Redundant Tapes Mean Safe Storage
by Richard Adhikari (08/13/99; 9:00 a.m. ET), Planet IT

Do you know whether your archived data is safe?

Every day, like his counterparts in corporate IS everywhere, Vance Bufalo backs up copious amounts of corporate data.

Bufalo is senior engineer, Network Engineering, at Ameren (formerly known as Union Electric), the electric utility serving the greater St. Louis area, as well as Missouri, Illinois and part of Iowa. Bufalo backs up more than 100 gigabytes of data every weeknight from about 120 Windows NT, NetWare and Unix servers, and more than 1 terabyte on weekends when he does a full system backup.

What would happen if he, or anyone in corporate IS, lost any archived data?

Such losses aren't uncommon, says Mohamad Nourmohamadian ("Call me Mo Nour, it's easier for everyone"), president and founder of tape- and CD-ROM-controller developer Ultera Systems. Nour is also its CEO and chief technology officer. "Everybody has lost data in large archives with 4 mm and 8 mm drives," Nour says. And, as tape technology offers higher and higher capacities, the amount of archived data lost when a cartridge fails increases.

Cartridges now offer close to 100 GB of capacity, "more than some companies' whole databases," Nour says.

One way to avoid losing archived data is to use RAID-type subsystems with tape. Such technology is called RAIT (redundant array of independent tapes) or RAIL (in reference to tape libraries).

The RAIT Stuff

As RAID does, RAIT or RAIL lets you do two things: mirror your tapes, and create an array in which one tape is used to store metadata about the data on the other tapes. That way, even if one of the tapes in the array goes down, you can restore all the data.

"It's another way to apply RAID," says Farid Neema, president of storage analyst firm Peripheral Concepts. He describes RAID as "primarily a space-saving technology because, instead of duplicating the disk or tape, you have just one additional redundant element over four or five or seven drives."

The advantage of RAIT or RAIL is that the systems offer redundancy. "You don't want a backup operation to stop under any circumstances," and that can happen with stand-alone tape backups if there's a problem with either the drive or the cartridge, Neema says.

Ultera offers two tape-controller families that let you implement RAIT or RAIL. One, the Striper 2, lets you create RAID or RAIL subsystems. It has a 16-bit Ultra SCSI interface rated at 40 megabytes per second and implements data striping, where incoming data is recorded a byte at a time in parallel on up to four tape drives simultaneously. The fifth drive runs the parity tape, which is used to store metadata about the four others. Or, you can use Striper 2 in mirroring mode to make up to five copies at one time.

In other words, RAIT and RAIL subsystems implement various levels of RAID, only with tape instead of disks.

Striper Specs

Each tape drive in a Striper 2 array has its own dedicated, 10-MB-per-second channel, and, to the host computer, the cluster looks like a very large, very fast single drive. Striper 2 also has a data scrambling feature, which lets you load the tapes in any order; has flash memory for easier updates; and offers support for a GUI for tape-subsystem management.

The scrambling feature eliminates one of the problems posed by RAIT and RAIL arrays: the need to load tapes from which data is being restored in the order they were loaded when the backup was made.

Striper 2 controllers are compatible with all commonly used tape drives, including 8 mm, AIT, Mammoth, DAT and DLT.

Ultera also offers plug-and-play mirroring controllers, in its Imager family. These can mirror the data being backed up onto two drives or two autoloaders at the drives' or autoloaders' top speeds. By cascading the Imagers, users can produce four, six or more copies without sacrificing backup speed, according to Ultera.

There are software-based RAIT and RAIL implementations available, but Ultera's hardware-based solution is preferable because "software-based RAID on tape uses host cycles, so it can impact performance," says Fara Yale, chief analyst for Dataquest Computer Storage Service at Gartner Group/Dataquest's San Jose, Calif., offices.

It's still an education sell, though. Once corporations understand the implications of RAIT and RAIL, they buy the product, Nour says. "I used to personally go on sales pitches with our partners, like Andataco, and potential customers would refuse to pay $70,000 for a subsystem just because we had RAIT, even though we'd demonstrate that if you turn off one of the tape drives or eject a tape the backup would continue," Nour recalls.

"Then I'd tell them this is an archiving or offline system, so if you have a bad cartridge and can't restore it, that won't be a problem, and suddenly they'd be interested." He'd then do another demo, where data was backed up on five tapes but only four were loaded for a restore, and the restore would be successful.

That's when the potential customers would sign up at once.

Mirror, Mirror

Most corporate users look on Striper as a handy device for mirroring data to cope with their shrinking backup window.

Take Ameren's Bufalo, for instance. His main concern is being able to back up all his data nightly and, in keeping with corporate policy, make a second set of backup tapes to be shipped off-site by 11 a.m. on weekdays.

Sheer logistics make it impossible for him to complete his backup and clone the backup tapes for an off-site set by 11 a.m. Two of his three backup servers, Windows NT boxes with STK 9714 tape silos, service more than 50 clients each within the corporate firewall and also are used for tape recovery by about 1,500 clients in the main office complex during the day. This means they can't be used for backups after 9 a.m.

It takes Bufalo 12 hours, from 6 p.m. to 6 a.m., to create the first backup tape set. That gives him only three hours to clone them.

There just wasn't enough time to do the job, so Bufalo decided to mirror the backup tapes, generating both sets at the same time. Backup software offered by tape vendors didn't meet his needs. "Legato [Systems], Cheyenne, Palindrome and all of the vendors offer cloning where, after a tape is backed up, they do a tape-to-tape copy for you, but no one offered tape mirroring," Bufalo says.

Then Bufalo turned to Ultera, whose products he was already using on other servers, DLT 4700s, which are seven-tape, one-slot silos. He got Ultera to custom-design a subsystem. This is a 19-inch rack unit with space for eight individual RAIT controllers, each of which is used for one set of drives, and it mirrors tape libraries.

When Bufalo backs up data at night, both STK 9714s automatically load their tapes and mirror the data being backed up. Each drive and the robot are mirrored on their own controllers, for redundancy. "As an electric utility, we're a 24x7 shop and take our tape backups extremely seriously because we're always looking out for the disaster-recovery scenario," Bufalo says. He's simulated a disaster, got the hardware vendor to deliver a new server, and restored data from off-site tapes successfully.

Worth Every Penny

This robustness didn't come cheap, but to Bufalo it was worth the cost. "The Ultera is a fairly expensive solution. It ran us about $6,000-$7,000 per channel and we have five channels on one server and six on the other," he says. "But you have to remember our requirements: Be able to meet our backup window and be able to ship a copy off-site in time."

Now, the off-site tapes are ready for shipping by 8 a.m. every day.

Bufalo also likes Ultera's "very high degree of support." Yet he doesn't have any use for the tape-striping features of the Striper 2 because "up to now, our tape drives can go faster than we can pump data to them" with the mirroring.

Analysts agree that RAIT and RAIL are useful for disaster-recovery planning. They're "great ways to make duplicate sets of your tape cartridges to send off-site for your disaster-recovery programs," says Gartner/Dataquest's Yale.

Peripheral Concepts' Neema agrees. "Most people are using RAIT for mirroring," he says. "I've seen relatively few RAIT applications like [Level] 3 + 1 [where they're actually using one tape in a subsystem for parity]."

Although RAIT has been around for about five years, it hasn't really taken off because the technology wasn't adequate, Ultera's Nour says. "Up to now, neither the service nor the RAIT system was fast enough to keep up with a single tape drive, let alone five tape drives in parallel," he says.

But server subsystems have gotten faster. Also, the size of backups has increased tremendously, and that's driving the demand for RAIT and RAIL. "People are paying a lot of attention to tape backup at the OEM [original equipment manufacturer] level because they have to reach backup numbers for databases, which are huge," Nour says.

Is your archived data safe?

- Planet IT

For further information, call (949) 367-8800, email inform@ultera.com or visit Ultera's web site at http://www.ultera.com.