Scaling Storage with Virtualization

Virtualization is a powerful storage tool that encompasses many other technologies and techniques, including RAID. The power of virtualization comes from its ability to create scalable storage by utilizing the flexibility of SAN connections. This section examines the fundamental ways virtualization applies simple geometric connection flexibility to build scalable storage.

The Virtualization "Lens"

To make it easier to discuss storage virtualization processes, we will use the word "lens" to pinpoint functions at a certain point in the I/O path. Just like an optical lens is used to define the scope and range of a visual image, the storage virtualization lens narrows or widens the "view" of block addresses in a storage address space.

To clarify, the virtualization lens does not perform any magnification, and of course it does not (or shouldn't!) distort the data in any way. Another major difference is that virtualization can seamlessly assemble address spaces from many different sources simultaneously so they can be "viewed" as a single, composite structure.

Figure 12-4 shows a virtualization lens in the I/O path between a host system and a pair of storage LUNs.

Figure 12-4. A Virtualization Lens in the I/O Path

Fan-Out of a Virtualization Lens

Fan-in was discussed in Chapter 5 as a way for a single subsystem port to export multiple storage address spaces, where they can be accessed by multiple host systems. Fan-out is the inverse process and refers to the number of downstream targets that can be accessed by a single virtualization lens. Figure 12-5 shows a virtualization lens with a fan-out ratio of 1:5.

Figure 12-5. Fan-Out of a Virtualization Lens

Multiple hosts can connect through a single link between a switch and a SAN virtualization system. Each of these host sessions can then connect to multiple downstream LUNs after passing through the virtualization lens.

Recursivity of Virtualization

Virtualization lenses can be used sequentially. For example, a geometric "virtualization tree" can be formed to aggregate storage from several downstream LUNs, which, in turn, are formed by LUNs that are downstream from them. Figure 12-6 shows recursive virtualization in a tree topology with three layers of lenses.

Figure 12-6. Recursive Virtualization Lenses Forming a Tree of LUNs

NOTE

Multilevel tree topologies are not necessarily recommended. They are shown merely to illustrate a simple idea. There is always some latency incurred with moving I/Os through a SAN virtualization system. This could cause problems, depending on the product implementation and the I/O requirements of the application (read transaction processing.) Don't assume all SAN virtualization products are the same in this regard. They aren't.

Comparing the "Views" of Volume Managers, SAN Virtualization Systems, and Storage Subsystems

Scalability and flexibility in storage networking depend on the ability to locate and use downstream storage resources. By virtue of being connected to storage over a network, as opposed to a device interconnect, volume managers and SAN virtualization systems have much better "views" of storage resources in the SAN than disk subsystems do. Figure 12-7 compares the view of a storage virtualization product with disk subsystems.

Figure 12-7. Comparing the Downstream Storage "Views" of Storage Virtualization Products with Disk Subsystems

Volume managers and SAN virtualization systems have essentially the same downstream views of the network, because both of them connect to SAN switches. Any resource "seen" by one can be seen by the other. In most SANs, the views of storage are managed by fencing technology that blocks access to SAN resources. Examples of fencing technology include LUN masking (see Chapter 5) and switch-based technology such as virtual SANs (VSANs) and zoning, which are not discussed in this book.

There is a big difference between the upstream views that volume managers and SAN virtualization systems have. The volume manager is limited to presenting I/O responses upstream to a single system. The SAN virtualization system, on the other hand, often works with many upstream host HBA initiators. This gives the virtualization system much stronger capabilities for centralizing management of the storage.