December, 1996
Contemplating the "Big Boys" of Optical Storage
Seldom seen and little understood, large-format jukebox systems labor at the heart of many mountain-sized applications in a range of industries. If your storage needs run - or will run - to the extra-large, you'll find one of these systems in your karmic corner.
Large-format optical jukeboxes are the unsung heroes of data storage. They toil quietly behind the scenes in banks, insurance companies and government agencies. Their role in most organizations is mission-critical. Their fortý is storing huge amounts of information.
Yet unless you actually use them, you probably don't know much about large-format jukeboxes. It's only when you sit down for some serious long-range capacity planning that the limitations of competing storage systems become glaringly apparent. Suddenly, those big optical jukes seem like a very bright idea.
Pick Your Size: LP or 45?
Jukeboxes, like the drives to which they provide shared access, come in two main sizes: 12" (about the size of a long-playing record -- remember vinyl?) and 5 1/4" (the size of an old 45 RPM disk). (One vendor, Kodak, makes a 14" juke.) These systems differ in three important ways.
1. Capacity. Jukeboxes with 12" drives store 8-15 GB per disc. Jukes with 5 1/4" drives store up to 2.6 GB/disc. The heftier media translate into much larger jukebox capacity.
How much larger? The upper limit on most single 12" jukeboxes is over one TB (1,000 GB). For single 5 1/4" jukes, the upper limit is about 600 GB.
You can chain many 5 1/4" jukes together to store more. But this makes the system harder to integrate and administer. It can also reduce performance and reliability.
2. Performance. Since each disc stores more, 12" jukes put more data under the drive's read-head. This data is essentially online. It's immediately accessible. The more data you get "under head," the less you need to swap discs to find the data you want.
Think of it this way: A 15 GB 12" disc holds about as much as six 2.6 GB
5 1/4" discs. But data under head with the 12" drive can actually be 12 times as much. That's because some 12" drives are dual-headed. Heads on either side of the drive can access data -- without flipping the disc.
All current 5 1/4" drives are single-headed. You have to flip the disc to read the other side. So there's only 1.3 GB of data under head.
Why is all this so important? Because disc swapping is by far the biggest drag on jukebox performance. With lots of data under head, it's much more likely that the data you want will be on the same disc that someone else just accessed. This is particularly true if most of your data requests are for files stored within a given time period -- the previous 30 days, for example. The result? Less disc swapping. Faster retrieval. Higher productivity. A better bottom line.
3. Permanence. Most 12" jukeboxes hold WORM (write once, read many) drives. They use a write process that's as permanent as you can get with electronic storage. Your data is very secure.
Most 5 1/4" jukes use a magneto-optical (MO) write process. MO media is rewritable. So it's not as permanent as 12" WORM media. Your data is less secure on MO.
Large-Format Jukebox Drawbacks
Despite their advantages, large-format optical jukes aren't for everyone. Three limitations:
1. Lack of standards. Discs created with a 12" drive from vendor A won't work in 12" drives from vendors B, C or D. So if you're using a jukebox from vendor A and vendor D comes out with a better box, all you can do is look at it longingly. Unless, of course, you want to convert all your discs so they'll work with the new box. Fun stuff.
The lack of standards also means the drive you buy today probably won't read the higher-capacity media your vendor brings to market tomorrow. This puts a very big log across your upgrade path. You'd have to buy a new drive to use the new media. And then you'd have to cross your fingers and hope the new drive will read your old discs. If it won't, you're stuck with buying a new drive AND all new media, PLUS converting the data. More fun stuff.
The 5 1/4" optical marketplace is much farther ahead in the standards arena. Those vendors are working on a universal data format that will let any disc be read in any drive, regardless of vendor or platform. And nearly every manufacturer has made a commitment to make their new drives read their old media. (Don't count on reading the new media with your old drive.)
2. WORM emphasis. Most large-format optical jukeboxes give write once, read many storage. You can't erase old information and replace it with new.
Drives used with most 5 1/4" jukeboxes are multifunction models. Use the same drive and media for either WORM or rewritable storage. More versatile.
3. Cost. You'll probably pay a lot more for the same amount of storage with a 12" jukebox as compared with 5 1/4". Take two terabyte solutions from Sony. Their 1.1 TB WDA-E550 jukebox, with two 12" drives and 76 12" platters, lists for $250,000. But for a 1.34 TB system with two Sony OSL25000 5 1/4" jukeboxes, four 5 1/4" drives and 516 5 1/4" discs, you'd pay only $150,000.
Each additional 12" platter from Sony lists for $460. The 5 1/4" discs are $90 each. The cost per megabyte is roughly the same for the two formats: 3.4 cents per MB for 5 1/4", 3.1 cents for 12".
Six Ways to Boost Performance
Are you buying a large-format jukebox to make an immense database available to multiple users? Then you'll want fast data retrieval. Here are six ways to speed things up in that department:
1. Use lots of cache. Caching is a data access technique that copies frequently-used files into a high-speed memory buffer, or cache. The cache, not the slower optical disc, satisfies requests for those files.
Here's a typical scenario where caching is crucial. A bank has a customer service department that handles account inquiries. Most relate to the current month's statements. Each day, those statements move from the appropriate optical disc(s) to the cache buffer after the first statement file is requested. (The statements could also be read into cache the previous night.) All subsequent requests for account data are retrieved from the cache.
Caching also lets you use predictive file loading. Let's say a secretary in the same bank requests the first page of a 20-page document to see if it's the file she wants. The jukebox anticipates that the secretary may also want to glance at the other pages as well. So it loads the entire document into cache. This eliminates the need to reaccess the optical disc if the other pages are requested.
2.Use more than one drive. Let's say your jukebox is in a high-production environment where a large number of users are continuously accessing data or images from optical storage. You'll get better retrieval speeds if you use more than one drive. Most jukeboxes take at least two drives. Some let you use up to six.
There are two ways to connect multiple drives. The first is to simply daisy-chain additional drives to the existing one. The drives won't be able to read and transfer data at the same time with this approach. But you'll be able to prefetch discs and load them into idle drives while the busy drive is reading and transferring data.
Result? Your users don't have to wait while the robot swaps discs. Plus, you put more data under head.
Here's how it works. Let's say your juke has three drives. While drive 1 is in use, the robotic arm fetches the next discs to be read and loads them in drives 2 and 3. When drive 1 is done, drive 2 kicks in. When drive 2 is done, drive 3 goes to work. While any given drive is in use, the other two are being loaded with discs and then waiting their turn to read and transmit data.
The second way to add drives is to connect each additional drive to its own host adapter. This approach lets all the drives read and transmit data simultaneously. Very fast. But also very expensive. And not always possible with every jukebox and host.
There's an important trade-off when you add drives to your jukebox: You typically have to give up several storage slots to make room for each additional drive. In other words, you gain speed but lose capacity.
3.Reshuffle your discs. In a jukebox, disc slots are stacked vertically. Discs typically get written sequentially, starting with the disc nearest the drive. When that disc is full, the drive moves on to the next optical disc.
This means that older data is closest to the drive and the most recent data is farthest away. That's crazy. Newer data is typically in greater demand, and thus retrieved more frequently, than older data. So you might want to reshuffle your discs periodically and put your most recent data closest to the drive.
Two caveats:
1. You can't use this strategy with all jukes, since some don't let you move discs.
2. Moving the discs might confuse the jukebox's volume manager. This module's job is to keep track of the position of discs in the juke. When discs are repositioned, the volume manager must reorient itself. In the meantime, the jukebox won't read or write data.
Suggestion: Reshuffle your discs after hours. That way, your users won't have to wait for the volume manager to catch up. (Some volume managers are more adept than others at reestablishing the disc arrangement. If you plan to reshuffle your discs often during the day, look for a volume manager that's quick on its feet. So to speak.)
4.Center the drive. Some juke-boxes let you position the drive(s) anywhere you want. For optimum retrieval speed, the most strategic place to put drives is in the center of the box. Why? Because it reduces the distance to the farthest slot by about half. This increases retrieval speeds. And it means you don't have to shuffle discs manually very often.
5.Use a separate drive for writing data. Unfortunately, WORM drives can't read and write data at the same time -- at least, not yet. So when a drive is writing data, users have to wait to retrieve files. And vice versa.
Solution: Dedicate a drive to only writing data. You can write to a compatible standalone drive and manually insert the discs into the jukebox. Or you can add a drive to the juke and designate it as a write-only device. The drawback with the second approach is that you give up disc slots to accommodate the additional drive.
You say another 12" drive will break your budget? Then you might want to batch your write requests and run them after hours with the drive you're already using. 12" drives work well for this application, since you can leave recordings unattended for long periods of time.
6. Resequence requests for data. Configure your app so it can prioritize simultaneous retrieval requests. Translation: Have the software decide the best order in which to respond to multiple data requests.
Let's say 25 users all ask for data at about the same time. The jukebox initially queues the requests in the same order they were received. But the application determines that users 3, 7, 14 and 23 all want data from the disc that's currently in the drive. So those requests are sorted to the top of the queue. The data requested by those four users is then read before the next disc is loaded. The result? Fewer disc swaps.
You may need additional system resources to resequence your requests, since that requires robust multitasking. But if retrieval speed is at the top of your priority list, the added expense is probably worth it.
Philips LMS Changing Rapidly
Jukeboxes are famous for providing fast access to vast amounts of information. But what if your application doesn't require a terabyte? Or even several hundred gigabytes? What if, say, 75 to 150 gigs will do it for you?
Storage can be a quandary with those types of applications. You might waste capacity -- and money -- with most standard 12" jukeboxes. But you'll also need more capacity than a few standalone 12" drives can provide. One solution? The RapidChanger from Philips LMS (Colorado Springs, CO 719-593-7900).
The RapidChanger mini-jukebox comes in two versions. The LF6600, a six-disc, single-drive model, stores 72 GB. The LF6602, a 12-disc, twin-drive model, holds 144 GB.
Neither model uses traditional jukebox-style robotics for disc swapping. Instead, the case containing the discs moves back and forth over the drive until the right disc is positioned for insertion into the drive.
The advantage? Discs don't travel as far to get to the drive head. This reduces disc swap time significantly. In fact, the RapidChanger's swap time -- 2.5 seconds -- is roughly half that of most standard 12" jukeboxes. (The trade-off, of course, is that jukes can store much more data than the RapidChangers can.)
Performance is also enhanced by the dual-head Philips drive used by both RapidChanger models. Data can be read from both sides of the mounted platter without flipping it. This means better swap time. It also means faster access, since the dual head doubles the amount of data you can access at a given time.
Think of the RapidChanger as an entry-level jukebox that's well-suited to departmental apps and small PC networks. If you outgrow it, just transfer the discs over to a larger box. Continue using the RapidChanger as a write device or as an archive for discs written with the jukebox.
FileNet Sires OSAR Family
If you need between 480 GB and 1.4 TB, then the 12" OSAR products from FileNet (Costa Mesa, CA 714-966-3400) are worth a look.
The OSAR line includes four models. The first two are sub-terabyte models designed for mid-sized applications. The OSAR-40GTL holds up to 480 GB (in other words, 9.5 million images) and takes up to three 12" drives. The OSAR-50GTL, a 600 GB (12 million images) model, includes two drives. The average cartridge exchange times are 5.0 seconds for the 40GTL and 5.1 seconds for the FileNet OSAR-50GTL.
Farther up the capacity ladder are the 1.28 TB OSAR 107GTL and the 1.48 TB OSAR 123GTL. The 107GTL takes as many as six drives, the 123GTL up to four. Average cartridge exchange time for both devices is 4.7 seconds.
FileNet jukeboxes use Philips dual-head drives.
Average access time is 100 ms. Sustained data transfer rate in read mode is 2.7 Mbyte/s.
FileNet's boxes also use a dual-picker mechanism. The robot fetches a new disc with its first picker and moves it to the front of the drive. When the drive ejects the old disc, the robot grabs it with the second picker. The robot then loads the new disc and returns the old disc to its slot. The advantage? The robot doesn't have to make separate trips to put the old disc away and bring the new disc to the drive.
Good Old Brand New Cygnet
The newest name in the 12" optical jukebox arena has actually been in the marketplace since 1986. Cygnet Storage Solutions (San Jose, CA 408-954-1800) was formed this past October after DynaTek Automation Systems (Bedford, CN 902-832-3000) acquired the assets of ATG Cygnet, which filed for bankruptcy this year.
Cygnet sells the former ATG Cygnet's series 1800 optical jukebox along with CD authoring and mastering systems developed by DynaTek. Expect Cygnet to stick to their longstanding emphasis on the Federal and government markets, where it has won a major market share.
Cygnet's Series 1800 12" jukebox is the most flexible product on the market. It runs all current 12" drives, including the 8-gig rewritable drive from Nikon. You can configure this jukebox to store 176 GB-1.692 TB.
The entry-level models in the series are the 264-GB ASM and the 348-GB 1602. Need more capacity? Consider the 1802, which handles up to 732 GB. Or the 1803, which tops out at 1.692 TB.
Each model is compatible with the next. Start small and expand as your needs grow. Add up to five drives. Store as many as 141 discs. Like FileNet, Cygnet uses a dual-head picker. The arm speeds up to reach discs farther from the drive. This minimizes the difference in retrieval times for near and far discs.
Sony Stuffs the Discs
Among today's jukeboxes, Sony's product -- the WDA-E550 -- offers the highest per platter capacity at 15 GB.
Configure Sony's box lots of ways. The base model includes one drive and 76 discs for a total of 1.14 TB. Other drive/disc combos include two drives with 999 GB, three drives with 840 GB and four drives with 690 GB.
Boost capacity by adding Sony's WDA-EX5 expansion unit. This lets you go all the way up to 2.31 TB with two drives. With six drives, the jukebox and expansion unit will store 1.71 TB.
The WDA-E550 uses Sony's new 15-gig jukebox drive, the WDD-530. (If it squeaks, use WD-40. That's a joke, son.) The drive's dual-head design lets you simultaneously read or write on both sides of the disc.
This lets you achieve sustained data transfer rates of up to 2.7 MByte/s.
Seek times range from 150 ms to 600 ms. On-board memory includes a four meg data buffer and a 256 KB read-ahead cache.
The WDD-530 jukebox is designed to be used with Sony's WDM-1500 optical disc. (According to Sony, WDM-1500 discs will last 100 years. But who ya gonna sue if they don't?)
With extra hardware, the drive can also read the previous generation of Sony 12" optical media.
Kodak Optimizes Optistar
If capacity upgrade potential is high on your priority list, then you should consider the Optistar Storage System Series 2000 jukebox from Kodak (Rochester, NY 716-724-4000).
The Optistar uses Kodak's 14" WORM drive, which holds 14.8 GB of user data. The company plans to increase the per-platter capacity to 25 GB by next June and to 200 GB by the year 2000.
Choose from several Optistar configurations. The base model includes one drive with 50 slots.
Add another drive and a 50-slot expansion bay, or skip the extra drive and pick up more capacity with an 80-slot expansion bay.
Cache ranges from 1-128 MB. Average disc exchange time is 6.5 seconds. Access times for the 14" drive range from 9 ms to 700 ms, depending on the location of the data on the disc. The sustained data transfer rate is
1.0 MByte/s.
In addition to the 14.8 GB media, the Optistar system supports Kodak's earlier 10.2 GB and 6.8 GB optical discs as well.
The drive has only one head, so you have to flip the disc to read the other side. Kodak is currently working on a dual-headed upgrade, scheduled for release next June. That drive boosts capacity to 25 GB.
The company is also working on a half-height version. This will let you insert four drives into the Optistar jukebox without reducing the number of slots in the two-drive model.
Related Articles:
