Why Storage Networks Can Be Better Than Direct Attached Storage for Exchange
Guest Post By: Brien M. Posey
Of all the decisions that must be made when planning an Exchange Server deployment, perhaps none are as critical as deciding which type of storage to use. Exchange Server is very flexible with regard to the types of storage that it supports. However, some types of storage offer better performance and reliability than others.
When it comes to larger scale Exchange Server deployments, it is often better to use a Storage Area Network than it is to use direct attached storage. Storage networks provide a number of advantages over local storage in terms of costs, reliability, performance, and functionality.
Storage Area Networks have gained something of a reputation for being more expensive than other types of storage. However, if your organization already has a Storage Area Network in place then you may find that the cost per gigabyte of Exchange Server storage is less expensive on your storage network than it would be if you were to invest in local storage.
While this statement might seem completely counter intuitive, it is based on the idea that physical hardware is often grossly underutilized. To put this into prospective, consider the process of purchasing an Exchange mailbox server that uses Direct Attached Storage.
Organizations that choose to use local storage must estimate the amount of storage that will be needed to accommodate Exchange Server databases plus leave room for future growth. This means making a significant investment in storage hardware. In doing so, an organization is purchasing the necessary storage space, but they may also be spending money for storage space that is not immediately needed.
In contrast, Exchange servers that are connected to storage networks can take advantage of thin provisioning. This means that the Exchange Server only uses the storage space that it needs. When a thinly provisioned volume is created, the volume typically consumes less than 1 GB of physical storage space, regardless of the volume’s logical size. The volume will consume physical storage space on an as needed basis as data is written to the volume.
In essence, a thinly provisioned volume residing on a SAN could be thought of as “pay as you go” storage. Unlike Direct Attached Storage, the organization is not forced to make a large up-front investment in dedicated storage that may never be used.
Another advantage to using Storage Area Networks for Exchange Server storage is that when properly constructed, SANs are far more reliable than Direct Attached Storage.
The problem with using Direct Attached Storage is that there are a number of ways in which the storage can become a single point of failure. For example, a disk controller failure can easily corrupt an entire storage array. Although there are servers that have multiple array controllers for Direct Attached Storage, lowering servers are often limited to a single array controller.
Some Exchange mailbox servers implement Direct Attached Storage through an external storage array. Such an array is considered to be a local component, but makes use of an external case as a way of compensating for the lack of drive bays within the server itself. In these types of configurations, the connectivity between the server and external storage array can become a single point of failure (depending on the hardware configuration that is used).
When SAN storage is used, potential single points of failure can be eliminated through the use of multipath I/O. The basic idea behind multipath I/O is that fault tolerance can be achieved by providing multiple physical paths between a server and a storage device. If for example an organization wanted to establish fault tolerant connectivity between an Exchange Server and SAN storage, they could install multiple Fibre Channel Host Bus Adapters into the Exchange Server. Each Host Bus Adapter could be connected to a separate Fibre Channel switch. Each switch could in turn provide a path to mirrored storage arrays. This approach prevents any of the storage components from becoming single points of failure.
Although Microsoft has taken measures to drive down mailbox server I/O requirements in the last couple of versions of Exchange Server, mailbox databases still tend to be I/O intensive. As such, large mailbox servers depend on high-performance hardware.
While there is no denying the fact that high-performance Direct Attached Storage is available, SAN storage can potentially provide a higher level of performance sue to its scalability. One of the major factors that impacts a storage array’s performance is the number of spindles that are used by the array. Direct Attached Storage limits the total number of spindles that can be used. Not only are the number of drive bays in the case a factor, but there is also a limit to the number of disks that can be attached to the array controller.
SAN environments make it possible to create high performance disk arrays by using large numbers of physical disks. Of course capitalizing on the disk I/O performance also means that you must have a high speed connection between the server and the SAN, but this usually isn’t a problem. Multipath I/O allows storage traffic to be distributed across multiple Fibre Channel ports for optimal performance.
Finally, SAN environments are ideal for use in virtualized datacenters. Although neither Microsoft nor VMware still require the use of shared storage for clustered virtualization hosts, using shared storage is still widely considered to be a best practice. SANs make it easy to create cluster shared volumes that can be shared among the nodes in your host virtualization cluster.
Exchange mailbox servers are almost always considered to be mission critical. As such, it makes sense to invest in SAN storage for your Exchange Server since it can deliver better performance and reliability than is possible with Direct Attached Storage.