Could Tape Libraries Become The Center Of SANs?

Vendors scramble to differentiate their products to cash in

The tape industry has seen the drive market settle down to low unit growth, especially regarding directly attached drives. However, all tape libraries require drives and that's where the growth is for the total industry. According to Freeman Reports, the library market will expand 15% this year. That's why so many of the drive manufacturers are going into the library business, even for libraries that use competing drive technologies.

In searching for some way to differentiate, it appears likely that libraries could soon become the center of the SAN (Storage Area Network) implementation. This could result in libraries housing not only the tape drives and robotics, but also RAID (Redundant Array of Independent Disks), SAN switches/hubs/routers, and backup software and it could apply equally well to NAS (Network Attached Storage) because the hardware can be essentially the same. For this article, the focus will be SAN.

What's Driving The Differentiation Need?

Fifteen percent market growth is the good news. The bad news is that the number of companies competing in this space also grew, which means that there is a larger group of companies chasing that 15% market growth. Clearly, the pressure is on them to differentiate. It's no longer adequate to just come up with a clever carousel design or perhaps a nifty stackable design.

So much of the performance in a library is determined by the drive technology that there is only so much that the library design can do to make the robotics performance stand out. What good is a cartridge exchange time of six seconds if the drive requires a minute to become ready? What's more, scalability in libraries is questionable. Is it smarter to pay lots of money for a library with ten drives and 500 slots than to buy ten autoloaders and put them in a rack? In some of the larger libraries, the slots average $500 each and the drives cost $8,000 or more. That values a one drive, seven slot autoloader at $11,500, which is 30% more than the real price of reliable autoloaders today. The software has to know where the cartridges are anyhow, so why do all cartridges need to be accessible to all the drives? The centralized software that is running across all the hosts has to know where the backed up files were stored anyhow, whether on one library or several different ones.

The point here is that--given today's technology and markets--there just isn't enough functionality in a library to create a competitive advantage between manufacturers. If the library is going to be cost-effective and competitive, it has to do more in the same floor space than to just provide the same functionality as a group of autoloaders.

With the entire buzz surrounding SANs, the natural place to look for growth is the environment in which libraries are used. Consider a typical SAN setup in today's world (See Fig).

The RAID disk system is the primary storage for the system while the FC bridge provides the high-speed communication around the SAN. Until last year--and even to a major extent this year--the tape library was a SCSI device and, thus, could not communicate with the other SAN elements, that is, the RAID and the host server, because they frequently are Fibre Channel devices. That made the Fibre Channel bridge a necessity in order for all these components to exchange data with the tape library.

Now, the issue of backup enters. In the classic approach, the server would have backup software installed and the server would conduct the backup. Of course, the major drawback to this is that the server had to usually shut down all the other applications while the backup ran. While Open File Manager software from St. Bernard was often used to mitigate the problems of trying to back up open files in this environment, it is still a fact that the more applications that are running during a backup, the more likelihood there is of problematic restores.

At this point, some end users solved this problem by having a dedicated backup server whose function was nothing but backup. It's a direct solution, but costly; it adds another server just to house the backup software.

What Has Been Happening Recently?

There have been several developments over the past year.

First, most library vendors have added FC interfaces to their libraries. Some have embedded the FC host bus adapters inside the libraries. With these features, the libraries can "speak" FC. Other vendors provide an external FC to SCSI bridge box to accomplish the same thing, but it's not inexpensive; many of these bridge boxes are priced at $5,000-$7,000 to the end user.

Second, most manufacturers have enabled their libraries to be split logically. This means that one large library can be made to "appear" as several smaller libraries, each supported by a different host and--in some cases--a different platform type. Now, Unix servers and AS400 servers could use the same physical library (but not the same physical cartridges). This, again, raises the issue of large scalable libraries versus a group of small ones. Since heterogeneous platforms frequently cannot share data, why pay extra to have each of them with access to the other's data? If "split" libraries make sense, why buy the big libraries unless they add more value?

Third, most of the vendors have announced Remote Management Capability for their libraries. Now, the libraries have either an Ethernet port or a RS232 for communicating with the outside world. Usually, this connection is for management and diagnostics, but still, it's a path to the outside world.

Fourth, many vendors have gone to Intel-based motherboards with an integral PCI bus. The PCI bus enables the use of SCSI or FC host bus adapters and the motherboard also provides a home for the software, which splits the libraries logically. Not only is common Intel architecture used, but also many of these libraries are running Windows NT or 2000 just to control everything. The library literally has a storage server inside the library.

Last, some manufacturers have begun providing rack space inside their libraries to house the server(s). The fact is that the libraries take up most of the floor space now when compared to RAID or FC hubs and routers. This suggests that all the SAN components or most of them could be inside the tape library.

So, looking at today's libraries, we see many of the functions that are essential to SANs:

* Backup software host

* FC to SCSI bridges

* Partitionable libraries

* High Speed Data path to the host

* High Speed Data path between Disk and Tape

* Hot-swappable components

* Serverless backup 

Where Is It Going?

Recognizing that the libraries are acquiring all these capabilities leads to the question of whether they can be the entire SAN or, maybe, the major part of the SAN. A major issue is how the disk system is used in the system. No question, disk can be used as a cache for the tape. It certainly makes the tape look much faster. However, even though most applications will require more than 13 rack units for the disk, integrated disk is still a way to enhance library value. The library firmware can determine when to migrate data from the disk to tape based on locally established algorithms. In the meantime, the disk can be functioning as system storage. In fact, the same software can determine how much of the disk to use for primary storage and how much for cache. It can be an adaptive system in as much as it can adjust space allocation based on usage rates.

One subject not yet discussed is standards. The FC world is still in conflict over which standards should be used. If the disk and tape and FC switches are all inside the library, the standards become much less important because the library manufacturer is the single source for all of them and they are the ones who must insure interoperability. In this case, the standards issue represents much less of a problem when it comes to interoperability.

In summary, there are two forces that are intersecting in the tape library and SAN markets. One is the need for library manufacturers to add more value. The second is the technology, which is always looking for a market need to fill. It is quite likely that these forces will produce larger and more functional libraries.

Max Bosetti is the principal in Storage Metrix (Thousand Oaks, CA).