Running BakBone's NetVault software with an Exabyte X80 tape library yielded impressive backup and restore performance.
By Jack Fegreus
OpenBench Labs tested BakBone Software's NetVault backup software in a high-end e-business scenario. We installed NetVault on a dual-processor Siliconrax-Sliger Rax2100 Web server running version 7.0 of Red Hat Linux and connected a rack-mountable Exabyte X80 library via a QLogic QLA12160 Ultra160 SCSI host bus adapter (HBA). We expected to see dramatic backup results.
We got a lot more than we bargained for.
The premise of this test was to examine how easily a data center could automate storage functions on a server running Linux. Tape-library automation has long been highly regarded as the best way to avoid the cost of juggling numerous devices on the network and manually handling media. This ability to automate the entire data-management process for highly reliable, unattended backup is precisely why tape libraries cost so much less per megabyte than standalone tape drives.
Realizing these benefits, however, requires a sophisticated marriage of hardware and software. Simply plugging an automated library such as the Exabyte X80 into a SCSI HBA and launching a backup program not designed for library automation are equivalent to wheeling a V12 Barchetta into your local service station for a tune-up.
The Exabyte X80 can be configured with two to eight Mammoth or Mammoth-2 (M2) tape drives and 40 to 80 data cartridges. With 80 cartridges, the configuration represents a native storage capacity of 4.8TB, which in practice, turns into 10TB to 12TB, depending on the compressibility of the data. The library sports a number of well-designed features, including an LCD menu display for configuration and status information, a five-cartridge access port to add or remove data cartridges without interrupting library operations, and an Ethernet port for future browser-based library management.
More importantly, the most impressive aspect of the X80 is its robotics. The big challenge in tape-library design is to minimize the distance traveled by robotic arms to speed cartridge picking and-even more importantly-speed the cataloging of the data-cartridge library. There is nothing more boring than listening to the high-pitched whine of robotics arms as they travel up, down, left, and right scanning bar codes on the tape cartridges and then plugging them into the drives one by one to verify the contents of the tape.
To ease the system administrator's pain during the tape-inventory process, the X80's data cartridges are placed in eight removable magazines-each of which contains five tapes-on a rotating carousel. A laser bar-code scanner easily reads the rapidly spinning carousel and, picking a cartridge from the carousel, dramatically reduces linear arm travel. This makes the X80's robotics very fast. In normal operation, the X80 can pick and deliver a tape cartridge in less than 10 seconds, and the M2 drive can position the tape in the correct location in less than 17 seconds. That's less than 30 seconds to identify a tape and get a backup or restore job into process.
Getting this all into play with NetVault turned out to be far easier than we anticipated. In its incarnation under BakBone Software, all of NetVault's technical pyrotechnics are hidden away under two simple X-Windows GUIs. The easy-to-use nvconfigurator takes care of the standard configuration data, and the nvgui module is impressive in its logic and simplicity. Even the most recalcitrant operations staff will have no problem with nvgui.
OpenBench Labs is compelled to offer one essential tip: Change the buffer for each tape drive, which defaults to just 257KB. With 512MB of RAM in our server-you can never have enough-we set each drive's buffer to 30,721KB.
The nvgui interface comes up with eight central options for backup, restore, client management, device management, status monitoring, media management, job management, and log viewing. We easily added the library via the client management option. It's a simple process in which NetVault scans for attached libraries; the systems administrator verifies the identity on any libraries found; NetVault scans the library for tape drives; and the systems administrator assigns the appropriate drives to open bays in the library. Now it's automatic-inventory time.
Once the initial inventory is over, the device-management screen displays the library, its drives, the slots with the identity of any inserted cartridge-both the tape label and the bar-code number-and the five entry/exit ports for moving tapes in and out of the library. From here, tapes can be jockeyed within the library, safe from the touch of erring human hands.
Tape handling gets even more sophisticated within the deceptively simple backup interface: Choosing a set of files for backup is a simple point-and-click exercise within a file tree. Then a target backup device is chosen. Here, radio buttons are used to implement sophisticated rules for triggering which tape the library chooses for the backup. This significantly reduces the risk of the wrong cartridge being put into a drive, or worse, the chance that a backup is not performed.
With the library set up, all that was left was to measure the performance of the NetVault-X80 combo. With our robust buffer setting, we sustained an average of 17.6MBps backing up 4GB of data. During the process, throughput peaked at just over 27MBps, as predicted by our obltape benchmark. With two backups running simultaneously, aggregate through-put averaged 25.5MBps.
It was on restores, however, that Net-Vault delivered a whole new class of performance: With an average throughput of 14.7MBps, NetVault was blowing through backups and restores like a Barchetta blowing off a Beetle.
OpenBench Labs scenario
- Tape-library automation hardware and software
What we tested
- BakBone Software's NetVault v6.0.1 backup software (www.bakbone.com)
- Exabyte X80 tape library (www.exabyte.com)
How we tested
- Siliconrax-Sliger Rax2100 Web server running Red Hat Linux v7.0 (www.siliconrax-sliger.com)
- QLogic QLA12160 Ultra160 SCSI controller (www.qlogic.com)
- OpenBench Labs obltape benchmark 1.0
- OpenBench Labs 4GB backup file set
- NetVault's sustained throughput on restore operations was comparable (approximately 85%) to its throughput on backup.
- With multiple backups running simultaneously on two drives, throughput rose by 45%.
The benchmarks are available free and can be downloaded from www.openbench.com.