Latest Developer Headlines · Open Source Targets Microsoft Exchange [BusinessWeek] · McDonald's Offers WiFi [Seattle Post Intelligencer ] · Hackers Contest Makes a Mess of Internet [Sify] · Hacking Contest Threatens Web Sites [Information Week] · Microsoft Launches Digital ID Software [CNet] submit | more ...

What's New on ACM Queue · A Conversation with Jim Gray · How Much Storage is Enough? · You Don't Know Jack about Disks · Programming Without a Net · Web Services: Promises and Compromises

The Old Days: They weren't that Good

This original, geometric programming model that coincided with the physical organization of data posed several challenges for the programmer. It helps to understand some of the limitations of this physical model.

The specifics of the geometry seeped into the operating system (OS) software. For example, two drive manufacturers might develop comparable drives of identical capacity, but one might get more of the capacity via TPI, the other from BPI, and a third vendor with less capability might require an additional disk in the drive. This would result in drives with very different values for cylinders, tracks, and sectors per track. What's worse—or better, depending on your point of view—some disk controllers were able to get one or two more sectors per track by using more sophisticated electronics, by, for instance, shortening the gaps between sectors. The upshot was that even a given disk drive could have different geometry values depending on the controller to which it was attached.

The complexity of supporting all drive offerings was considerable. As explained previously in the areal density discussion, each generation of disk drive would increase both the sectors per track (BPI effect) and the number of cylinders (TPI effect). This means the OS or controller needed to include a table of drive geometries, the size of which was multiplied by the number of generations it contained.

Handling flaws was an even bigger problem. When areal density was low enough (around 2 megabits per square inch, which was state-of-the-art in the mid-1970s), it was possible to build a stack of disks with no flaws, or bad sectors. As areal density increased, this quickly became impossible. Because the OS was looking at the raw device, it was responsible for managing bad-block flaws so they did not jeopardize the user data.

Through many generations of drives, controllers, and operating systems, the developers put up with these problems. But then came the last straw: zoned bit recording (ZBR).

An artifact of the cylinder, head, sector (CHS) model was the assumption that every track had the same number of sectors. As drive designers looked for innovative ways to deliver more capacity, they saw that the longer tracks on the outer diameter of a disk could hold considerably more data than an inner track. To take advantage of this, the drive was divided into radial recording bands or zones (i.e., clusters of nearby cylinders). All the tracks within a given zone had the same number of sectors. A track in a zone near the outer diameter of the disk, however, might have 50 percent more sectors than a track in a zone near the inner diameter of the same disk. This would be true for a 3.5-inch drive. The advantage ZBR provides varies by media size and is a function of the relative size of the outer radius of the recording band to the inner. Drives today usually have 15 to 25 zones.

ZBR added great value: 25 percent or more capacity for no additional material cost in a 5.25-inch drive, the prevailing form factor when ZBR first appeared. It forced the industry to adopt a more intelligent interface—one that would hide the complexities of ZBR and, at the same time, hide the geometry and bad-block flaw problems by pulling that functionality into the drive, as well.

WHAT CHANGED

The advent of intelligent interfaces—first SCSI, then ATA—completely changed the nature of disk support. Disk drives used microprocessors to manage these higher-level interfaces, and programming the host went from a low-level, know-all-the-signal-lines-and-timing engineering task to an inter-computer communication problem that any experienced programmer could manage.

These intelligent interfaces are based on a programming model that is essentially like a tape—sequential sectors or block addresses. Although drives still have block random access capability, the physical geometry is no longer seen in the programming model. Behind this new model or interface, drive vendors have been able to produce higher-capacity, more reliable, and faster-performing storage.

Comment on this Article in the ACM Queue Forums
Previous Page (2/7) The Basics: TPI and BPI	Next Page (4/7) Inside a Drive Today