Virtual Worlds: Virtualization Layers In The SAN.

Saving money in both personnel and hardware

SAN virtualization makes physical storage devices act and appear as logical volumes in a storage area network. IT administrators can then allocate the volumes on demand to application servers without resorting to awkward host zoning and LUN mapping procedures.

Samuel Tam, CEO of Vicom said, "You need to be able to have centralized services, and those services now include host zoning, LUN masking, and being able to map storage resources in a secure manner to all the users and servers attached to the network. Our idea is that we would take it to level of granularity, allowing a very small chunk of storage partition to be available, to be dynamically allocated to any users of servers. We're talking about true virtualization, where it's done at each individual block level."

This method of SAN virtualization, which represents a distinct layer in the storage stack, can save money in both personnel and hardware. It impacts personnel by allowing servers to access centralized storage while IT administrators manage the storage as a single logical entity. Since storage is growing 100% a year in many cases and additional headcount is unlikely, easier storage administration is welcome news. In fact, IDC reported that single storage administrators can manage 7.5 times more data when it is centralized than if it is dispersed. The hardware budget improves because IT administrators can allocate storage much more efficiently. It's also relatively easy to reassign storage on the fly, whether for an emergency need or temporary disk space for migration, testing, or financial closes.

Although some products have offered virtualization types for years, SAN virtualization as a separate layer from a server virtualization layer is just now coming into its own. It offers two different functions. The first is the access function, providing virtual volumes for security, storage consolidation, storage on demand, and SAN file systems. Control functions offer SAN services for serverless backup, archiving, migration, and virtual devices for Storage Resource Management (SRM).

SAN Virtualization Layer

Since software server components divide storage space into volumes, the physical layout of the actual storage devices is irrelevant. This contrasts with the present method for dynamically allocating storage, which is combining LUN zoning with port zoning. The method offers a high level of security, but there are distinct problems connected with the technology as the data is held captive in layers and within different operating systems.

According to DataCore, the top three technical issues in dynamic reallocation are:

1. Hosts communicate with physical storage devices using SCSI protocol over Fibre Channel. The destination address is a combination of a Target I.D. and a LUN.

2. Some hosts take ownership of any visible LUNs on attached fibers.

3. Two or more hosts connected to the same LUN will automatically compete for the same storage device.

To get around technical limitations, IT often zones switches to block other servers from connecting to the same storage device. "LUN Masking" relies on each application server to ignore the neighboring servers' LUNs by using special host-based device drivers to provide a filtered view of the storage network. However, some hosts cannot accept a LUN Masking installation so they often cannot share storage assets. Both workarounds are time-consuming, which affects downtime, backup and capacity chores, and allows large areas of storage to remain unused.

In contrast, the IT administrator can create, expand, remove, move, and selectively present a virtual volume to users, independent of its storage subsystem. It can consist of storage space in different subsystems, even subsystems with different characteristics. This results in a highly efficient space-assigning operation, saving money both on added storage hardware and on scarce man-hours and storage personnel. LDC reports that consolidating data in one location can lower data management costs from 55% to 15% of the total storage budget.

Killer Apps?

Investment firm Morgan Keegan lists several top applications for SAN virtualization and different architecture types (See Table).

Storage sharing. IT administrators must regulate storage access to prevent storage sharing conflicts. They do it now with a combination of port and LUN zoning, but SAN virtualization makes it much easier to map servers to storage as virtual volumes. This would produce an integrated access framework spanning all SAN resources in an organization.

Data sharing. Port and LUN zoning enables storage sharing, but no data sharing as access to each LUN is restricted. Managers usually use NAS servers at the logical layer to solve the problem, the NAS devices acting as intermediaries between data stored in specific LUNs and the application servers. This helps, but you then have the problem of an additional latency of a NAS and increased file transactions over the LAN.

To answer this problem, a SAN File Systems approach is emerging. The IT administrator installs a network file system such as NFS or CIFS onto each application server to capture and forward file operations over TCP/IP. Using SAN virtualization as a key enabler, the file server processes the operation and sends the data traffic to the application server over a SAN instead of the LAN. A similar procedure can handle clustered file systems.

Performance optimization. Storage I/O demands vary for different applications. For example, both data warehousing and OLTP applications generate frequent small bursts and are usually measured in 110 operations per second. However, multimedia applications often produce fewer bursts and more frequent large, streaming blocks of data that are measured in MB per second. The IT administrator optimally tunes storage arrays to handle one or the other type of data traffic, but must choose between tuning for performance with multiple storage devices, or capacity with a single, shared storage device. SAN virtualization can optimize both throughput and capacity metrics.

Storage on demand. Managing open systems increasingly depends on a multiple-tier-distributed architecture. In this architecture, IT can scale layers independently of each other for optimal resource utilization. SAN virtualization offers this capacity in precise increments at the physical layer while remaining invisible to the logical and application layers. This type of non-disruptive scaling is especially important for SSPs and their heavy storage-on-demand capabilities.

SAN Services. A storage central point such as a SAN needs careful data protection policies in place, such as backup, archiving, replication, and migration. These protection measures can be carried on at the SAN itself, skipping the LAN, which results in higher speeds without disrupting business applications.

SAN Virtualization Players

Available virtualization products rely primarily on disk-based virtualization, but are good approaches to the problem of virtualizing storage space for dynamic reallocation. The field can be confusing in spite of the shared "virtualization" buzzword, since, many of the products represent different aims and architectures, and many companies define SAN virtualization differently. Augie Gonzalez, director of product marketing at DataCore, agreed. "For example, Gartner means that SAN virtualization looks at capacity planning and leveling resources. DataCore is using an SRM product to implement tools--you're running out of disk space, take some action, allocate more capacity." DataCore's SANsymphony is based on network storage pool architecture and, allows organizations to add and redistribute capacity on the fly with a drag-and-drop interface in heterogeneous environments.

Seagate's XIOtech has a different approach. For the last several years, they have offered a proprietary environment starring its MAGNITUDE line and REDI (Real-Time Data Intelligence) Storage Manager, which it calls MAGNITUDE's virtualization cornerstone. It allows users to share storage across heterogeneous servers running on different operating systems, simplifies storage management by virtualizing physical devices, and provides high availability and fast I/O optimization.

Vicom has an approach to a heterogeneous environment. It's based on Fibre Channel, though CEO Samuel Tam says they'll be ready to support SoIP and InfiniBand when those technologies become available for SAN environments. Vicom's Storage Virtualization Engine (SVE) provides disk-level virtualization of Fibre Channel, SCSI, or SSA storage for any Unix and NT servers. Storage pools can scale upwards to more than 500TB.

Other virtualization products range from StorageTek's Nearline procedures, which virtualize tape storage into faster disk storage by sending it to a buffer, and VERITAS' Volume Manager, which allows disk virtualization in a multi-vendor environment.

The upcoming new technologies represent a hot new market, likely to take off in 2001. It's this type of approach that makes storage based networks become more and more practical to more and more users, from Storage Service Providers and enterprises to mid-level organizations.

Editor's note: The above products are a sampling of virtualization products, not a roundup. If readers know of other virtualization products, I'll be happy to hear about them at [e-mail]. We will continue our coverage of virtualization throughout the year to come.

Archiecture            Description                 Strengths
SAN controller  implements virtualization  Regulate access policies
                intelligence on a central   for SAN servers without
                   device with a Fibre     server-resident software,
                    Channel switching            can use JBOD
                        platform
SAN server        Central approach for     Regulates access policies
                    widely available        for SAN servers without
                  computing platforms,     server-resident software,
                   expansion board for           can use JBOD
                     custom hardware
SAN Metaserver     Removes bottleneck      Cost versus performance,
                  by separating access          highly scalable
                  control traffic from
                  storage data traffic
SAN Cluster     Implements virtualization  Cost versus, performance
                intelligence in  software        can use JBOD
                   on each application
                         server
Archiecture            Challenges
SAN controller   Design complexity, cost
                     vs, performance
                   bottlenecks due to
                     centralization
SAN server      Performance cannot use a
                   switching platform
SAN Metaserver     Requires additional
                server-resident software,
                  must use RAID storage
                 provided by the storage
                         devices
SAN Cluster        Requires additional
                    server-resident,
                   software sacrifices
                scalability since mapping
                     is mirrored to
                       each server