This research explores the definition, heritage, adoption and value of storage virtualization. It also explores the role played by external (heterogeneous) storage virtualization in the market today.

• The virtualization of storage dates from the 1980s, when the Small Computer System Interface (SCSI) standard was ratified, and is widely deployed today in disk storage, with storage arrays providing the most popular vehicles for the technology.
• Storage networks would not be possible without storage virtualization. For the future, its most important contribution may be resource mobility under software control.
• External (heterogeneous) storage virtualization has been in the market for over 10 years and is available from a variety of vendors, but has shown only limited success in the market. It creates the same proprietary vendor lock-in as does a conventional storage array, even though choice is available for the back-end disk.
• External (heterogeneous) storage virtualization negates the value-added software capabilities of the back-end disk arrays.
• Much of the virtualization capability touted for the external storage virtualization appliances can be achieved with a well-appointed midrange or high-end disk array, and often at a lower cost.

• Use business requirements and storage strategy to drive the selection of storage virtualization functions and characteristics.
• Compare the overall outcomes from alternative scenarios involving nonexternal virtualization controllers when considering external, or heterogeneous, storage virtualization solutions.
• If deploying external (heterogeneous) storage virtualization solutions, select low-end, simple-to-configure, low-maintenance redundant array of independent disks (RAID) devices as the building blocks for back-end disks to avoid unnecessary expenditure.

Through at least 2015, disk storage arrays will continue to be the primary source for storage virtualization.

Through at least 2015, penetration of external storage virtualization devices will rise, but will not exceed 5% of the disk storage arrays shipped.

Virtualization creates an abstraction layer between a physical object and the managing or using application. Not all virtualization techniques should be viewed from the perspective of server or hypervisor virtualization. Storage virtualization refers to the presentation of a simple file, logical volume or other storage object (like a disk drive) to an application in such a way that enables the physical complexity of the underlying storage to be hidden from the storage administrator and the application. Examples of this in common use include RAID, volume management, virtual memory, file systems and virtual tape.

The focus of this research is block-level storage virtualization. We note that the term "virtualization" is also active in the marketplace in the context of server virtualization, which is a form of application partitioning of individual servers and is different from the virtualization that occurs in storage. What both of these forms of virtualization have in common is the enablement of resource mobility under the control of software.

By accident and good luck, the storage industry implemented a de facto standard for virtualization as a part of the SCSI standard ratified in 1986. That de facto standard, the SCSI Logical Unit Number (LUN), was originally intended to enable up to eight devices on one bus. By the late 1980s, virtually every operating system had written its storage stack to support the SCSI command set and the use of the SCSI LUN. As a result, when the industry began to adopt RAID, volume management, caching and various forms of replication, it was possible to do so with few, if any, modifications to the software stack in the server. This was, and still is, possible because, regardless of how much complexity is hidden by the storage array, storage network or host bus adapter (HBA), the result for the server is a SCSI LUN that appears to the server software stack as a simple local disk.

It is hard to imagine any storage network (storage area network [SAN] or network-attached storage [NAS]) without virtualization. The process enables the physical storage in this environment to be shared across multiple servers, and enables physical devices behind the virtualization layer to be viewed and managed as if they were one large storage pool with no physical boundaries. To facilitate the sharing in manually provisioned arrays, the physical volumes (RAID sets) in the arrays are broken up into smaller volumes (volume management) to match the individual needs of the aggregated group of servers. The virtualization flexibility provided by a storage disk array or appliance in a storage network is, therefore, a key attribute that enables effective provisioning for servers. It also allows maintenance and upgrades of physical storage to be hidden from the servers.

Two other virtualization features are important when working with a SAN. The first is the ability to mask or hide volumes from servers that are not authorized to access those volumes, and the second is the ability to bind volumes to particular servers; both together provide an additional level of security in the SAN, which can be reinforced by zoning and port binding in Fibre Channel switches.

• Most of today's disk storage arrays go beyond the basics of RAID and volume management to support the ability to expand RAID set capacity in one form or another without hardware. A more recent and even better capability is thin provisioning, which assigns a very large logical volume to a server, but only provisions physical storage to the level required by the application. More physical storage is then consumed as the application expands its use of the large volume. This technique eliminates over-provisioning, improving utilization, and hides the physical expansion from the application and eliminates administrator intervention in growing the size of the volume.
• Snapshot copies increasingly are used for rapid recovery, and to drive test and development activities. These snapshots appear to the application server as SCSI LUNs, but of course are frozen to the time at which the snapshot was taken.
• Remote replication is key to many disaster recovery deployments. These replicas differ by geography, but again appear as SCSI LUNs to the servers to which they are presented. Some arrays can now automate storage tiering in a manner that is totally invisible to the server, which allows storage teams to lower costs without impacting performance or other operational metrics.

Within a storage network environment, there are three general locations where LUN virtualization can be employed: inside the host or application server, inside a disk array or inside an appliance within the storage network (see Table 1).

• Host-based volume managers are the oldest storage virtualization tools, and, in general, operate only for the benefit of the individual server and are not aware of the broader storage network environment and other servers. Therefore, they are often redundant with volume managers in the storage network. These forms of host-based virtualization are most effective when the storage is directly attached to the server (direct-attached storage [DAS]), and not in a storage network. But server virtualization has breathed new life into host-based volume management because of the importance of VMware's Virtual Machine File System (VMFS) in setting up storage for server virtual machines (VMs). And third parties are providing storage virtualization functions as add-ins to VMware and Hyper-V.
• Disk arrays have evolved to become the primary location for RAID sets and volume virtualization, especially within the storage network. In this model, the disk array holds the data and is, therefore, in a position to easily perform data-oriented actions, like snapshots or migrations, without having to move the data to an intermediate location. Its weakness is that the logical volumes it creates and manages must physically reside within the array. This means that when the disk array is full, any volumes that need to be expanded must be physically moved to another array. Grid or scale-out architectures, which create virtual arrays by combining multiple nodes, have eased this restriction by providing more scale.
• A SAN appliance, which is essentially a server dedicated to and residing in the SAN, or perhaps on a switch blade, is another evolution in the storage virtualization story. It has the advantage of being able to see multiple arrays and hosts, but the disadvantage of not being near either the data or the application. To manipulate data, the appliance first has to retrieve the data from the storage device. Then it does the work and returns the data to the storage device. Similarly, the appliance is not able to synchronize with the needs of the application without information found only in the application server. This model has gained some traction in the market, but its penetration remains small compared with host and disk storage array virtualization.

Table 1. Pros and Cons of Virtualization Implementations

Location	Advantages	Disadvantages
Host-based	Close to the application Can span storage arrays	Typically aware of only the local server and its storage
Array controller-based	Close to data Can span hosts	Typically can't span multiple arrays
Switch-/appliance-based	Can span storage arrays and hosts	Close to nothing: switch failure isolates data from control
Source: Gartner (June 2010)

Storage virtualization is widely deployed – in everybody's server, disk storage array and network – and, today, most vendors of midrange and high-end disk storage arrays market their features as virtualization capabilities. In the late 1990s, however, vendors did not use the word "virtualization" in marketing. Instead, they called out the various types of virtualization by functional names, such as RAID and volume management.

By 2000, a small group of mostly array-startup companies began to use the word "virtualization" as the primary descriptor for their products, causing confusion about the meaning of storage virtualization. What set some of these companies apart from traditional disk array vendors is that, for the most part, they did not want to be in the hardware business. Rather, they developed the software to make a storage controller, and used channel or OEM partners to load this software on a server to produce a controller and to mate the controller with disks to make a complete storage array. In other words, these products allowed a connection to an external disk, which is chosen by the user or the channel.

Today, controller/array products that support external disks are marketed by both large and small companies as "external disk storage virtualization," "heterogeneous storage virtualization" or sometimes just as "storage virtualization controllers." The result is that, when users talk about storage virtualization products, it isn't clear whether they are talking about external disk support or not.

Many external storage virtualization products are sold today. Some of the most successful include Hitachi Data Systems Universal Volume Manager (UVM) – a software license for the Universal Storage Platform (USP)-V/USP-VM series, IBM SAN Volume Controller (SVC), NetApp V-Series, EMC RecoverPoint and DataCore Software SANsymphony. Others include, but aren't limited to, EMC Invista, FalconStor Software IPStor and HP SAN Virtualization Services Platform (SVSP). For a discussion of successful use cases in the market, read the companion research, "External (Heterogeneous) Storage Virtualization: Successful Use Cases."

These products can virtualize externally connected disk storage arrays from heterogeneous vendors into their SAN environments. Some also simultaneously virtualize internal disk. With the exception of the two EMC devices, which are intended as complements to the arrays under management, external storage virtualization devices become the user's primary storage controllers and, therefore, the platform on which value-added storage virtualization software executes, as well as the platform that provides all the storage connections to application servers connected to the SAN. What users often miss is that the value-added software capabilities of the back-end arrays are disabled in this scenario, reducing their function to disk volumes protected by RAID. As a result, any investment made in software for the back end, or for hardware resources like processors and memory to run that software, are written off. The external storage virtualization product requires these investments to be made all over again. This leads to the obvious conclusion that users should not put full-featured disk arrays behind virtualization appliances, because it is a waste of money. Instead, they should buy low-cost, basic RAID disk arrays for this purpose.

Using full-featured disk arrays behind the virtualization appliance is justified when:

• The disk arrays exist in the storage environment and have a moderate amount of useful life left in them.
• The data must be migrated from the array.
• The disk array also has servers directly attached to it.

At the end of their useful life, however, these full-featured arrays should be replaced by those inexpensive, feature-poor and low operator/administrator maintenance disk arrays, rather than another full-featured disk array. This is a consolidation or transition strategy that virtualizes multiple storage arrays behind a single manufacturer's external storage virtualization controller(s) with the result that a single vendor lock-in is created around the storage virtualization appliance.

When deploying an external, or heterogeneous, virtualization product in new infrastructures, it should come with the same degree of scalability and enterprise-class properties as the products it front ends. In performance-sensitive environments, for example, administrators should be aware that, without enough cache, latency may be added to the input/output (I/O) traffic. The converse situation is that external storage devices may be able to add addition cache and other hardware resources. Firmware needs to be of the same maturity and reliability as on the back-end disk arrays being replaced. If the storage appliance has write cache, then it needs to have the same sophistication in algorithms and protection (batteries and mirrors, for example) as the disk arrays behind it do. Otherwise, the virtualization appliance may reduce data integrity and reliability. Also, although a virtualization appliance may support many interconnects, it is possible that the internal processing power of the virtualization appliance will be the limiting factor on throughput. As a result, the virtualization appliance could be the bottleneck in configurations supporting multiple disk arrays behind one virtualization appliance, leading to performance degradation across multiple storage devices. Potential customers should ask the following:

• Is the virtualization appliance explicitly supported by SAN management software?
• Is the virtualization appliance properly integrated with each element of the server software stack, including path failover software and synchronization for snapshots?
• Does the proposed disk component have management tools, and, if so, are they integrated and supported by the virtualization appliance and other software?
• Does the virtualization appliance vendor have the resources to do problem replication, problem determination, problem management and problem resolution?

Since their initial release in 1999-2000, Gartner estimates that approximately 20,000 external storage virtualization systems have been sold. This sound like a big number until you realize that, in the past five years, users have bought and deployed over 1.2 million disk storage arrays, making the overall market penetration between 1.5% and 2%. For products that have been on the market for more than 10 years, this is very limited success. The reasons for this vary depending on whether you ask a vendor or an end user, but the bottom line is that much of the virtualization capability touted for virtualization appliances can be achieved with a well-appointed midrange or high-end disk array, and often at lower cost. In addition, while messages such as "unify storage management," "negotiate the best price for storage" and "buy your storage from any vendor" are all true, the reality of the scenario is often very different from user expectations.

Finally, the hidden part of the value proposition is that it may break one proprietary relationship only to replace it with another lock-in just as strong, or even stronger, as the one it replaced.

Gartner recommends that organizations considering deploying external disk virtualization appliances should build alternative scenarios for storage infrastructure that rely on functionally rich storage arrays and migration tools, and then compare the benefits, operating costs and risks of each solution.

This research is part of a set of related research pieces. See "ATV: Storage Virtualization Is Changing Storage Deployment" for an overview.