Hewlett-Packard's StorageWorks Ultrium 460 racks up impressive capacity and performance scores. By Jack Fegreus
By Jack Fegreus
"Super drive" is the new buzzword in the tape storage market. According to industry analysts, super drives have tape cartridges that can hold at least 100GB (uncompressed) of data. In that category you'll find Mammoth 2 from Exabyte, SuperDLT (SDLT) from Quantum, SuperAIT (S-AIT) from Sony, and Ultrium format LTO drives from Hewlett-Packard, IBM, and Seagate.
One of the most recent entries in the super drive group is HP's StorageWorks Ultrium 460, which is built on the second generation of Ultrium technology—Ultrium 2.
Enhancements to the Ultrium 2 specification over Ultrium 1 can be found in many of the same areas that Quantum made to SDLT 320 over SDLT 220: primarily in track density and tape speed (see "DLT tape scales upward with SDLT 320," InfoStor, March 2003, p. 40). Using the same media as Ultrium 1, Ultrium 2 lays down 512 data tracks instead of 384 tracks for a 33% increase in track density.
Figure 1: The obltape benchmark helps define a performance range for tape drives by marking three critical performance points. The first point is the native throughput rate of the drive with no boost from data compression firmware. Then, using a stream of patterned data that was calibrated to produce a compression ratio on the order of 1.9:1, we determine an optimistic practical upper level for performance. Finally, we determine a pessimistic performance level by streaming random non-compressible data to the drive.
The adaptive tape speed (ATS) range on the Ultrium 2 has been increased across the board—from the minimum to the peak streaming tape speed—by about 40%. All Ultrium drives use an ATS algorithm that monitors the host's data rate to vary the tape speed to keep the drive's data buffer full or empty (depending on whether a restore or a backup is in progress), and thereby minimize any mechanical stops and starts to reposition the tape. At the same time, the main buffer on the HP StorageWorks Ultrium 460 drive has been increased to 64MB from the 16MB used in the earlier StorageWorks Ultrium 230.
Given just these changes, the areal bit density of Ultrium 2 formatted tape is approximately 86% greater than that of the earlier Ultrium 1 format. All things being equal, the change in areal bit density would nearly double the throughput specs of the new drives over the first generation of Ultrium drives. But all things are not anywhere near to being equal.
Increasing the areal bit density dramatically increases the probability of errors at the tape head. These errors would then result in an equivalent increase in tape repositioning, which would leave throughput rates at the level of sludge-like ooze. As a result, the real operating speed within the range of the ATS feature is dictated by the ability of the read-and-write channels to cope with lower electrical signals that result from physically smaller magnetic bits packed more densely on the tape that is now passing by the read/write head more quickly than before.
To solve this potential problem, the Ultrium 2 design employs a digital signal processing technique in a PRML-4 (Partial Response Maximum Likelihood) read channel, which provides more-accurate tracking than before. In essence, the channel compares the measured signal from the tape with a known waveform to correctly decode weaker signals and read/write data at a higher bit density. As a result, the throughput specs for Ultrium 2 are double those of Ultrium 1. Throughput of streaming uncompressed data peaks at 30MBps. In a typical backup scenario, expected throughput will be in the range of 50MBps to 60MBps with compression. This means that each Ultrium 2-formatted half-inch cartridge will hold 200MB of uncompressed data and more than 400MB of compressed data.
Figure 2: The HP StorageWorks Ultrium 460 had no problems with backward-compatibility using HP, IBM, and Seagate tape cartridges formatted as Ultrium 1. Unlike many backward-compatibility schemes, the Ultrium 460 continues to function as an Ultrium 2 drive in terms of tape speed. As a result, it distinctly outperforms Ultrium 1 devices when using Ultrium 1 tapes.
The Ultrium 2 format is designed to allow an Ultrium 2 drive to use older Ultrium 1 cartridges. The ability to read Ultrium 1 formatted tapes, combined with a higher ATS speed range, pegs top non-compressed (native) data throughput on the Ultrium 460 drive using an older Ultrium 1 tape cartridge at 20MBps. This compares with 15MBps for an Ultrium 230 drive.
To test the performance of the HP StorageWorks Ultrium 460 drive, we calibrated the drive using our obltape benchmark. The benchmark generates two very different types of data streams: purely random data (which is not compressible), and data that falls into a pre-set frequency pattern (which is explicitly designed to produce a compression ratio in the range of 1.8:1 to 2:1). We ran the tests on the same hardware running both SuSE Linux 8.1 and Windows 2000 Server.
The obltape benchmark starts by allocating a large block of memory from which it then streams patterned or random data to the device. By streaming data directly from memory, the benchmark eliminates bus bandwidth contention with other devices. In this way, the benchmark more accurately represents the data-transfer rate of the tape device than the overall system throughput. The data can be streamed in block sizes of 2n KB, where n ranges from 0 to 8. This simulates the differences in the way backup applications read data off of a disk drive. In particular, high-end backup applications tend to use 64KB reads on Windows and 128KB reads on Linux. For that reason, we used 64KB data blocks on tests run on Windows 2000 Server and 128KB data blocks on Linux tests.
All testing of the HP StorageWorks Ultrium 460 was conducted on a dual-processor HP Netserver 1000. We used a QLogic Ultra160 SCSI host bus adapter (HBA) to connect to the tape drive.
The results proved to be quite interesting on a number of levels. First, while we have previously measured higher streaming I/O rates from RAID drives when running on Linux, when it came to measuring streaming I/O from tape there was an insignificant difference between Linux and Windows.
This was definitely not the case with the Ultrium 460. Using fresh media, with data compression turned off and using 128KB data packets on SuSE, we measured uncompressed throughput for the Ultrium 460 at 29.6MBps. Using 64KB data packets on Windows 2000 Server, the throughput lagged by about 12%—at 26.1MBps. By way of comparison, using an SDLT 320 drive with the same equipment, we measured native throughput at 15.6MBps on both Linux and Windows.
Turning the Ultrium 460 tape drive's hardware compression circuitry on and sending packets of patterned data brought throughput up to 55MBps on SuSE with a compression ratio of 1.86:1. On Windows 2000 Server we measured throughput at 48.9MBps with a compression ratio of 1.87:1. These performance results easily placed the HP StorageWorks Ultrium 460 as the best-performing half-inch, single-reel cartridge drive.
Perhaps even more remarkable in their implications for real-world performance were the results measured when we tested our worst-case scenario: data compression turned on at the drive while it receives a stream of non-compressible data. The most typical example of this comes when .gz or .zip files are included in a backup set. This event disrupts data flow and can cause a repositioning halt as the drive wastes internal cycles trying to compress the data. In the past, we have measured a drop in throughput of about 10% when streaming random data to the drive with compression on as compared to having no compression (i.e., native throughput).
On both of the Ultrium 1 drives we previously tested—an HP Ultrium 230 and a Seagate Viper 200—detection circuitry built into the drive signaled the drive to turn off compression and we measured a drop of only 0.5% in throughput. On Ultrium 2 drives, all of the circuitry has been redesigned and separate ASICs are used for the interface, servo system, formatter, digital signal processor, and read/write pre-amplifiers. The result of all of this wizardry is no measurable loss in throughput during any of our tests sending non-compressible data to the drive when compared to native uncompressed throughput.
For a final test, we ran the Ultrium 460 on Linux with an Ultrium 1 formatted tape cartridge. We measured uncompressed native throughput at 19.9MBps, compared to 14.2MBps using the Seagate Viper 200. With hardware compression on and streaming a compressible data flow, we measured throughput rates of 37.9MBps on the Ultrium 460 and 28.7MBps on the Viper 200.
Divorced from the constraints of disk drive performance and software overhead, the HP StorageWorks Ultrium 460 sets a high-water mark for backup performance. Clearly, backing up data from single-spindle drives to the Ultrium 460 will be sheer folly when it comes to performance. Nonetheless, as we have seen in numerous RAID performance tests, ensuring a minimum 60MBps throughput rate off a RAID 5 stripe set is not that big of a challenge.
The interesting challenge will come for software. For the first time in a very long while, an off-the-shelf tape drive has high enough throughput levels that will not mask inefficiencies or unnecessary overhead in backup software packages.
Jack Fegreus is lead analyst for the CCI Group, which conducts research studies on the functionality and performance of IT software and hardware. He can be contacted at JFegreus@CCIcommunications.com.