|
Raiding The Stacks: Array Alternatives
Tape arrays--collections of drives in a single enclosure--are nothing new. They provide a powerful performance punch, letting you stream multiple jobs to the array at one time. Although the multiple drive libraries we tested can do the same, you'll pay a lot less for a multiple drive array than you will for a library. The disadvantage of arrays is you have to manually switch tapes and track rotation schemes.
But a new twist to arrays merits serious attention: the application of RAID-based tape backup. Introduced last year as a Cheyenne ARCserve module, the ARCserve RAID Option lets you configure a group of tape drives for RAID Level 0 (striping), 1 (mirroring) and 5 (striping with parity). Although fault-tolerant, RAID Level 5 uses an entire tape for parity. Mirroring, on the other hand, uses one tape for every tape you deploy. But the protection is worth it; lose a tape drive or a tape and you'll be able to rebuild the entire set. R
AID improves performance by writing to multiple tapes simultaneously--a process further enhanced in an array if you follow a ratio of two DLT7000 drives per controller card.
In addition to libraries, we also tested a tape array, Storage Solutions DLTa-Ray 210, priced at $44,665, shipped with four DLT7000 drives in a rackmount enclosure. It comes equipped with two removable chassis, each of which holds two drives, a power supply and two SCSI connectors. We attached each group of two drives to separate Adaptec 2940 Ultra-Wide controllers.
The DLTa-Ray 210 slightly outpaced the libraries when configured as a RAID Level 5 device under ARCserve, but it still didn't break any speed records. Our tests had the DLTa-Ray turning in a 268.47-MB-per-minute throughput on the large data set backup. But performance isn't the only point; with a RAID-configured array you get superior reliability at an excellent price.
In an interesting twist, Cheyenne added RAID support for tape libraries in version 6.5 of ARCserve.
We loaded 10 tapes and configured the ADIC Scalar 458 from Advanced Digital Information Corp. as a RAID device. ARCserve automatically set up two RAID sets (eight of our tapes). You can have as many RAID sets as your slots will hold (each set is comprised of as many tapes as you have drives). And, if you really want to scale, ARCserve will let you configure multiple libraries of the same type as an array.
How We Tested DLT Libraries
To test DLT7000 libraries, we used a Micron Electronics Vetix 2Lxi equipped with dual Pentium 200-MHz processors, 128 MB of RAM and three 4-GB SCSI drives as our server. Each library was attached to an Adaptec 2940 (single-ended) or 2944 (differential) SCSI adapter. Because the majority of the libraries shipped with only two drives, we chose to use one SCSI adapter for testing. Our mass-storage data source was a RaidTec Corp. FibreRAID Fibre Channel JBOD (Just a Bunch Of Disks) array, which was loaded with 51 GB of data.
T
o generate real-world performance data, we put each DLT (digital linear tape) library through a series of backup and restore procedures using Cheyenne's ARCserve version 6.5 for Windows NT. We also performed tape changing robotics testing.
Performance tests consisted of using ARCserve's built-in job reporting features to measure the speed in MB per minute of a variety of jobs. All tests were performed with compression turned off. The large data set file backup job consisted of a full backup of 51 GB of data. The image backup consisted of a full disk backup of one of our Micron's 4 GB drives, which had been loaded with a full 4 GB of data. Image backup creates a true sector-by-sector image of the disk. The two-stream file backup job used two drives, in tandem. Each drive backed up a directory (one from the RaidTec device, one from one of the server's internal drives) that contained 3.5 GB of data.
Restore testing consisted of three functions. We created a restore job made up of 16,193 files (1.1 GB), ran
domly selected from our 51-GB backup set. We used this 1.1 GB of data to force the units to read from a multitude of noncontiguous regions of two backup tapes. For the image restore job, we used the previously created image backup and restored it to a
blank disk of equal size as the original backup source. The two-stream restore tested how well the library dealt with multiple restore jobs that ran simultaneously.
To give each of the unit's robotics tape changers a workout, we used ARCserve's full and quick inventory tasks. The full inventory reads the barcode, loads each tape into a drive, reads the header and ejects the tape. The quick inventory scans the barcode labels of each tape and compares them to the information in ARCserve's database.
|