The Hard Drive Click Of Death: Archaic Technologies And Their Protracted Viabilities

It hasn’t been a good year for computers in the Dipert household. At the end of May, the Windows partition on my dual-O/S MacBook self-destructed. A month ago, the ‘perfect storm’ of a dog, a cat, and a glass of Merlot proved fatal for my geriatric-but-dependable Dell laptop. And late this past Saturday night, when the MacBook locked up and its hard drive began emitting the ‘click of death’, I knew I’d be spending the next day or few reinstalling from scratch its OS 10.4 foundation, Windows XP virtualized O/S, and full application suites for both operating systems.

That’s right…I didn’t have a recent full-image backup, although I should have known better, and freezing the dead HDD hasn’t temporarily revived it, either. Yes, folks, this is one of those cases I mentioned before, where S.M.A.R.T. statistics are of no help in alerting me to impending failure and in fact, arguably, gave me a false sense of security. Fortunately, I’d archived my voluminous Microsoft Outlook database of emails, RSS feeds, calendar items, contacts and to-dos just 24 hours earlier. Also, the bulk of my data files (specifically, music and photos) are on network drives, so they weren’t lost. Similarly, the combo of FEBE (which I’d used to archive configuration files to the network) and Foxmarks (which backs up and synchronizes bookmarks with the ‘cloud’) got me up and running again with a fully configured Firefox in short order. As this writeup (written on the MacBook in VMware Fusion-powered Microsoft Word and posted via Fusion-powered Internet Explorer) suggests, I’m well along in the path to recovery. Thank goodness for football on television yesterday afternoon and evening to help me pass the time…

Although it’s a pain to re-install everything from scratch, this system was frankly overdue for a substantial housecleaning, so the hours I’ve put into the project aren’t a complete waste. I’ve long been skeptical about the robustness of the 160 GByte Seagate 2.5" (5400.3) HDD that expired Saturday night. Over a year ago, I experienced disruption of OS X’s HFS+ partition, and I’ve had plenty of problems with Windows XP NTFS partition data corruption as well (which, perhaps naively, I’d until now blamed on Apple’s Boot Camp software). My system had recently inherited a baffling bug wherein whenever I mated a new USB device, virtualized Windows XP would lock up at the ‘Found New Hardware’ step with services.exe consuming all available CPU resources. And although I hadn’t noticed anything particularly wonky going on with OS X, I’ve installed, upgraded and un-installed plenty of software and peripheral drivers over the past nearly two years, so a clean sweep is probably justified here as well.

Near term, I’ve upgraded to a 250 GByte Western Digital 5400 RPM HDD, whose added capacity enabled me to (among other things) double the virtualized Windows XP drive’s size from 20 GBytes to 40 GBytes. In the near future, however, I’ll probably migrate to a flash memory-based solid-state drive, for reasons that reflect some big-picture technology-trend thinking I’ve been doing in recent weeks as I work towards completion of my next hands-on project (a HDD-vs-SSD benchmarking writeup for early-November publication in EDN). I don’t think many of you will argue with me that the overall trend in computing is towards portable, therefore ruggedness- and power consumption-critical hardware, in which form factor and weight are also critical parameters. Why, then, are we still using rotating magnetic mass storage that’s powered by a motor spinning at 4200, 5400, 7200 or 10000 RPM, with a rapidly vacillating read/write head suspended only a few millionths of an inch above its platter mate?

Cost, specifically cost/GByte, is of course the key reason for HDDs’ near-term continued dominance. Even though MLC (presumably) NAND flash memory-based 32 GByte SSDs are already down to $80 (after rebate), they’re still nowhere near the cost-for-capacity metric of the more fiscally attractive HDD options. Yet, as I’ve pointed out before, the capacity threshold at which the fixed cost of a HDD (regardless of its capacity) dips under the Moore’s Law-fueled and density-proportional cost of a SSD alternative is steadily increasing. And, as I’ve argued before, with the passage of time fewer and fewer folks will require the increasingly burgeoning capacities delivered by HDD suppliers, particularly as alternative ‘cloud’-based storage becomes increasingly popular.

A related analogy will, I think, clarify my point. Back in the mid 1980s when flash memory first burst on the scene as a competitor to mask ROM, OTP PROM and UV EPROM, it was significantly smaller in per-device capacity compared to the already-mature alternatives as well as being significantly more expensive on a per-bit basis. As such, the only applications that used it were ones in which its unique combination of non-volatility and in-system upgradeability were valued versus the semiconductor competitors. Historical note; flash memory was cheaper than EEPROM ‘out of the chute’ and therefore quickly stole many ’sockets’. EEPROM retains an application niche where full bit alterability (both program and erase) is necessary.

As time progressed and the flash memory market began to take off, the alternative semiconductor technology suppliers maintained their per-chip density lead (at least at first) but discovered that the high-end market was evaporating…among other factors, as the amount of per-chip code grows, the probability of a bug in that code also grows, thereby making engineers reluctant to put that code in a device that’s not easily updateable. Fast-forward to today, and nobody makes EPROMs anymore. OTP technology is virtually non-existent except in antifuse FPGAs, and mask ROM technology is similarly rare outside of memory arrays integrated within ASICs.

It’s my contention, now that flash memory has turned its competitive attention to rotating storage, that the same slow-but-steady obsolescence will occur with HDDs. They’ve served us (pretty) well for years, and today they’re a huge business, but their shortcomings are increasingly difficult to ignore. When SSDs’ capacity at a given price point becomes sufficient for a given application, semiconductor advantages (read performance, power consumption, ruggedness, etc) will be irresistible. So-called ‘netbooks’ have already largely made the transition to flash memory-based storage. Notebook PCs and other portable computing, communications and entertainment devices will follow them, if they haven’t already done so. Was anyone surprised, for example, that the HDD-based iPod classic received nothing more than a Toshiba-provided capacity upgrade in last week’s intro? How long do you think it’ll be before the iPod touch completely takes over the high end of Apple’s portable media player lineup and the iPod classic is discontinued, as was the case with the iPod mini-to-nano transition of a few years ago?

Ironically, what’ll probably pace the HDD-to-SSD Darwinist conversion more than any other factor is flash memory suppliers’ degree of courage to bring the necessary manufacturing and test capacity online. Desktop PCs will probably be the last HDD holdouts, but eventually the overall shrinking hard drive market will erode rotating storage’s cost competitiveness to such a degree that the SSD will triumph here, too. Then you’ll see legacy HDD-related operating system features such as defragmentation (which is unnecessary with an SSD, and in fact only results in undesireable incremental block erase cycles) disappear. Storage-to-system interfaces derived from HDDs’ inherently serial access scheme will also vanish, replaced by more semiconductor-compatible parallel alternatives. And I’ll even go out on a limb and predict that the average per-system amount of DRAM will even shrink…when SSD-based virtual memory paging latencies are a fraction of what they were in the HDD days, they’ll be much more tolerable in exchange for an overall reduced system memory bill-of-materials budget.

Visionary or Don Quixote; which one am I? Let me know in the comments.