Before you deploy Exchange Server 2013, you should carefully plan
the messaging architecture. As part of your implementation planning,
you need to look closely at preinstallation requirements and the
hardware you will use. Exchange Server is a complex messaging platform
with many components that work together to provide a comprehensive
solution for routing, delivering, and accessing email messages,
voice-mail messages, faxes, contacts, and calendar information.
Successful Exchange Server administration depends on three things:
If you’re using Exchange Online, Microsoft
provides the hardware. Otherwise, for on-premises implementations,
Exchange Server 2013 should run on a system with adequate memory,
processing speed, and disk space. You also need an appropriate
data-protection and system-protection plan at the hardware level.
Exchange
Server 2013 requires two different types of server hardware. You want
to select hardware for Mailbox servers with scaling up in mind while
selecting hardware for Client Access servers with scaling out in mind.
Scaling up typically means adding additional or faster, better CPUs and
memory to existing servers to meet capacity needs. Scaling out
typically means adding additional servers to meet capacity needs.
Key guidelines for choosing hardware for Exchange Server are as follows:
-
Memory
. The minimum random access memory (RAM) is 8 gigabytes
(GB) for servers with both the Mailbox Server and Client Access Server
roles, 8 GB for Mailbox servers, and 4 GB for Client Access servers. In
most cases, you’ll want to have at least twice the recommended minimum
amount of memory. The primary reason for this is performance. Most of
the Mailbox server installations I run use 16 GB of RAM as a starting
point, even in small installations. In multiple Exchange server
installations, the Mailbox server should have at least 2 GB of RAM plus
5 megabytes (MB) of RAM per mailbox (with a minimum of 8 GB
regardless). For all Exchange server configurations, the paging file
should be at least equal to the amount of RAM in the server plus 10 MB. -
CPU
. Exchange Server 2013 runs on the x64 family of
processors from AMD and Intel, including AMD64 and Intel 64. You can
achieve significant performance improvements with a high level of
processor cache. Look closely at the L1, L2, and L3 cache options
available—a higher cache can yield much better performance overall.
Look also at the speed of the front-side bus. The faster the bus speed,
the faster the CPU can access memory.
Exchange Server 2013 runs only on 64-bit hardware. The primary
advantages of 64-bit processors over 32-bit processors are related to
memory limitations and data access. Because 64-bit processors can
address more than 4 GB of memory at a time without physical address
extension, they can store greater amounts of data in main memory,
providing direct access to and faster processing of data. In addition,
64-bit processors can process data and execute instruction sets that
are twice as large as 32-bit processors. Accessing 64 bits of data
(versus 32 bits) offers a significant advantage when processing complex
calculations that require a high level of precision. -
SMP
. Exchange Server 2013 supports symmetric
multiprocessors, and you’ll see significant performance improvements if
you use multiple CPUs—not just multiple cores in a single CPU. Although
the clock speed of the CPU is important, so are the number of logical
processor cores and the number of threads that can be simultaneously
processed. That said, if Exchange Server is supporting a small
organization with a single domain, one CPU with multiple cores may be
enough. If the server supports a medium or large organization or
handles mail for multiple domains, you will want to consider adding
processors. When it comes to processor cores, I prefer two multicore
processors to a single processor with the same number of cores, given
current price and performance tradeoffs. An alternative is to
distribute the workload across different servers based on where you
locate resources. -
Disk drives
. The
data storage capacity you need depends entirely on the number and size
of the data that will pass through, be journaled on, or stored on the
Exchange server. You need enough disk space to store all data and logs,
plus workspace, system files, and virtual memory. Input/output (I/O)
throughput is just as important as drive capacity. Rather than use one
large drive, you should use several drives, which allows you to
configure fault tolerance with RAID. As part of your hardware planning,
it’s important to point out that Exchange 2013 supports multiple
databases on the same volume, allowing you to have a mix of active and
passive copies on a single volume. Keep in mind, however, the
input/output per second (IOPS) capabilities for the underlying physical
disks. Also note that even if you’ve been assigned multiple logical
unit numbers (LUNs) for use from storage these different LUNs may be
spread over the same physical disks. -
Data protection
. You can add protection against unexpected drive
failures by using redundant storage. For the boot and system disks, use
RAID 1 on internal drives. However, because of the new
high-availability features, you might not want to use software RAID for
Exchange data and logs. You also might not want to use expensive disk
storage systems either. Instead, deploy multiple Exchange servers with
the required server roles.
If you decide to use software-based redundant storage, you can use
disk striping without parity or disk striping with parity for data
volumes. Disk striping without parity offers good read/write
performance, but a failed drive means that Exchange Server can’t
continue operation on an affected database until the drive is replaced
and data is restored from backup. Disk mirroring creates duplicate
copies of data on separate drives; you can rebuild a mirrored unit to
restore full operations and can continue operations if one of the
drives fails. Disk striping with parity offers good protection against
single drive failure, but it has poor write performance. For best
performance and fault tolerance, RAID 10 (also referred to as RAID 0 +
1), which consists of disk mirroring and disk striping without parity,
is also an option. -
Uninterruptible power supply
. Exchange Server 2013 is designed to maintain database
integrity at all times and can recover information using transaction
logs. This doesn’t protect the server hardware, however, from sudden
power loss or power spikes, both of which can seriously damage
hardware. To prevent this, connect your server to an uninterruptible
power supply (UPS). A UPS gives you time to shut down the server or
servers properly in the event of a power outage. Proper shutdown is
especially important on servers using write-back caching controllers.
These controllers temporarily store data in cache. Without proper
shutdown, this data can be lost before it is written to disk. To
prevent data loss, write-back caching controllers typically have
batteries that help ensure that changes can be written to disk after
the system comes back online.
If
you follow these hardware guidelines and modify them for specific
messaging roles, as discussed in the next section, you’ll be well on
your way to success with Exchange Server 2013.
Real World
Mirroring can be implemented with software RAID 1 on Windows Server.
As software-based RAID is implemented using dynamic disks, it’s
important to note that beginning with Windows Server 2012 dynamic disks
are being phased out in favor of Storage Spaces. However, for mirroring
boot and system volumes on internal disks, Microsoft recommends
continuing to use dynamic disks and RAID 1.
If you decide to use software-based redundant storage, remember that
storage arrays typically already have an underlying redundant storage
configuration and you might have to use a storage array–specific tool
to help you distinguish between LUNs and the underlying physical disks.
Herein, I focus on software-based redundancy implemented with RAID or
Storage Spaces rather than the underlying hardware redundancy
implemented in storage arrays.
Windows Server is transitioning to standards-based storage beginning
with Windows Server 2012. This transition means several popular tools
and favored features are being phased out. Officially, a tool or
feature that is being phased out is referred to as deprecated.
When Microsoft deprecates a tool or feature, it might not be in future
releases of the operating system (while continuing to be available in
current releases). Rather than not cover popular tools and features,
I’ve chosen to discuss what is actually available in the current
operating system, including both favored standbys and new options. One
of these new options is Storage Spaces. With Storage Spaces:
-
Simple volumes can stretch across multiple disks, similar to disk striping with parity (RAID 0). -
Mirrored volumes are mirrored across multiple disks. Although this
is similar to disk mirroring (RAID 1), it is more sophisticated in that
data is mirrored onto two or three disks at a time. If a storage space
has two or three disks, you are fully protected against a single disk
failure, and if a storage space has five or more disks, you are fully
protected against two simultaneous disk failures. -
Parity volumes use disk striping with parity. Although this
is similar to RAID 5, it is more sophisticated in that there are more
protections and efficiencies.
|