|
SPREADING THE DLT PERFORMANCE SPECTRUM From the ValuSmart Blade 640 to the SDLT 320, Quantum extends the DLT value proposition. |
||||
by
Jack Fegreus |
|
Quantum last month announced that the company is acquiring Benchmark Storage Innovations, the company that Quantum helped fund to satisfy the SOHO end of the market with low-cost high-value drives. During the same time frame, Quantum extended the enterprise end of the DLT technology spectrum with the introduction of Super DLT 320. This latter introduction significantly improves the top end of SDLT performance. Devices on both ends of the Quantum DLT spectrum were tested by openBench Labs and the results render an interesting price/performance profile. While many of the specifics of the Benchmark acquisition have yet to be announced, Benchmark’s ValuSmart line of products will probably be rebranded as the Quantum ValuSmart line. At the foundation of that ValuSmart line is the half-height ValuSmart Tape 80 (VS80) drive, which provided the first DLT alternative to 4mm DAT DDS technology as a built-in backup device on PC servers. For media, the VS80 uses the same DLTtape IV cartridges also used by Quantum DLT4000, DLT7000, and DLT8000 tape drives. This differs from the current top end of DAT technology dubbed DDS4, which requires a special DDS4 cartridge to record at full speed. |
|
|
||||||
|
Unfortunately, you can’t just pop one of the more than 70 million existing DLTtape IV cartridges into a VS80 and expect full functionality. The VS80 writes data in DLT1 format: 168 tracks of data having a bit density of 123K bpi, which translates into 40GB of data per cartridge without compression. As a result, a VS80 drive can read but not write a DLTtape IV cartridge that was previously written in DLT 4000 format, which has a density of just 82K bpi. Any cartridge must be either blank—brand new or thoroughly bulk degaussed—or previously written to using another VS80 or DLT1 tape drive. Benchmark utilized the VS80 in the ValuSmart Tape 640 Blade, the first rack-optimized 2U DLT autoloader to be introduced. A carousel containing eight DLTtape IV cartridges rotates around a center-mounted VS80 drive. The result is a DLT autoloader that has a street price of around $3,500, which is less than a standalone SDLT or LTO drive and about $2,000 less than a six-cartridge DDS4 DAT autoloader. As a result, the ValuSmart Tape 640 Blade provides a very cost-effective way to take advantage of the basic benefits of tape automation: random access to multiple cartridges and the ability to change tape without human intervention.
The default mode is stacker or sequential mode. In this mode, the ValuSmart Tape 640 Blade automatically unloads the current tape cartridge when it becomes full and replaces it with the cartridge in the next highest slot if additional tape capacity is needed. Additionally, the operator can put the autoloader in ‘circular’ mode, which will automatically reload the original cartridge when all of the other cartridges have been cycled through. |
|
If the ValuSmart Tape 640 Blade detects that the SCSI controller is using multiple LUNs to address the device, then the switch puts the autoloader into random mode. In this mode, the ValuSmart Tape 640 Blade does not automatically load the next tape cartridge into the drive. Instead, the autoloader waits for a command from the software to load a specific tape. To test the performance of the ValuSmart Tape 640 Blade and later an SDLT 320 drive, openBench Labs used its oblTape v1.0 benchmark. We ran the tests on a dual-processor HP Netserver 1000 running under both Linux—SuSE Linux 8.0 with the 2.4.18-4 kernel— and Windows 2000 Server. For a SCSI connectivity we used a QLogic Ultra160 SCSI Host Bus Adapter (HBA). Our tape benchmark generates two very different types of data stream: purely random data and data that falls into a preset frequency (compression) pattern. The benchmark’s patterned data stream was originally calibrated using Quantum DLT 7000 tape drives, which implemented the Digital Liv Zempel (DLZ) compression algorithm. That algorithm provided an approximate 2-to-1 compression ratio on normal data. OpenBench Labs devised a means of generating patterned data that consistently produced a compression ratio on the order of 1.9-to-2.1 on those DLT 7000 devices. 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 data 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 often use 128KB reads on Linux. For that reason, we chose to use 64KB data blocks on tests run on Windows 2000 Server and 128KB data blocks on Linux tests. In our benchmark tests, the VS80 in the ValuSmart Tape 640 Blade performed just as we expected it would: It behaved like a DLT1 drive. Under both SuSE Linux 8.0 and Windows 2000 Server, native performance with no hardware compression was a consistent 2.87MB per second. Interestingly, this is exactly the same base throughput that we measured with the HP DAT40e. We measured the openBench Labs worst-case scenario of hardware compression with incompressible data at 2.5MB per second. Once again, this performance was identical with the worst-case performance of the HP DAT 40e. |
|
|
It was at the high end where Linux and Windows performance diverged on the VS80. The VS80 showed a high degree of sensitivity to the size of the data blocks being transferred. In both cases, the results of data compression were low when compared to the new generation of drives such as SDLT. Using 64KB blocks, compressed data performance on Windows showed just a 1.25X improvement by clocking in at 3.59MB per second. Using 128KB blocks under SuSE, performance improved by 1.48X to 4.24MB per second. It was the HP DAT40e, however, that reached levels of data compression never before measured by openBench Labs. On both Windows and SuSE, we pegged throughput for our compressible data stream at 7.62MB per second. Remarkably, throughput soared by a factor of 2.66X. While we never got to this level of compression when running a set of actual disk backups, compression and throughput remained consistently higher on the HP DAT40e in every test. We traced the extraordinary performance exhibited on the highly compressible patterned data stream to improvements in the SCSI firmware and the larger buffers in the DDS4 compression engine chain. Using a firmware-based "intelligent disconnect," the DDS4-4 drive optimizes the time to burst data at 40MB/second into a mini-buffer which is used to hold the data before the compression engine. An 8MB buffer then holds up to 20 compressed DDS "groups" of data. When the drive is given an opportunity to access the bus again, the compression and group buffering is held off so that available data can be loaded into the drive and processed. As a result of the tuned SCSI bandwidth utilization, the HP DAT40e can move data through the compression chain faster. |
|
|
On the opposite end of the performance spectrum, SDLT 320 represents the latest entry in nine generations of Quantum drives. This new SDLT model introduces changes in two key design areas. Compared to the older generation of SDLT drive, an SDLT 320 drive writes data 45% denser on tape: 193K bpi. By way of comparison, the data density on an LTO tape is 124K bpi. This makes the areal bit density of an SDLT 320-formatted tape roughly 56% greater than that of an LTO-formatted tape. In addition, Quantum sped up the tape movement from 116 to 122 inches per second. What's more, the SDLT 320 enters the market with a street price that is just under the price of competitive LTO drives like the Seagate Viper 200. By far and away, the most important take-away from this discussion is that we are talking about an SDLT I type tape. In other words, the SDLT 320 does not require any new media investment by current SDLT customers. All of the advantages introduced by the SDLT 320 are immediately available to any site currently using SDLT I tape cartridges. SDLT320 drives are also backwards-read compatible with DLT type IV media, thereby bridging the DLT-to-SDLT generational and technological gap. With data that is not compressed, a single LTO cartridge can pack 100GB of information. The first- generation of SDLT drives shoehorned in 110GB of information in a similar sized format. With the Quantum 320, those exact same SDLT I tape cartridges become cavernous vaults for 160GB of information. And now, if we assume the standard metric that, on average, data stored on disks can be compressed by a 2-to-1 factor by the compression hardware on today’s tape drives, then each SDLT I cartridge can theoretically hold 320GB of data when an SDLT 320 drive is employed. There is one caveat, however, concerning the reuse of SDLT I tapes previously written with an older model SDLT drive, which is now dubbed the SDLT 220. Because of the differences in tape density and recording speed, the SDLT 320 has two running modes, which are automatically set by the density of data recorded on the tape. If a blank tape is placed in the drive, the drive naturally defaults to SDLT 320 mode. If a tape previously written with an older SDLT drive is loaded, the SDLT 320 defaults to SDLT 220 mode even if it reinitializes the tape. Special software is needed to force a mode change on the drive. On Windows, this is done by the required special device driver. Linux, however, has no need for such a driver and so that task will have to be accomplished with a backup software package that knows about the SDLT 320. |
|
|
||||||
|
|
The results we measured using oblTape proved to be quite interesting on a number of levels. First and foremost, there was some expectation among advocates of both Linux and Windows 2000 that there would be inherent measurable differences in the streaming performance of each tape drive as it moved between the two operating systems. For all practical purposes, there was none. Using fresh media,
with data compression turned off and writing 128KB data packets on SuSE, we measured native throughput of the SDLT
320 to be 15.6MB per second. Using 64KB data packets on Windows 2000 Server, the throughput was identical, 15.6MB
per second. Turning compression on and sending packets of random incompressible data dropped throughput to 12.2MB
per second. Switching to our patterned data stream, the throughput performance rose by a factor of 2.06X to 32.1MB
per second on both SuSE and Windows 2000 Server. These performance results pegged the Quantum SDLT 320 as the best
performing half-inch single-reel enterprise-grade cartridge drive. |
|