 4 GB of memory installed with a recommended maximum of 2 GB per processor which would equal 4 GB minimum. However, there are several other factors that must be considered when you determine the amount of memory needed for each Unified Messaging server. These include:
- The size of the global address list or the combined size of all
- The UM language packs that are installed and available.
- Whether Voice Mail Preview is enabled or disabled.
- The number of incoming calls.
The following table shows the minimum supported and recommended maximum memory configurations for Exchange 2010.
|Number of users||Recommended memory per UM server|
After the required number of processors has been estimated for a specific server role, baseline memory recommendations can be applied. Exchange 2010 on the 64-bit edition of the Windows Server 2008 operating system can efficiently use more than 64 GB of memory.
With effective planning and an understanding of the basic processor and memory requirements for specific Exchange 2010 server roles, you can create a balanced and cost-effective topology.
For more information about how different memory configurations perform, see Understanding Memory Configurations and Exchange Performance.
There are significant benefits to be gained by running Exchange 2010 on multiple processors. The performance benefit depends upon the specific processor that's used. Check with your server hardware vendor to see whether the benefits of processors are Exchange-specific for a given hardware architecture.
The processor on a server should maintain a load of about 60 percent during peak working hours. This percentage level allows for periods of extreme load. If the processor usage is consistently greater than 75 percent, processor performance is considered a bottleneck.
There are several ways that server CPU can affect performance. These include:
- The processor clock speed, measured in megahertz (MHz) or
- The number of processors used
- The type of processors used
- Whether Voice Mail Preview is enabled or disabled
Exchange can make full use of multiple processors, so using servers with more processors improves performance. However, the relationship between the number of processors, the number of processor cores, and performance is complex. The optimum number of processors and cores is partly determined by the Exchange server roles deployed on the server and whether the Unified Messaging server role is also deployed on the same physical server as other server roles.
Much of the network interface subsystem is tuned automatically. Server-based network adapters are capable of detecting the type and level of traffic that passes through the network interface, and they self-tune to reflect this information. We recommend that you ensure that the latest device drivers are maintained on the server.
Every incoming call that's received from an IP gateway will generate IP-based network traffic and consume some of your available network bandwidth. Before you deploy Unified Messaging, you should perform an analysis of the network traffic to determine current usage patterns and find any potential issues. On most networks, bandwidth demand isn't evenly distributed throughout business hours. Because all the IP-based calls are routed directly to your UM servers from the IP gateways on your network, and because this IP-based network traffic consumes some available bandwidth, you should follow these recommendations and guidelines:
- Place your PBXs physically close to your IP gateways.
- Place your IP gateways and your UM servers on the same
well-connected network or within the same physical site.
- Place your UM servers on the same well-connected network or
within the same physical site as other computers that have Exchange
2010 server roles installed, including Mailbox, Hub Transport, and
Client Access servers.
- End your wide area network (WAN) connections close to where
your telephony equipment is located.
- In branch office scenarios or over WAN connections, use the
G.723.1 codec instead of the G.711u or G.711A codec to minimize the
network traffic that's passed between your IP gateways and your UM
The network requirements or recommendations for UM servers are as follows:
- UM servers using gigabit (1,000 megabits per second
(Mbps)) or 1 gigabit per second (Gbps) Ethernet adapters.
- UM servers connected to multiple-switched, fast Ethernet
networks of gigabit Ethernet connections.
- A round trip time from the IP gateway or IP PBX of less than