Section 7.3. Storage Virtualization

   

---

7.3 Storage Virtualization

Storage virtualization is very often positioned as a new technology that is a wonderful solution for all problems. In reality, the term storage virtualization represents a slew of technologies, some of which are new, and some of which have actually been around for quite some time.

Simply put, storage virtualization offers the ability to specify functional and application characteristics of the desired storage requirements without paying any attention to the details of the underlying hardware (e.g., type, vendor, location). As a simple illustration, consider the following example: An administrator specifies the need for an 80GB volume to back up data, but unfortunately all the hardware underlying the backup application has a maximum disk size of 40GB. To solve this dilemma, a volume manager might virtualize two separate physical disk drives into a single larger volume. With the growing attention to total cost of ownership (TCO) and the exploding amounts of storage, storage virtualization offers the hope of providing manageability, ease of use, better utilization of storage resources, and a reduction of the storage TCO.

Storage virtualization can be implemented in various ways, each offering different functionality and advantages. Indeed, there are several ways of classifying storage virtualization. One way is to specify the physical location; another is to specify the functional layer at which the virtualization lives. The following sections examine the various ways of classifying storage virtualization.

7.3.1 Virtualization in the (Host) Server

Virtualization has existed in the host server for quite some time, although it usually goes by a different name . Good examples include the Logical Volume Manager (LVM) and Hierarchical Storage Management (HSM). A volume manager (such as the Windows 2000 Logical Disk Manager, or LDM) can provide virtualization in several ways:

• By turning inexpensive unreliable disks into reliable disks using RAID software

• By providing larger volumes through virtualization of smaller disks into larger volumes

The Logical Disk Manager is described in Chapter 6.

Hierarchical Storage Management provides a way for applications to access files transparently without regard to whether the file is residing on a hard disk or on removable media such as tape. The HSM that is shipped by Microsoft is described in Section 7.8.

Virtualization in the host server represents a mature and stable technology. However, there is the perception that it contributes to islands of storage where certain storage effectively is dedicated to certain servers. This perception is especially true, given the lack of popularity and availability of clustered file systems and volume managers with cluster support that allows multiple servers to truly share and simultaneously access the same storage devices, file systems, or volumes.

7.3.2 Virtualization in the Storage Hardware

Virtualization in the storage hardware has also been around for quite some time. A good example is RAID implementation in the storage subsystem. One advantage is that the implementation is efficient because it is hardware based. Another advantage is that this implementation is robust and secure because of the close proximity of the data repository. Consider a data storage device that is managed by virtualization in the host (server). A malicious user may circumvent the software virtualization in the host or deliberately bring a host online without the required virtualization present in the host and then access the data repository.

The disastrous effects of this scenario can be prevented with a storage unit that implements its own virtualization. Another advantage is that the solution works for a multitude of servers, irrespective of the actual operating system used on the server. The drawback is that the data virtualization is managed in a proprietary way by the hardware vendor, which can cause interoperability issues.

7.3.3 Virtualization in the Storage Network

Instead of doing virtualization at the two end nodes (the host server or the storage hardware subsystem), the virtualization is done in the storage network, either in the switches or routers or via special hardware deployed just for the purposes of virtualization and storage management.

7.3.4 In- Band Virtualization

In in-band virtualization , the virtualization occurs in the active data path. The host software does some I/O, and a device that sits in the data path between the host server and the storage hardware handles the I/O.

The advantage is that the solution works for a range of activities, including file-oriented and block-oriented access of storage. The appliance can provide a single point of management.

The disadvantage is that the in-band virtualization appliance device can become a single point of failure and a performance choke point. We can alleviate these problems by having multiple such appliances, adding more memory to the appliance, using higher-performance components , or building caching into the appliance. However, these tactics add cost and create a more complicated solution that requires clustering of the management devices.

7.3.5 Out-of-Band Virtualization

With out-of-band virtualization , the data path along which the storage management information flows exists apart from the control path. One can compare out-of-band virtualization with asymmetric clustered file systems, in which a metadata server controls the file system metadata access. The virtualization server typically communicates with virtualization clients , which are lightweight software running on the host server or devices in the storage network such as fabric switches or HBAs.

The advantage is that the data access can be much faster because the data does not have to go through an additional device. Another advantage is that one can have a single point of control.

The disadvantage is that one ends up with a different problem involving managing multiple out-of-band virtualization devices. Multiple virtualization devices are typically needed for several reasons, including

• To provide redundancy

• To provide the needed performance

• To conform to what the fabric topology dictates (e.g., multiple SAN islands that are physically disjointed from each other would each require a virtualization device)

Disk virtualization refers to software and firmware that has been shipping as part of disks for years . The functionality maps bad sectors to good ones transparently to the operating system. The functionality also hides the intricacies such as number of spindles, heads, and so on from the operating system.

Block virtualization also refers to technology that has been present for quite some time. Good examples on the Windows NT platform are the Logical Disk Manager and the FtDisk driver that provides RAID functionality and the ability to present a single large volume to an application that may be larger than any individual physical disk's capacity.

File virtualization refers to functionality that abstracts the location and characteristics of files and directories. A good example is Hierarchical Storage Management, or HSM.

File system virtualization refers to functionality that abstracts multiple file systems into one perceived file system. A good example of such functionality on Windows NT is the distributed file system (Dfs) described in Chapter 3.

Tape virtualization refers to functionality that abstracts tape media and tape drives. In the absence of tape virtualization, a host server typically has a dedicated tape unit, its own tape libraries, and its own tape media. Tape virtualization typically simulates tape devices and tape media, caching the I/O happening to the simulated tape onto disk instead. The cached I/O streams at a convenient time to a real tape device. The advantage is that the I/O completes faster, and the need for numerous tape devices is minimized.

---

Top