11.1 What Is Storage Virtualization?

The common denominator for storage virtualization initiatives is the abstraction of storage from its physical deployment and configuration. Without this logical abstraction, the host system would have to discover and manage every physical disk or tape subsystem under its control. It would have to be aware of the data paths between the host and its storage resources and would have to accommodate the unique characteristics of each target device. Storage virtualization hides this physical complexity from applications residing on the host system, instead presenting a simplified view of storage resources. With a virtualization mechanism in place, the application sees only specified storage volumes that may in fact reside on a very complex assortment of physical storage devices on a multipathed network infrastructure.

RAID controllers are an obvious example of storage virtualization engines. As discussed in Chapter 5, a RAID controller sits between the host and the physical disk drives on which storage data is placed. Each physical disk supports a logical unit that executes SCSI read and write commands. Without a RAID controller to manage the physical drives, the host would have to attach to and manage each one individually for example, as a JBOD. The RAID controller simplifies connectivity and management of physical drives, however, by appearing to the host as a single storage resource. The host system is not aware of the physical configuration of disks sitting behind the controller, their individual attributes, nor even how many disks are installed in the RAID chassis. When querying the RAID controller, the host sees one or more LUNs that the RAID controller has logically aggregated from its attached drives. In addition to striping data across several physical disks to enhance performance, the RAID controller may provide mirroring for data integrity. These internal RAID functions are transparent to the host, which benefits from both simplified administration and the efficiency provided by the RAID system.

Although RAID storage devices were not initially marketed under the banner of storage virtualization, they are an atomic component of a virtualized solution. Just as a single RAID controller provides a simplified logical view of its attached storage devices, multiple storage systems can be virtualized to appear as a single storage pool, as shown in Figure 11-1. As with an individual RAID controller, this superset of abstraction requires an intelligence that sits between the host and multiple downstream storage arrays. Various schemes have been offered to solve the problem of where to place the virtualization engine for greatest efficiency.

Figure 11-1. Physical and logical views of a virtualized storage network configuration

Storage virtualization takes on the burden of configuration and administration of physical storage devices but also enables more efficient utilization of storage resources. Without a virtualization intelligence, individual hosts would be assigned separate storage arrays. In that case, one host might use only 40 percent of its assigned storage capacity, whereas another host might use 90 percent. You would not be able to assign the excess capacity of one host for use by another host, and you might purchase additional physical storage even though a surplus actually existed. Storage virtualization aggregates the total capacity in the SAN and, with the proper implementation, can automatically balance utilization across multiple hosts. This reduces the total cost of the shared storage system and helps to consolidate storage administration.

The virtualization engine also screens the host systems from changes on the back-end storage network. Because the host accesses storage via the virtualization intelligence and has no awareness of the physical devices in the storage pool, new storage can be added nondisruptively. Networking storage via SANs facilitates scaling of storage resources, and virtualization masks the physical adds, moves, and changes to the network. If virtualization is properly implemented, the ability to hide physical changes in the storage configuration should eliminate or at least minimize down time.

If myriad physical storage devices now appear as a single logical storage pool, the capacity of the individual devices that compose the pool can be more efficiently utilized. As shown in Figure 11-2, traditional direct-attached storage cannot leverage unused capacity. Because each server is the exclusive owner of its own fixed storage resources, there is no means to balance capacity among multiple servers. One server may have excess capacity, while its neighbor is exceeding its allotted storage. Whereas SAN technology facilitates sharing of storage capacity, virtualization makes the sharing transparent.

Figure 11-2. Inefficient storage utilization with direct-attached storage

As shown in Figure 11-3, the virtualization of storage into a single pool enables more efficient utilization of the total storage capacity. Available disk resources can be used on demand, and the total capacity can be resized nondisruptively.

Figure 11-3. Maximizing capacity utilization with storage virtualization

In addition, the composition of the storage pool itself can vary, depending on performance, availability, and budget requirements. To boost performance, the storage pool can present logical volumes that ultimately reside on RAID arrays with large cache memory. For availability, the RAID arrays can be mirrored to other RAID systems or to cheaper JBOD banks. For economy, the entire pool can be composed of JBODs, or even SCSI disks that are front-ended by SAN bridges. Because the potentially eclectic assortment of physical storage devices is hidden behind the virtualization engine, the complexity of storage capabilities and redundancy does not need to be managed by individual servers.

Storage virtualization strategies can be extended over distance by use of SAN extension or IP storage technology. This makes it possible, for example, to implement campus or metro storage pools that further streamline storage operations and amortize storage costs over more servers. If multiple departments within an enterprise are brought into a common virtualized storage scenario, the unique requirements of each can be accommodated by policy-driven mechanisms to enforce service level agreements. This setup represents a higher level of virtualization intelligence, because the virtualization engine must not only aggregate diverse physical storage but also determine which applications require special treatment in terms of performance and availability.

The discussion thus far may imply that these capabilities are already available in virtualization products. Although some features may be available, the higher-level functionality for automatic sizing of volumes, nondisruptive changes, and quality of service are still under construction. The ideal of application-aware virtualization, which identifies data types and automatically determines which storage policies should be applied, may not be available for some time.