December, 1996
Put Up Your Jukes!
It's the Great Storage Face-off - 5¼" vs 12" optical jukeboxes. But the fight is fixed. There's a sure winner: The user organization that chooses the right media size for their application needs.
Storage choices abound. In optical jukeboxes, two main media sizes (5¼" and 12") yield a wide variety of costs and configurations. In general, 12" jukes cost more, hold more per disc (12 GB vs 2.6 GB), keep more data online and support WORM only, not rewritable.
But which is best? Which combination of media and device is the "fastest"? What kind of performance can you expect in a production setting?
To find out, Doculabs compared two storage devices: a Hewlett- Packard (CITY /STATE/ PHONE???) SureStore Optical 600fx 5¼" jukebox with the new higher-capacity 2.6 GB 5¼" discs, and a Philips LMS (CITY/ STATE/PHONE???) 12" juke, their LF6602 RapidChanger20.
We looked at how juke design, retrieval and data delivery speed, capacity, cost and media. Consider these factors when choosing a jukebox.
How and What We Tested
We tested retrieval times by simulating a real-world app. We sent multiple requests for 2.9 MB TIFF images to the storage subsystems at once. Both devices held about 46 GB, with images stored at random on the 17 HP and four Philips platters.
The jukes were attached to a Gateway 80 MB P133 with Pegasus 1.02 optical driver for Windows NT (which may have limited the performance of both devices), Adaptec 2940 SCSI and a Novell 4.1 10baseT Ethernet LAN. To test the performance of the devices themselves, we turned off caching.
A single-tasking test measured retrieval of 2.9 MB images from each device. A multitasking test sent 50 retrieval requests in batch mode to each device. We ran each test against each storage device five times.
We also tested two single-tasking scenarios: heavy swapping (lots of random retrievals, mostly on different discs each time) and light swapping (most retrievals from the same disc or from discs already loaded in the drives). In the multitasking test, requests came in simultaneously rather than sequentially. Each device got 20 simultaneous requests.
What We Found
Both devices performed about the same under heavy swapping. The 12" Philips handled 50 requests in 10:58 minutes (ten minutes and 58 seconds), averaging 13.2 seconds per request. That's marginally better (2.2% faster) than the 5¼" HP unit, which took 11:13 min total, 13.5 sec average.
But under light swapping, the Philips took 6:38 min total (8.0 sec avg) -- 7.5% slower than the HP's 6:08 min (7.4 sec avg).
Decreasing the number of swaps -- by requiring more sequential retrievals -- greatly sped up both machines. The HP improved by 45.2%, the Philips by 39.4%.
In the multitasking test, the Philips took 1:49 min to process 20 requests (5.5 sec avg). That was 53.8% faster than the HP, which needed 3:59 minutes -- more than twices as long -- to handle all 20 requests (avg 11.9 sec).
Why We Got These Results
Three primary factors affected performance:
1. Jukebox mechanics and architecture. Disk swap time is the most time-consuming part of retrieval. The design of the swapping mechanism makes a huge difference.
The HP juke uses a robotic arm to pick the disc from a slot and mount it. A dual picker move diagonally between columns of discs. HP claims their juke averages six seconds to pick a disc from the slot and mount it in a drive.
Instead of a robotic arm, Philips uses a "RapidChanger" shuttle that they claim swaps discs in 2.5 seconds on average. Discs travel less to get to the drive head than in HP's box. Also, no flip is required -- the dual heads access both sides of the media.
2. The amount of data under head. Once the disc is mounted, access and transfer are relatively fast. The bulk of total retrieval time consists of disc swapping. For this reason, the more data is available under the head at a given time, the faster retrieval will be.
The amount of data available under the heads is determined by media capacity (same amount of data spread over fewer disks) and the number of drives and drive heads (dual heads don't require a flip).
Media Capacity: The more data per disc, the fewer swaps per retrieval. Result: Better performance.Single-Head vs Dual-Head Drives: Optical discs store data on both sides. A single-head drive can only read one side of a disc at a time, so only half the disc's capacity is accessible at a given time. Accessing the rest requires a physical flip of the disc.If performance is critical, look for a device with robotics and architecture designed for efficient swapping. Also, look for a device that makes as much data available under head as possible to minimize disk swaps.
3. Single vs multitasking. A multitasking environment can initiate a thread for every request posed to the jukebox. Multiple requests can be received and processed by the jukebox at once. This lets users to things like resequencing requests to minimize swapping.
Note that the HP performed slower in the multitasking environment, mainly because we used a single-processor PC. Operating systems like Windows NT scale up easily to run on a two- or four-processor machine. This can improve performance in large access-intensive apps with a high retrieval frequency. Magnetic cache was not used in testing. This would have made both products much faster in all tests.
Related Articles:
