4.4 Storage Virtualization

---

Virtualization fulfills a role for storage similar to what an operating system does for a computer system; it makes programming and operation simpler by automating resource management housekeeping functions. The abstraction of storage enables organizations to pool a wide range of storage technologies from different vendors and to easily add capacity and automatically move data among devices, independent of operating systems or network infrastructure. In separating the representation of storage to the server operating system from actual physical storage, a centrally managed storage resource can be created whereby vast amounts of physical capacity can be pooled across the enterprise for dependable presentation to application servers, wherever they may be.

The virtualization technology allows disks of different capacities and data protection levels to be created from a set of physical disks. Virtualization occurs within the storage pool, which contains all the storage assets that an organization chooses to deploy, regardless of vendor. Initially, the pool scales to satisfy the needs of a modest number of application servers, but over time, long-distance extension technologies can be used to scale the pool to a global dimension.

The goal of virtualization is to standardize and centralize storage management in a heterogeneous storage and/or host environment. This includes such functionality as LUN masking, LUN mapping, disk pooling, dynamic multi-pathing (DMP), and imaging. It could even include data replication. Virtualization decouples the relationship between physical disk and logical volumes, allowing customized logical volume sizes to be presented to the applications based on need rather than physical limitations. There are several approaches to virtualization, depending on where the virtualization is implemented: host, storage unit, or appliance.

Host

This method of virtualization is either packaged with the operating system or made available as an add-on product. It allows administrators to put one or more physical disks into a logical group and then split that group into one or more logical volumes. From there, data can be stored on those logical volumes as though they were physical disks of a specified size. Host-based virtualization also allows administrators to access advanced storage management functions such as mirroring, RAID sets, redundant paths, and hot backups (by using mirror splits or images).

Management of the virtualization must be done on a host-by-host basis, however, which eliminates the opportunity for centralized global storage management.

Storage

This approach allows for movement between different RAID groups without data loss, as well as automatic migration of data from one RAID group to another based on the frequency of data access. This method of virtualization also permits the creation of multiple data mirrors, which provide additional availability when one of the mirrors is split for hot backups. In moving these functions off the host and onto the storage subsystem, the host is relieved of the overhead burden. Storage-based virtualization also takes the availability of storage beyond what host-based virtualization provides, offers the flexibility to modify LUN size, and enables multiple hosts to see the same LUNs, which is particularly critical with high availability clustering and remote replication.

Appliance

This scheme provides virtualization between hosts and the storage, providing administrators with the same level of control and centralization across the storage architecture. There are two kinds of appliance-based virtualization products: in-band and out-of-band. With in-band, the appliance is physically located between the host and the storage. The appliance takes the disk requests from the host and fulfills the host’s request from the storage attached to the other side of the appliance. This functionality is transparent to the host because the appliance presents itself as disk. In-band virtualization reduces the amount of administrative time needed to load drivers onto each operating system platform. Since in-band virtualization’s presentation of storage to the host is indistinguishable from that of a normal disk drive, drivers need not be loaded on each host. However, there is a risk of overloading the appliance if several hosts make simultaneous disk requests, so the implementation of redundant appliances should be considered.

Out-of-band appliances logically present themselves as if they are located between the host and storage, but they actually reside elsewhere. A special driver is installed under the host’s disk driver so that the appliance driver receives logical-to-physical block mappings from the appliance. When the disk driver accesses the disk, it is actually accessing the appliance’s driver. The appliance driver then translates the block information and accesses the correct storage blocks. The virtualization appliances use pointers to manage the disk pool. Because it has no direct interaction with the host or the storage during I/O operations, the system delivers nearly the original disk performance.

Virtualization appliances are the management platform. They treat hosts and storage subsystems as commodities and perform all functions as a third-party overseer. They eliminate vendor-specific management by handling RAID levels, point-in-time backups, remote replication, LUN masking, and security.

Some vendors claim that SANs and virtualization are the same thing. But SANs do not include software to handle all storage media types and accommodate secondary storage. On the other hand, most virtualization products are limited to disk subsystems, and hard disks only address part of the storage equation. True storage virtualization software encompasses all devices and media types, as well as the ability to incorporate secondary storage devices such as tape and optical libraries, where data is stored away from the primary environment.

---