To test backup and restore time and throughput, we used a test-bed consisting of a file server containing sample data, a connecting switch, and a backup server running the backup software. The backup server also held the database and media catalogs and provided the SCSI connection to our tape devices.
Our file server was an IBM xSeries 230 with a 1-GHz PIII and 1GB of RAM. The units ServRAID 4M Ultra160 controller was connected to a 64-bit PCI slot and configured for RAID 5. Each of the servers five hard drives ran at 10,000 rpm with 18GB capacity. The total data volume was 68GB and contained our 52GB workload. The data was handed off through a 64-bit Linksys Instant Gigabit Network Adapter EG1064 to a Linksys ProConnect II Layer 2 Management 4-Port Gigabit Ethernet Switch. Also connected to the switch was our backup server, a white-box unit with dual AMD Athlon 2200+ processors and 1GB of RAM. This server was equipped with a Linksys Instant Gigabit Network Adapter EG1064 and an Adaptec SCSI Card 29160 with an Ultra160 LVD interface. Both were connected via a 64-bit PCI interface for maximum performance.
Our test workload was a cross section of last years PC Magazine production data. We used 129,111 files in 12,151 directories, which amounted to exactly 52,483,388,473 bytes (just over 52GB). We chose Veritas Backup Exec 8.6 because it is one of the most advanced yet easy-to-configure backup solutions.
We attempted to back up file-server data that closely represented a real-world mix of environments. This included a significant number of files that exceeded the 100K size of typical Microsoft Word or Excel files. The largest files in our tests were 133MB.
To prepare our backup server for testing, we did a clean installation of Microsoft Windows 2000 Server, which included Veritas Backup Exec. We wrote this setup to an image file using Symantec Ghost Corporate Edition 7.0 and then loaded it before installing each new tape device. We also added a verify pass, and we chose settings that would minimize the amount of data logging.
Our first test involved simple data backup (with performance efficiency measured in several ways), and our second measured data restore speeds, representing each units ability to find specific data on the tape and then write it back to disk. The restore test used data that was not located sequentially on the tape; thus it required tape forward and seek operations. We restored 97 files in 18 directories, totaling 157,187,873 bytes (just over 157MB).
For the automated devices only, we ran an additional test, which recorded how fast each one could inventory four preloaded tapes. This required each unit to load every cartridge into the drive and read its label.
A Closer Look at
A Closer Look at the Results
In our tables, backup time represents the period during which data was actually being written to tape. During the verify time, the recorded data was checked against the original source. The job completion time is the total of backup time, verify time, and all other overhead; this includes calibrating, positioning, and rewinding the tape, as well as the time the backup software required to process the request.
Comparison of the results revealed both significant and insignificant differences. LTO Ultrium 1 and SDLT devices all scored relatively close across the board on the backup test. SDLT devices had a slight edge in backup and verify operations.
Matters became more interesting when we examined the performance of the Sony AIT-3 devices. Both the single-drive Sony SDX-D700C and the automated Sony StorStation AIT Library LIB-81 were far behind all the other products in backup and verify speeds. This could take on real significance if your data volume is extremely large and your backup window is very small.
But both devices restore speeds were far superior to those of the other products. And since restores are manual operations, this could save administrators a lot of time. When you have ten file-restore requests to process, the difference between 1.5 minutes and 6 minutes per request is no joke. This speed is due to Sonys Remote Memory In Cassette (R-MIC) technology, which uses a built-in chip to store catalog and file location information. This way, the tape need not be rewound for the device to read the catalog from the tape, as with DLT, LTO, or SDLT formats.
LTO Ultrium 2 is the newest technology in our roundup, and at testing time HP was the only manufacturer shipping it. This format greatly reduces backup time—and especially verify time—compared with LTO Ultrium 1. Yet when compared with Quantums SDLT technology, LTO Ultrium 2s overall performance advantage was not eye-catching. Restores were still considerably slower. Throughput can vary significantly depending on the mix of data, and most advanced backup applications offer plug-ins to back up databases or mail servers. Since such backups usually involve data from a consistent source, much higher speeds can be achieved. HP claims a native transfer rate of 30 MBps, a throughput we were not able to duplicate with our setup and data set.
The inventory test revealed relatively minor differences among the devices—with the HP StorageWorks 1/8 Autoloader in first place. But when we retested the Overland PowerLoader SDLT 320 with a second drive installed, it raced through the test in 3 minutes 50 seconds.