Selecting tape systems to support storage consolidation. (Tape/Disk/Optical Storage).

Automated tape storage systems can play a central role in helping IT administrators consolidate storage to improve resource utilization, reduce management overhead, and improve overall data availability. Their value, however, depends to a significant degree on the architecture of the library system--simply buying a large capacity tape system instead of several smaller ones may not give IT departments the consolidation benefits they are looking for. The strongest support for consolidation will be found in tape storage systems that support mixed uses, that provide a controlled SAN interface, that supply advanced monitoring and alerting capabilities, and that are designed for efficient scalability (see Figure 1).

Tape Systems Can Support Mixed Use Operations

Consolidation means using a single resource to manage and store data that has, in the past, been managed and stored in several separate independent systems. Those systems frequently have a wide range of differences that caused them to be developed separately, including different user communities, primary applications, operating systems, server platforms, and backup requirements. When the storage for these applications is combined, it is rare to have all the differences eliminated. Much more likely is that different portions of the data will need to be treated in different ways. One set will use Legato or CommVault to backup their data while another has built its system around the capabilities that Veritas or CA offers.

The library architecture that best supports mixed use is one that allows libraries to be partitioned easily into different logical or virtual libraries. Each of these logical libraries presents itself to software applications as if it were a separate library, with its own independent storage pool and set of drive resources. The independence makes it easier to manage separate processes and to use different media types for different kinds of applications. With effective partitioning, one section of a library could be used for archiving a large digital asset management project--using an application program like ADIC's StorNext Management Suite and a fast-seek drive type--while other sections backup operational data using one or more standard backup packages. Partitioning can also provide long-term investment protection, as well, by allowing legacy technology to remain online and accessible while a transition is made to a newer technology. Without partitioning capability, each library is owned by a single applica tion and can normally use only a single media technology.

Even when a library is only consolidating backup operations and using a single kind of media, partitioning can provide very powerful advantages by letting users apply different backup policies to different kinds of data. Users might, for example, have a separate cluster of Linux servers running a specific application that needs different backup programs or procedures. Some people are beginning to use different backup programs for different standard applications--using one program for email and another for print and file servers. Partitioning is also a great way of separating processes for different customers, either external organizations or internal departments. It allows different policies to be managed more easily and it gives IT departments a simple way of charging departments back for the specific resources they are using.

One of the reasons that partitioning hasn't been used extensively up to now is that it has been relatively complicated. For most libraries, it has required running separate software on an external library control server--that has kept it more expensive and more complex. And supporting mixed media has never been a primary goal of library manufacturers who have a strong interest in seeing to it that libraries use the drive that they themselves manufacture. That's changing now, however. The latest versions of more intelligent libraries like ADIC's new iPlatform libraries, embed a server-class main control engine in the library which integrates a large number of management functions, including native partitioning and very flexible mixed media operation. When you evaluate mixed-use library support, look for integrated partitioning functionality that is easy to manage, that allows flexible re-sizing of the logical libraries, and that lets the user change the partitions easily.

Tape Systems in Consolidated Environments Need Controlled SAN Technology

It's not an accident that the movement to consolidate storage resources is occurring precisely at the time that Fibre Channel SANs have become ubiquitous in the data center. SANs provide a flexible, high throughput environment that eliminates much of the requirement to bring large amounts of data over a site's communications network, and that allows different parts of the backup process to share tape drives to help use resources more effectively. SAN operation is so much an essential component of the consolidated environment that among ADIC customers today, nearly 80 percent of the installations of our enterprise-class libraries--the ones that support consolidation--are installed in a SAN.

The tape systems that best support consolidation are the ones that have SAN support built into the library architecture--instead of using add-on components--and that provide a controlled and managed SAN interface--instead of forcing users to create a separate connection between each drive and the fabric. Using integrated controller technology in a library is analogous to using a RAID controller to manage I/O to disk spindles; it provides better management, can increase or tune performance, gives users more flexibility, and can increase reliability.

One obvious reason for using a controller is that it dramatically reduces the need for switch ports, keeping costs low and making installation and ongoing management easier. ADIC controller technology that is built into the Scalar family storage networking libraries, for example, can aggregate up to four high-performance midrange tape drives (LTO, SDLT, or AIT) for each port on the SAN fabric--that's a reduction of 75 percent in switch ports and cabling. Another way of thinking about this effect is to ask how much of a switch port's 200MB/second bandwidth is wasted when it's used for a single drive that won't sustain more than 50MB/s of data flow.

Controllers also provide a great deal more flexibility for drive and transport interface. Several important drive technologies today are available with SCSI interface only, and controllers allow them to extract full value from SAN implementation. Controllers also have the potential to easily provide different interface standards--iSCSI interface technology will be supported in controllers long before they will be supported directly in tape drives.

It may seem counterintuitive, but controllers can also provide better throughput performance than drives directly connected to a fabric. Remember that FC SANs are transmitting SCSI packets over a Fibre Channel transport. Something has to extract the SCSI data from the FC protocol and specialized controllers can do that more effectively than protocol conversion systems built into each drive. Carefully controlled benchmarks have shown a 20 to 40 percent write-performance advantage for drives connected through a library controller over drives connected directly to fabric ports.

In the long run, though, the most important reason that a controller technology makes sense is that it provides much better management for the library system, and as libraries support larger, heterogeneous, consolidated environments, the management increases in its value. For example:

* Security and zoning: As systems link more drives with more hosts, it makes management much easier to manage access between hosts and devices from the library. Controllers technology can provide both channel zoning and even LUN or device-level mapping. When the technology is integrated into the library, this can be managed through an easy-to-use graphical interface associated with the library much more easily than in the central fabric.

* Address persistence: Tape drives get changed and added frequently. If drives are directly attached to the fabric, installers normally have to update address tables in all connected hosts whenever a drive is changed. The right kind of controller technology matched to drive subsystem design eliminates the problem, presenting a persistent device address even when drives are changed.

* SAN-aware management: Because they are in the data path, controllers can participate in verifying and trouble-shooting connectivity all the way from the host down to individual drives. This is a process that at ADIC we call data-path conditioning. It includes the discovering of devices in the SAN from the library, and proactive in-band data path continuity checks that send a signal from each host to each drive to verify that paths are intact and ready to send data long before a backup begins. Path verification plays an important role during normal operations by confirming that paths are in good working order, and it can provide dramatic time savings when a problem needs troubleshooting.

Libraries in Consolidated Environments Need Better Monitoring and Alerting

When storage and backup operations are consolidated, the end-user is putting all his data eggs in one basket. As the old adage goes-you want to watch that basket carefully. That means being able to see what's going on operationally, and also making sure that if anything goes wrong, there's a way of finding out about it as early as possible and solving the problem quickly Historically, libraries haven't provided very much of this functionality themselves, sending users instead to various external programs to find out what's really going on. That is changing in the latest generation libraries, however, and it is allowing them to provide much better support for consolidated operational environments (see Figure 2).

The first big advances are on the side of watching for fault conditions and telling administrators and service teams about it. New libraries, like the ADIC iPlatform models, have increased dramatically the amount of sensing data that they process from different points in the library, and they have even extended fault detection to the backup system around the library. Today, library boards and components have become small systems themselves, fitted with sensors that gather temperature, voltage, and current characteristics at multiple points. Library drive subsystems include local sensors, processors, and memory that gather component status and operational parameters continuously. The library can even proactively query drives about their status while they are not being used, and it can generate heartbeat signals which are sent in-band from the external hosts to the library to confirm that connectivity is sound. The data from all these sources can now be carried back to a server-class central processing system t hat watches parameters and can tell when levels begin to drift outside optimal ranges.

Even more important is how the library communicates the results. In earlier generation libraries, users had to be watching a control panel on the unit or rely on external framework tools to pick up SNMP traps issued by the library and turn them into a message to an administrator. Unfortunately, framework tools, very useful in network management, have not extended their range to storage very effectively. The new generation libraries, are incorporating an alerting system into the library itself. The Scalar i2000, for example, can detect a suspected fault condition in the library or in the paths between the host and the library, and turn that into an email message to a specific list of individuals or pager networks. Proactive checking and active, flexible alerting means that users can now often learn about a fault condition well before a backup begins, giving them time to solve it and still keeping their backup processes intact. Problem solving also becomes easier and faster. Increased system monitoring and bett er conditional analysis routines now being built into the libraries can help identify faults much more rapidly to a single component. This translates into faster service and more uptime.

The sensing that libraries can now perform also extends to operational data, another important feature when users are managing several processes in a single resource. We are now seeing the first libraries to gather data transfer data directly from the drives and to present that to the user as an aggregated throughput number for the library. The first-generation iPlatform libraries, for example, show the user exactly how much data was written to and read from the library, and to each logical library, hour by hour. This kind of data helps an administrator quickly verify that backups were complete, and it provides a useful capacity planning tool.

Consolidated Environments Need Effective Scalability

Data storage and backup needs for any specific server or process are notoriously unpredictable. When users consolidate several processes into a single resource, the aggregated storage needs can change rapidly. This condition means that an automated tape system that is serving consolidation needs to offer an effective system for scaling. Effective means that it can add capacity and drives to match even large-scale needs while maintaining performance and reliability, that it can grow quickly with little impact on ongoing operations and, in an era of close attention to budgets, that it can grow cost effectively.

For tape systems, these requirements argue very strongly for a unified library strategy in which a library can grow without requiring that the user purchase separate libraries and join them using some kind of pass-through or elevator mechanism. The most effective architecture for a high growth, consolidated environment is one in which a library can grow in modules, but in which the modules form a single library with an uninterrupted interior space that is served by a single robotics system. This architecture is available from more than one supplier, but it is not the norm. This barrier-free sclability gives users a much wider practical scalability range than alternative approaches--the new Scalar i2000, for example, can grow from 100 tapes to more than 2,000 in a single, high performance library.

A more recent innovation, capacity-on-demand pricing, makes scalability easier to implement and allows users to pay for the capacity they need only when they actually use it. In ADIC's system, users are provided with extra installed tape slot capacity when they purchase a capacity-on-demand system. They can purchase this extra capacity in 100-slot increments and enable it using a software key whenever they need to increase the usable capacity of the library. When the space in the initial library configuration is all enabled, ADIC delivers additional physical modules without charge, providing additional capacity ready for them to purchase and activate. In this model, the only charge is for the enabled slots and there is no obligation for the owner of the system to purchase any of the extra cartridge positions.

Consolidating storage makes sense for many organizations. It means fewer devices and systems to manage, reduced management overhead, better resource utilization and improved overall data protection. Tape systems can provide essential support in consolidated environments, but the requirements for tape systems change when they are employed in that role. New generation tape storage systems like the ADIC iPlatform Scalar i2000 provide new combinations of capabilities that are designed to provide better support for consolidated storage installations. These include support for mixed use, controlled SAN interfaces, advanced monitoring and alerting, and broad-range, cost effective scalability.

Kevin Honeycutt is executive director, enterprise partners, at Advanced Digital Information Corporation (ADIC, Redmond, Wash.)

www.adic.com