Featured in CAD SYSTEMS November '98


Storage Devices: Where and What Kind?
By Donald Jenner

  Mass storage is a complex issue; the question is, "Mass storage where and what 
  kind?" It can be at the local workstation, and either kept private, or shared 
  out across a peer-to-peer local network. It can be part of a dedicated server 
  system. It can be removable or fixed. 

  Some obvious items to consider: 

  Zip drives, from Iomega, have held pride of place in the small-removable 
  category for a long time. Imation (a 3M spinoff) is out to give Iomega's 
  superb marketing team a technology challenge with its superdisk. 120mb 
  superdisks cost less (especially in 10-packs), are smaller and seem a bit more 
  rugged. Superdisk drives ("superdrives") can manage both 120mb superdisks and 
  standard 1.44 3.5 inch floppies. Best of all, Imation is not the only source 
  for either drives or disks; e. g., Mitsubishi is making a range of superdrives 
  in different sizes and with different interfaces for different applications. 
  Some computer manufacturers are including the dual-capability superdrives in 
  lieu of 3.5 inch floppy drives, as a way to differentiate their products. 
  I use my superdrive with my CTX notebook computer when I am in the office. 

  This is simpler than either pulling the battery and putting in the floppy 
  drive, or connecting the floppy externally, and gives me broad access across 
  the LAN. The superdrive I have connects through the parallel port. Like 
  similar connect-through-the-parallel-port storage devices, the connection 
  looks like a SCSI connection to Win95/98; if you like the superdrive concept, 
  but use SCSI at the station where you are connecting this, get a SCSI 
  superdrive, as the parallel port version did not work on SCSI-equipped 
  systems, in our tests. The version we have gives limited portability (it is 
  too big to be carried all the time), and its metal case (thankfully, not 
  grape-colored...) makes for seemingly greater reliability when I do take it into 
  the field. 

  What about removable hard disk systems? My favorite of a couple years back, 
  the Avatar Shark 250, never seemed to go anywhere. Now the company itself 
  seems to have disappeared, presumably a victim of Asian financial woes (its 
  parent was Bangkok-based). The remaining contenders are Iomega's one- and 
  two-gigabyte Jazz drives and Syquest's one-gig SparQ and one-point-five gig 
  SyJet. Both product lines offer lots of storage at comparable prices (they 
  pace each other; as I write, Syquest appears more aggressively priced than 
  Iomega). 

  Truly permanent storage is still the preserve of CD-R. A read-write variation 
  is also around, and the prices are about the same - about US$350-US$450 for 
  the drive. CD-R media run as little as US$2.50 in single-disk quantities. I 
  have stuck pretty much to CD-R; for me this is an archival medium (the minimum 
  rated lifetime is 25 years if you don't scratch the disk) and rewriting is not 
  useful. I have used drives from both Hewlett-Packard and Olympus (the former 
  with a tray, the latter using a caddy), connecting to the Intergraph TD225 
  through an Adaptec SCSI adapter. Both have performed flawlessly. Although I 
  have seen parallel-port versions of the Hewlett-Packard SureStore CD-R drive, 
  I am skeptical that it could sustain transfer speeds needed for reliably 
  writing data. 

  All of these storage devices are adjuncts to main system storage, merely. Main 
  storage, whether at the desktop or on the network, is generally permanent, and 
  it is huge, and it is cheap. E. g., quantity-one retail prices for 11.5gb hard 
  disks are around US$350. For most people, that is more hard disk capacity than 
  their computers can address as a contiguous whole. 

  There are some problems with these drives, in my view. First, inevitably these 
  big, cheap hard disks are EIDE, not SCSI devices. But SCSI has proven a 
  generally better technology for moving data around (it was, after all, a 
  development of the same Shugarts who made hard disks popular). SCSI is, put 
  simply, faster than EIDE in most cases, and appears to be more able in a 
  multitasking environment. Still, if your system uses EIDE, big cheap hard 
  disks are a great solution for local storage. 

  Second, integrating these disks is not without an element of adventure. This 
  is particularly true, should the idea be to upgrade a boot drive (not just to 
  add a secondary drive on one or another IDE channel). Again, SCSI usually 
  makes this easier (and there are some developments coming from Adaptec and 
  Seagate, according to those companies, which should enhance that ease of use). 
  Say perhaps, a good backup is not an option when integrating a major upgrade 
  of this kind. 

  Finally, even in a nice peer-to-peer network, storing stuff all over is just 
  sloppy. Big disks are good things in graphics workstations, especially if they 
  host 3D or animation (like, walk-throughs) or really large application suites. 
  But there needs to be a "data-central" for shared information, if nothing 
  else. 

  There are options: Since data-serving is generally not processor-intensive, 
  that same machine can be a print server, or even host a proxy server for the 
  Internet. So, buy a new box, or upgrade an old box, with a big hard disk; this 
  doesn't have to be a heavy-hitting system, so last year's Pentium is plenty 
  good enough (a 486-based system might do as well). 

  Another option, from Meridian Data, Inc., is a hard disk in a black box, the 
  company's Snap!Server. Available with capacity from four to twelve gigabytes, 
  according to the company's literature, these cost between US$1,000. and 
  US$1,800. - on the pricey side, given the capacity. But these are 
  plug-and-play devices. Give 'em some juice and plug in a connection to the 
  Ethernet hub; the system more or less configures itself, apparently. 
  Administration chores are done via web-browser interface. This is strictly a 
  file-server device; there is no capacity for other kinds of communication in 
  the unit, which is small (say, half the size of a really small desktop 
  computer). The advantage is ease and speed of deployment; where labor costs 
  are a greater concern than the capital cost of the device, Meridian's 

  Snap!Server merits consideration. 
  The "grown up" version of what Meridian offers comes from companies such as 
  Boxhill. Generally, one goes to a company like Boxhill to secure a 
  custom-built disk array, matching the special characteristics of the - usually 
  large - network of stations. Boxhill will configure its various solutions with 
  anywhere from four gigabytes to two-plus terabytes of accessible hard disk 
  capacity, with a full range of advanced RAID features to guarantee maximum 
  reliability. 

  Lots of companies offer RAIDs. The local storefront guy can buy an 
  off-the-shelf RAID controller package, stick it in a box with a simple 
  motherboard, four swap-able drives of suitable size and maybe a tape drive. 
  These will be more or less reliable, depending on the savvy the system builder 
  has in this special, rather arcane area. That is not what Boxhill is selling. 
  Boxhill's only business is building mass-storage and backup systems. The 
  company has several general categories of connectivity - via fiber, SCSI and 
  so on. The company has a variety of storage devices they can hook in through 
  these connections. Being canny systems engineers, the Boxhill team takes 
  advantage of hardware features early and effectively; for example, the company 
  claims to have been the first to implement RAID using onboard capabilities in 
  Seagate Barracuda drives, reducing the overall cost of advanced RAID systems. 

  The result is a custom-fit storage solution; Boxhill quotes only a 
  per-megabyte figure (around US40¢, tops), which is not particularly helpful in 
  judging potential costs. On the other hand, the kinds of subsystems Boxhill 
  builds are for systems where the normal practice would include soliciting a 
  fairly detailed proposal. 

  As a rule of thumb, I suggest the following for graphics-oriented shops, with 
  between six and twelve seats: Locate a lot of storage at each station, to 
  accommodate complex design needs. Locate about three times that amount of 
  storage at the network server, with archive and backup services alongside. 
  This has proven practical and effective (not always efficient). Larger 
  installations become sufficiently variable that a rule-of-thumb approach is 
  perhaps less than wise.