Tape Storage Is (Still) The Best Backup Medium.

Mark Twain said it best almost 100 years ago; after reading newspaper articles that he had died, Twain wrote the editor, claiming, "the reports of my demise are greatly exaggerated."

That's how it's been with tape storage. New storage technologies have repeatedly arisen, challenging tape's viability. Tape was supposed to have been replaced by more advanced products, but CD-ROM, magneto-optical (MO), DVD, and even RAID all have serious limitations as enterprise-level backup and archival storage media. Thanks to technology advancements, including some just around the corner, tape storage continues to be the most cost-effective archive and business continuity storage solution for the foreseeable future.

Best Uses For Tape Storage

Any data center operations manager worth his or her title knows the value of tape storage for protection of corporate data. It's reliable, stores large volumes of data, and it's mobile, making it ideal for offsite storage and, thus, disaster recovery.

Tape storage is also the most economical way to archive online transaction information. When systems supporting online transaction environments experience a problem, they must be able to recover every transaction right up to the point of failure. Typically, such systems write audit trails to disk. When the audit trail reaches a certain size, it closes and a new one opens. At some point during the day, the closed audit trails are pushed over to tape to free up disk space.

Centralization--A key Trend In Data Storage

For large and medium size enterprises, centralization of computing resources is a growing trend. Following the explosion of distributed, de-centralized environments in the '80s and early '90s, high speed networking has enabled a return to the "glass house" data centers of old. For data centers with high online transaction and data volumes, this trend is a real plus for tape backup. Why? Because large numbers of tape devices and host CPU's are in physical proximity with one another. As we'll see shortly, this is critical to upcoming advancements in tape storage performance and flexibility.

Myths About Tape Storage Performance

To understand the direction of innovations in tape storage, it is important to clarify some common misconceptions about tape performance.

* Faster is always better (Fig 1). The reality is that more than 95% of mainframes and servers today cannot stream data fast enough to saturate a tape drive running at 5MB/sec. File copies to tape and logical database or transaction monitor backups all require CPU processing time and this is the real bottleneck. You can attach your tape device(s) to your host with high-speed SCSI and Fibre Channel connections, but it probably won't help. Until the CPUs move data faster or until serverless backup (direct from disk to tape) becomes a reality, tape drives today are fast enough for the tasks they need to handle.

* Higher capacity per tape is better (Fig 2). This is true if you fill your tapes with data. However, most data centers don't. They tend to store logical sets of data on a particular cartridge(s), limiting the amount of tape actually used. The bottom line--large amounts of tape capacity goes unused. The other concern is that searching a full tape for a file or set of files, especially high capacity tapes, can take longer. In practice, it isn't always wise to fill your tapes to capacity.

* Data compression provides a 2:1 capacity and performance boost. While this is the specification for most tape drives, tests show that different compression algorithms compress at different rates. For example, a DLT drive claims 2:1, but its LVW compression achieves more of a 1.8:1 ratio. An AIT drive claims 2:1, but its more modern ALDC algorithm achieves about a 2.6:1 ration--almost 45% more performance. Despite this, although compression helps achieve tight backup windows and perform capacity planning, there doesn't appear to be any new revolutionary schemes that will improve performance significantly.

If The Host Is Still The Bottleneck, What's Next In Tape?

"Loyalty to petrified opinions never yet broke a chain or freed a human soul." Preconceptions, Twain thought, can be deadly, ensuring that people fail to consider new ideas or new ways of approaching problems and their possible solutions; and so it is with tape.

Right now, faster tape devices, higher tape capacities, and new compression algorithms aren't necessary. What must be overcome, however, is the current one-to-one or one-to-many relationship between CPUs and tape devices. Companies have to purchase one or more tape devices for every CPU that needs to be backed up or run the risk of slow backup over networks.

Bear in mind that directly connected tape devices using SCSI or Fibre Channel are the fastest and most reliable arrangement (Fig 3). While LANs go a long way to eliminating the physical wiring limitations of SCSI and Fibre Channel [80 feet (differential) and 150 feet (copper point-to-point), respectively], Ethernet is generally too slow and simply drops too many packets to provide the absolute reliability required in mission-critical transaction and database environments.

What companies relying on tape storage need is the proverbial "black box," providing a "many-to-many" relationship between CPUs and backup devices; in other words, a device that will enable a shared tape device pool (Fig 4). With the "black box" sitting between multiple hosts and a shared pool of tape devices, data centers could make better use of the tape capacity they already have, allocating available storage devices to hosts at different times of the day or for different purposes, as required. And they could add servers or backup devices as needed, not on a strictly one-to-one basis.

Speaking Everybody's Language

In order for such a "black box" to be effective, it would need to talk seamlessly between any tape device and any CPU. While many Unix and NT hosts support open architectures and communications protocols and, thus, standard tape devices, there are still a number of host vendors with large installed bases who require proprietary tape devices due to their proprietary operating systems. A dedicated tape storage server running the appropriate emulation software must deal with any combination of host machines and tape devices to succeed in the marketplace. In addition to flexibility, a big benefit of emulation would be cost savings. A company running a proprietary operating system or a mix of systems could buy standard tape devices, instead of more expensive proprietary devices. In many cases, standard tape devices are one-half to one-third the cost of proprietary versions.

A Real Life Scenario

Consider a large health services company on the West Coast. More than 900GB of data are backed up every night on its mainframe system. Because the particular mainframe being used can only push data from disk at about 1MB/sec per channel, the company had to purchase 36 tape devices and establish multiple channels to the drives in order to complete the nightly backup within the allotted window. Unfortunately, once the nightly backup is completed, almost all of the 36 tape devices sit idly during the day because they are attached to the mainframe. The system only requires a few of the drives to handle normal transaction auditing during daytime operations.

If it were possible to share those 36 tape devices among different servers and hosts, even if they were from different vendors, the company could avoid having to purchase separate drives for their other production servers and save substantially on their tape backup expense. It spent nearly $1 million on its proprietary tape units because the company uses a proprietary operating system. A centralized solution with emulation capabilities could have saved the firm at least $500,000 by buying off-the-shelf tape technology.

Other Cost Saving Tricks With Shared Tape Pools

Today, lots of companies make two copies of their back-ups, one to keep on-site, the other for off-site disaster recovery purposes. A black box approach could enable this to be done in a very flexible, user-defined manner. Simultaneous writes to two separate devices, even if they were from different vendors or used different tape formats, could be handled. Alternatively, the chore could be done "off line," doing a physical copy from one tape drive to another without involving the host CPU. This would allow users to make better use of their existing storage investment and would free up processor cycles on the host CPU.

Many tape libraries today can be partitioned logically, appearing to be two, three, or more separate tape libraries. With appropriate logic built into a tape server black box, these partitions could be assigned to different CPUs and operated concurrently, transparent to the user or partitions could be used to mirror backups. Again, the objective is to make better utilization of tape devices and capacity, yet provide scalability as needed.

Then, of course, there is the Holy Grail of tape storage--serverless backup. The objective is to eliminate the bottleneck caused by the server/host CPU. A clever vendor will soon come up with a way to make a tape server black box that directly transfers data from disk to tape at native I/O bus speeds, avoiding the host CPU entirely. When this happens, high-speed SCSI and Fibre Channel connections will be used to their full capacity for transferring data.

Will Disk Mirroring Replace Tape? Not likely.

The advent of RAID has brought significant improvements to the availability and reliability of on-line data. Using the mirroring option of RAID, data centers now have created access to data with no single point of failure. Except, of course, if the data center experiences a major disaster. That's where off-site storage of tape comes in to save the day, maintaining the advantage it's had for years.

High network bandwidth and advances in database and transaction software have made it possible to conduct remote disk mirroring; duplicating data to a distant site in real time. This is a nice alternative, if you can afford it. Yet the key question remains: how many companies can spend the money to completely duplicate their production environment?

There is still one major flaw with real-time mirroring; there is never a static "snap-shot" of data that can be used as a fallback position. Suppose a software program has a logic error in it that kicks in sometime during the day, writing corrupt data for a period of time before the error is noticed. Without a static backup, there may be no way to retreat to a point in time prior to the beginning of the data corruption. Tape provides the needed fallback security.

Keeping Pace With Explosive Growth

"Always do right. This will gratify some people and astonish the rest." Twain was prescient; tape has been doing the right thing for years, gratifying those who have grown to rely on its dependability and astonishing those who predicted its demise long ago.

One thing is certain. Overall data storage requirements have risen dramatically in the past three years. That growth will likely continue unabated for the foreseeable future. However, the need to archive it and have it available for recovery purposes do not change. With the Internet creating more 24/7 business opportunities, backup windows will be under tighter restrictions and recent transaction history will need to be readily available. Therefore, tape performance will continue to be critical to a company's overall ability to conduct business.

The tape storage market will continue to evolve, as it has through the years, to meet performance demands, but, until CPUs start streaming data at better rates, the opportunity for performance improvements lies in the increased flexibility and utilization that a tape server "black box" can offer. It will be interesting to see if anyone is up to the challenge.

Barry Mattingly is the director of business development at TapeLabs, Inc., (Los Angeles, CA).

Only 3% of mainframes and servers can transfer data at the rate of 3MB/sec or more. Most modern tape devices can accept at least 5MB/sec.

                        Tape Bandwidth Utilization
          Percentage of hosts transferring data at the rate noted
250KB/sec      52%
3.0MB/sec       3%
1.0-3.0 MB/sec 16%
75.1 MB/sec     7%
500-750KB/sec  10%
250-500KB/sec  12%
Source: Freeman, IDC, Dataquest

Data centers frequently put only one backup set on a given tape cartridge. This results in substantial under-utilization of tape capacity.

                         Tape Capacity Utilization
                           Backup Data Set Sizes
[less than]50MB 30%
800 MB           7%
400 - 800MB     15%
200 - 400MB     24%
50 - 200MB      24%
Source: Freeman, IDC, Dataquest

Direct SCSI or Fibre Channel connections between the host and tape device are used for 63% of mission critical database backups (higher reliability and speed). In contrast, 70% of general (non-database) file backup runs over the corporate LAN (less reliable, slower).

                 MISSION CRITICAL DATABASE BACKUP PROFILE
                          Directly Attached Backup Backup over the Network
Mission Critical Database         (63.0%)                  (37.0%)
1998                              (29.8%)                  (70.2%)
                          Central Administration
Mission Critical Database           --
1998                                --
Source: Database Operating Practices -- High
Availability and Data Protection, 8/98