Hal is a hardware engineer who sometimes programs. He is the former editor of DTACK Grounded and can be contacted through the DDJ offices.
I've just finished reading the article "Disruptive Technologies: Catching the Wave," by Joseph L. Bower and Clayton M. Christensen (Harvard Business Review, January/February 1995), which discusses the history of companies making hard-disk drives and the choices and mistakes these companies made. It's a fascinating article.
For instance, Bower and Christensen discuss the reasons new companies arise to eventually challenge (and sometimes displace) entrenched companies. An example is Seagate. Founded in 1980, the company had by 1986 grown to be a $700 million company, providing 5.25-inch hard drives for AT-compatible PCs. Seagate developed 80(!) models of 3.5-inch drives, but its principal customers wanted to stick with higher-capacity 5.25-inch drives. So Seagate continued to concentrate on the larger format, allowing upstarts Conner and Quantum to create a new market for 3.5-inch drives that could fit into small desktop cases and Compaq's portable PCs.
As the authors point out, the capacity provided by 3.5-inch drives was increasing faster than that demanded by PC users. Suddenly, a 3.5-inch drive was adequate for the needs of even power users with large desktop cases, and Seagate's principal customers left to find a volume source of the smaller, less expensive drives.
Seagate survived and has even prospered of late, with 1994 sales of $3.5 billion. But Conner and Quantum had combined 1994 sales of $4.5 billion, and a lot of that $4.5 billion came out of Seagate's hide.
The article introduces "the concept of performance trajectories--the rate at which the performance of a product has improved, and is expected to improve, over time...." (I've always called that a "performance trend.") The article also differentiates between "sustaining" and "disruptive" technologies. Transistors disrupted the vacuum-tube industry, for instance.
Bower and Christensen believe that the switch over from 5.25- to 3.5-inch drives was a disruptive technological change. Well, I don't agree with their every assertion; they seem to believe the 1.8-inch drive market is going to grow dramatically, and that storage capacity is still growing only 50 percent per year.
Their article doesn't include a whole lot of hard data. (It is important to differentiate between the data the authors had available when preparing the article, and what they chose to include in an article aimed at a business audience.)
I live in Santa Clara, the heart of Silicon Valley, just a long stone's throw from Intel headquarters. It won't surprise you that the local rag, the San Jose Mercury News, has a "Computing" section in its Sunday edition. It may surprise you, however, that this section is devoted to personal computing, written for the consumer. No corporate puff pieces that I can recall. This must be frustrating to certain local industry leaders who like to see themselves in print.
Since the section is read mostly by personal-computer consumers, it attracts a lot of ads from retail outlets, many of which are low-overhead outfits selling no-name PC clones at very attractive prices. Others are superstores, the best known of which is Fry's Electronics, which used to be a supermarket (as in food).
On a whim, in February 1993 I started saving the "Computing" sections. After a year I thought of a use for all that stacked-up newsprint: a database that would allow me to track the prices of various PC products over time.
The local superstores run full-page ads which feature two- or three-day sales on just a few items. Because of the limited selection, these ads are not a reliable way to track prices. Many of the low-overhead outlets run small ads that list lots more items than the full-page ads of the superstores. One of these, Hi-Tech USA (Milpitas, CA) has advertised consistently since I started saving the "Computing" section, missing only one Sunday in over two years. The prices in these ads (by the no-name clone outlets) are consistently low, if not always as low as particular items in the superstores' full-page ads. I decided to plot hard-disk prices as advertised by Hi-Tech.
I wrote a QuickBasic program that's a collection of disk-price data statements, one line per week. Every Sunday I take a minute to update this program. I wrote other QB programs to plot that data, using PCL5 to drive my LaserJet 3P.
When I plotted all those disk prices, I wound up with an interesting but confusing graph that had too much raw data. What I really wanted to know was the price of hard-disk storage, measured in dollars/MB versus time and the disk size that provided the lowest price/MB at a given time. I wanted this information for my own buying strategy, and I figured that anybody else who bought hard drives from time to time might be interested, too.
Everybody knows the old recommendation for choosing a hard drive: Buy the largest one you can afford. I don't know if that ever was a good buying strategy, but it certainly isn't now. Why?
Figure 1 is a plot of five time slices of disk price/MB versus disk capacity. This plot shows a few things: First, at a given time there's an optimum disk size, or "sweet spot," that provides the lowest cost/MB of storage. Buying a smaller or larger drive increases the cost per megabyte. Second, the optimum disk capacity increases with time. Third, the downward price trend is less consistent for smaller drives.
A hard-disk drive contains h platters, where h=1,2,3,.... IDE drives for PC clones invariably have either one or two platters. Since disk manufacturers can't use, say, 1.5 platters to produce a variety of disk capacities, they use higher or lower disk read/write technology. The "sweet spot" is for drives with two platters and the most cost-effective R/W technology. This technology moves with time.
Why two platters? It doubles the capacity to move from one platter to two, but it obviously doesn't double the cost. Why not three or more platters? I dunno. Maybe the physical size of the drive increases. Maybe the absolute price is too high (as opposed to dollars/MB). In any event, the cost penalty for buying away from the sweet spot is lower if you move up in size than if you move down.
The lowest-cost/MB, one-platter drives are known by the marketeers as "entry-level" drives. As of May 1, 1995, the entry level was 540 MB.
We want to closely track that sweet spot, both in terms of cost/MB and optimum disk size; see Figure 2. A glance shows that storage per buck is doubling every year (at least). This is one reason I think the old paradigm of buying the largest disk you can afford is dead.
I don't like coincidences in my data because it makes the data appear to be cooked. Regrettably, there are two coincidences in this article. The first is that I happened to start saving the Sunday "Computing" sections at the time that downward disk-price trends suddenly doubled from their historical (1986 to early 1993) rate. It used to take two years for storage per buck to double.
Could the sudden increase in the areal-density trend that occurred early in 1993 have been anticipated? Well, it was anticipated! Robert Scranton, director of storage systems and technology at IBM's San Jose Almaden Research Center is quoted in the article "Disks Make Capacity Drive," by Terry Costlow (Electronic Engineering Times, March 8, 1993), saying, "We think areal density will increase by about 60 percent over the next few years. That means disk-drive capacity will double every 14 to 16 months, instead of every two years." So he was a tad conservative; capacity per dollar has in fact been doubling every 12 months since he made that statement at the 1993 Technology Forum.
Let's take a close look at Figure 2. The first thing to notice is the price increase toward the end of 1993. Since there was no discontinuity (technology shift) in the optimum size, this would appear to be a supply/demand problem. In fact, a lot of PCs were sold in the fourth quarter of '93.
Next, you see the sudden jump in optimum disk size at the start of this year from 540 to 1260 MB! This suggests a couple of things: First, the IDE 540-meg addressing limitation threw a wet blanket over sales of IDE drives larger than 540 megs because of the slow consumer shift to EIDE drives. Second, Hi-Tech happened to start selling 1.2-GB drives four weeks before offering 850-MB drives. As soon as 850-MB drives were offered, they immediately became the new optimum size. But 1.2-GB (actually about 1.26 GB) drive prices were dropping steeply, and by March 26, 1.26 GB was once more the optimum size. Both 850-MB and 1.2-GB drives then cost about $.27/MB, giving the consumer two price points: $236 at 850 MB or $368 at 1.26 GB.
I did not overlook those large ads that feature only a few items; Access in Santa Clara, for instance, advertised a Maxtor 853-MB drive for $209 ($.24/MB) on April 30. Such occasional specials, however, are not included in my data.
Besides the discontinuity when the market moved from IDE to EIDE, some disk sizes just never became cost effective. An example is the 1.08-GB IDE drive. I dunno why, but your money is better spent on 850-MB or 1.26-GB drives. Figure 1 also shows this.
If your personal strategy is to bite the bullet and buy as big a drive as is reasonable, then your best buy is a disk whose size has just become the new sweet spot; for example, $338 for 1.26 GB. If you're a cheapskate like me, you'll want the best cheap choice ($209, 850 MB). These are April 30 prices; by the time you read this, the market will have moved on. And that's another point: Delay buying a new hard disk until the last possible moment, because prices continue to nosedive. Surely you can offload some inactive files?
(You do back up your hard disk, don't you? I have a colleague who has three 1-GB drives at home, and he does not back them up. Yikes!)
Picture a male high-school student with a brand-new, completely empty, 1.2-GB drive, a color scanner, and a copy of Playboy magazine. How long do you think it'll take him to completely fill that drive?
We PC users are divided into those who already work with images and those who will be working with images. As soon as you start dealing with images, no hard disk in the world is anywhere near as big as you need. I don't even want to think about video, but Quantum has just introduced a 9-GB, 3.5-inch drive for about $2500. Seagate and Micropolis have been shipping 9-GB drives for a while now using 5.25-inch technology.
850-MB drives come with 32-256 KB of on-disk buffer memory, have average access times of 9-15 msec, and average latencies of 8.3-5.56 msec (3600-5400 RPM). The fastest of these drives commands about a 10 percent price premium over the slowest.
Figure 3 includes the data from Figure 2 and the plots from "Disk Architectures for High Performance Computing," by Randy Katz, Garth A. Gibson, and David A. Patterson (Proceedings of the IEEE, December 1989). The 1982 to 1988 plots are (presumably) based on hard data. Katz predicted trends for another four years, up to 1992. Here is the second coincidence: When I extended Katz's 3.5-inch trend to February 1993, it met my Hi-Tech data perfectly. I didn't cook this data, honest! I predict that the new, steeper trend will extend at least until 1998, when, at $.04 a megabyte, a 9-GB drive will set you back $360. Let's hope we have a E2IDE standard by then, because 9 GB exceeds the 8.4-GB EIDE address limit!
Figure 1: Sweet spot versus time. Figure 2: IDE disk price/MB and optimum size versus time. Figure 3: Disk price/MB versus time.
Copyright © 1995, Dr. Dobb's Journal