Benchmark tests show that the actual performance advantages of LTO over Mammoth-2 fall far short of the theoretical advantages.
By Jack Fegreus
The Linear Tape-Open (LTO) consortium has defined two new specifications in the mid-range to high-end tape-drive market. These specifications, which zero in on tape formats for media interchange, are being driven by Hewlett-Packard (HP), IBM, and Seagate. LTO drives compete directly with Quantum's SuperDLT and Exabyte's Mammoth-2.
The LTO Accellis format will be based on a two-reel tape cartridge somewhat similar to 8mm and 4mm cartridges. The LTO Ultrium format is based on a single-reel cartridge similar to DLT cartridges.
The first format into the LTO market is Ultrium. HP, IBM, and Seagate have all announced Ultrium drives. HP recently announced two Ultrium tape drives: the HP SureStore Ultrium 230e (tested here by OpenBench Labs) and the HP SureStore Ultrium 215, a half-height version that will be available in early 2001. (This month, HP began shipments of a family of Ultrium-based autoloaders and libraries, ranging from a native capacity of 900GB to 70TB, with native transfer rates of 15MBps to 300MBps.)
The Ultrium specification calls for a half-inch, linear, bidirectional format that uses a single-reel cartridge 20% smaller than a DLTtape IV cartridge. While the size difference may seem insignificant, in today's density-conscious computing environments, a 20% smaller size can make a big difference in automation environments.
More specifically, the Ultrium specification calls for eight read/write channels and a native cartridge capacity of 100GB. The specification does not define standards for reliability, form factor, power consumption, or performance. All of these important aspects are left open to promote competition among vendors.
However, when it comes to tape format, the next seven years of Ultrium are rigorously defined: In two years, the number of tape tracks are to increase from 384 to 580, as a result of the use of partial-response, maximum-likelihood (PRML) encoding. PRML is already being used on the current Mammoth-2 and DDS-4 DAT drives to reduce signal interference among the tightly packed tracks. In fact, the crosstalk on the densely packed Ultrium tape is enormous and easily overwhelms the data signal from the tape.
With PRML, the channel compares the measured signal from the tape with a known waveform to interpret the data. Exabyte was the first tape-drive manufacturer to use an enhanced Class IV PRML (EPR4) technique similar to that used by many hard-disk-drive manufacturers. According to the Ultrium road map, by 2007, tape tracks will increase to 1,024, the tape will be extended to 800 meters from the present 580 meters, and thin-film media will replace the current metal-particle media.
The Ultrium format records eight data tracks simultaneously. While data is being written, separate read elements are used to verify that the correct data has been recorded and can be recovered on each track.
The Exabyte Mammoth-2 puts four pairs of read/write heads positioned at 90° intervals around the drum. On each revolution of the scanner, the Mammoth-2 writes four tracks of data while simultaneously reading the previous four data tracks.
The construction of the tape head for Ultrium drives is derived in large part from the heads used in existing tape products. The head includes two modules, each of which contains a complete array of four servo elements and eight read/write elements. This array of thin-film elements is contained in a small chip in the center of the module. When the writers of one module are recording data, the servo elements and readers of the other module are operational. Each module has a flexible circuit that connects the elements to the read/write preamps.
Packing 384 data tracks onto such a small tape means that there can be only a bare-minimum distance between servo elements (which ensure the tape heads are properly aligned) and data elements. In fact, the servo-track budget is such that the average of two servo signals is calculated to align the data elements to the correct position for writing and reading. In particular, the tape is divided into four data bands, with 96 data tracks in each band. In addition, the outer edge of each band is bounded by a servo track.
Streaming from memory
The result is a very fast tape drive. Results of the OpenBench Labs benchmark, obltape v1.0, pegged base throughput without hardware compression at 13.9MBps-about 18% faster than the Mammoth-2. This benchmark allocates a large block of memory from which it streams data to the device. By streaming directly from memory, the benchmark eliminates bus-bandwidth contention with other devices. The data can be streamed in block sizes of 2n KB, where "n" ranges from 0 to 8.
The tape benchmark generates two types of data stream: purely random data and data that falls into a preset frequency pattern. The patterned data stream was originally devised and calibrated using Exabyte Mammoth-1 and Quantum DLT 7000 tape drives. This data stream consistently produces a compression ratio on the order of 1.9:1 to 2.1:1 on these devices. These numbers are highly comparable with backup results on real data that is not stored in a compressed format.
When we ran the compressible data stream with hardware compression, throughput on the HP Ultrium 230e soared to 32MBps. This represents a compression factor of 2.3:1, which is in line with what OpenBench Labs measured with HP's new generation of DDS-4 DAT drives, tested earlier this year. Like Exabyte's Mammoth-2, the Ultrium drive uses a new Adaptive Lossless Data Compression (ALDC) algorithm that purports to provide an average compression ratio of 2.5:1 across multiple data types. In addition, the HP Ultrium 230e implements what HP calls "smart data compression."
HP compression circuitry now has a "pass-thru mode" which switches off compression for non-compressible data. According to HP, this "pass-thru mode" can be at least 10% more efficient. This assertion can be explicitly demonstrated in the OpenBench Labs tests. When purely random data is sent to the drive while compression is on, the drive will attempt to compress this data, which is typical of highly compressed jpg, zip, and e-mail archives. In doing so, the drive will waste its own embedded CPU cycles trying to compress non-compressible data, and throughput will degrade.
In normal testing, this loss averages about 10%. With the Exabyte Mammoth-2, this loss fell only 5%. With the Ultrium drive, however, performance degradation fell little more than 1%.
In particular, the Mammoth-2 delivered between 27MBps with highly compressible data-which is not the case for the jpg, executable, tar, and zip files in our mix-and 11MBps with no compression.
For the Ultrium drive, this envelope is a little higher: For compressible data, OpenBench Labs was able to sustain 32MBps throughput. With no compression, throughput dropped to just under 14MBps.
We finished our testing by backing up a 5GB data set with 30,000 files. With Pkzip, these files could be compressed by a factor of two; however, this took approximately an hour on a system with a 600MHz CPU. With the BRU backup software, OpenBench Labs sustained 15.2MBps throughput using the Mammoth-2 drive, and 15.9MBps using the HP Ultrium 230e drive.
OpenBench Labs scenario
- Linear Tape-Open (LTO) Ultrium tape-drive performance
What we tested
- HP SureStore Ultrium 230e tape drive
How we tested
- Dell 2400 PowerEdge server running Red Hat Linux v7.0
- Adaptec 29160 controller
- Enhanced Software Technologies BRU v16.0 backup software
- OpenBench Labs obltape v1.0 benchmark
- The theoretical performance envelope of the Ultrium was about 20% faster than that of the Exabyte Mammoth-2.
- Practical backup performance of the Ultrium drive was only about 5% higher in a real-world backup scenario.
The LTO cartridge is similar in shape to the DLTtape IV cartridge but is 20% thinner, which increases the potential storage density of LTO over DLT.