COEN 180
Data Storage on Rigid Disks

Figure 1: Hitachi Ultrastar Hard Drive

A hard drive consists of a single disk or stack of disks, which have a thin magnetic coating and rotate at high speeds. The disk surfaces are recorded using magnetic recording heads mounted on arms that are moved accross the disk surface by a high speed actuator. The information is written on circular tracks. There is an actuator for every disk surface and they are all moved in parallel by the actuator. The tracks on all the surfaces formed by the same actuator position form a cylinder. The tracks themselves are broken up into sectors (or disk blocks). The old format of 512B per block still remains in effect.

Hard Drives need to rotate at a constant speed. Alternating rotational speed with the head position takes too long to be practical. Constant rotational speed implies that the head sees more track in the outer cylinders during a fixed time period. If we use the same number of sectors per track all over the disk, then the data density would be highest in the innermost cylinders, wasting capacity in the outer cylinders. If we change the number of sectors per track continuously, then optimizing the access pattern becomes difficult. As a solution, the disk surface is broken into many bands consisting of contiguous cylinders. Within each band, the number of sectors per track is constant, but this number changes from inner to outer bands. Data density decreases within the band as we move from one track to the next one out, but remains constant among the several bands.

Figure 2: Actuator-Head-Surface Assembly (left) and Tracks Forming a Cylinder (right).

The platter consists of a rigid aluminium or glass platter, coated with various coats. The most important coat is the magnetizable thin film that actually stores the data. This coat is protected by an overcoat and on top of that a lubricant. A disk drive suffers from external forces that can bump the head into the surface ( a head crash). The lubricant mitigates the effects, which in the long run destruct the head as well as the magnetic surface. The aftermath of a head - surface smash can introduce particles into the assembly that further damage head and surface. To maintain the proper distance between head and surface magnetic disks are not kept at a vacuum, but make use of the surrounding air pressure. The spacing controls the focus of the head; if the head is further away from the surface, then it will read from and write to a wider area. To increase data densities, the head - surface spacing has decreased dramatically. As a consequence, the head can no longer be parked on the surface during power down (when the rotation ceases, the head will actually land). Rather, the head needs to be parked during power-down in an area, where the surface is formed to allow air to enter between it and the head so that the head can take off during power up. Disks will move their heads to this zone whenever they are idle, a policy essential for disk survival in laptops where the disks even have a gravity detector to take emergency measures while the laptop is hurdling to the ground.

Figure 3: Head - Surface Spacing consisting of the mechanical spacing, the lubricant and the overcoat thickness.

Data Access Mechanics

When the hard drive performs a read or write request, the actuator moves the head assembly first over the right cylinder and switches to the right head to find the data on the correct track. It used to be that one service was dedicated for "servo" information, essentially tracks layed out to find the right cylinder. However, the different temperatures in different parts of the disk assembly only allow this type of tracking if the track width is relatively large. Therefore, cylinder and sector data is embedded into the data stored on a track. This first portion of the access time is called the seek time. When the head has reached the correct track, then the access waits for the disk to rotate the desired sector under the head, the rotation time. On average, this happens after half the time for a complete rotation (given in RPM rotation per minutes). This value is known as the latency. This mechanics favors accesses to subsequent sectors on the same track or at least on the same cylinder. Zero latency disks are capable of reading from (or writing to) a track whenever the track is reached instead of waiting for the first sector of the desired data to appear. Actually, switching from one track to another in the same cylinder is not quite instantaneous but takes the head switch time. The seek time is a non-linear function of the seek distance roughly equal to a constant plus a square root type function of the seek distance. Once the first sector has been reached, the disk needs to transfer data to or from the surface. This time is the transfer time. In summary, the disk access consists of three phases:

Performance Parameters

The disk file performance characteristics of importance in the design of file systems are:
  1. Formatted Hard Drive Storage Capacity.

    Formatted Hard Drive Capacity depends on the data density and the overhead that the servo mechanisms and the breaking into tracks and sectors impose. Data density is given by track separation, track width, and the bit density per track. We will look at these factors one by one.

    Track separation is determined by the ability to control track misregistration (Figure 3). If the head is not in the center of the track, but slightly off to one side, the head might not obtain enough signal strength to reconstruct the written data. If the misregistration is worse, then the head might read magnetic flux emanating from the neighbouring track. Track misregistration during a read is a soft error, since the next attempt of a read succeeds. Track misregistration during writing leads to a more permanent error, since the written data might no longer be readable. Even worse, a more severe track misregistration can destroy already written (and possibly verified) data on an adjacent track.

    Figure 4: Track Misregistration. The black lines indicate the width of the read and the write operations. (a): No misregistration. (b) Track misregistration, that is acceptable for a read, but not for a write because it will not overwrite the previous magnetization. (c) Track misregistration that causes reads from two tracks and writes to two tracks, destroying data in both.

    Track width depends on the ability to control track registration and on the head to disk spacing. The wider the spacing, the larger the area that affects or is affected by the head operations.

    Data density is limited by physical factors, such as the superparamagnetic effect, where thermal energy causes fluctuation in the magnetization of a particle. Basically, depending somewhat on the magnetic medium, a too small magnetized area might spontaneously switch magnetization, destroying data. Data density is in general limited by the magnetic properties of the storage medium. For example, particulate matter (basically formed by suspending iron oxide particles in a carrier medium) can only achieve magnetic densities conmensurate with the size of the particles, therefore, this type of magnetic medium has been abandoned. Data density also depends on the head to surface spacing. If the flux changes are close together, then they influence each other (Figure 5 shows the effect for an inductive read head). This leads not only to a decrease in effective signal strength but can also move the position of the detected magnetization changes. By closing the spacing between head and magnetized material (near contact recording), the head measures the flux (changes) in a more focused area.

    Figure 5: Pulse Super-Imposition Theory for an inductive read head: As the two flux changes move closer to each other, not only is the amplitude diminuished, but the location of the peak is moved outwards.

  2. Seek Time:

    The seek time is given by speed with which the actuator assembly can move to the correct track. Disk manufactorers use different levels of aggressiveness in the move, a more aggressive strategy accelerates the actuator heavily, but then needs to let the actuator settle before reading a track. The weight of the actuator assembly, the diameter of the platters, and the servo strategy (finding the right track) are the key parameters.

  3. Rotational Delay /Latency:

    Latency is solely determined by the rotational speed. The aerodynamics of the head flying over the platter as well as the need to control the movements of the platters limits the maximum speed, in consequence, smaller diameters allow higher rotational speed.

  4. Access Times:Random access time consists of seek and rotational delay, followed by the data transfer rate. The latter is determined by data density and rotational speed. Access to subsequent blocks avoids the overhead of seek and rotational latency, so that access to large objects stored in contiguous sectors on the same track or at least on the same cylinder is priviliged. Zero latency disks can start reading whenever useful data appears under the head, they thus reduce the rotational delay of large accesses.
  5. Error Rate:

    We distinguish between soft and hard error. The prime example for a soft error is a misregistration during a read, a repetition of the read will yield the desired data. A hard error is a permanent inability to recover data. Error sources are interference and noise.

    1. Interference through cross-talk between different channels or through feedthrough.
    2. Interference by Track Misregistration.
    3. Interference because of Imperfect Overwrites / Incomplete Erasures.
    4. Interference through side fringing (when the head picks up flux changes from an adjacent track.
    5. Bit loss due to Intersymbol Interference.
    6. Defects or random properties resulting in Media Noise.
      For example, a spot on the surface does not retain magnetization because of a manufacturing problem or because of a previous head crash. A modern disk drive has spare sectors on each track and complete spare tracks to substitute for sectors that have these defects. Even without an outright defect, the magnetic properties of the medium vary.
    7. Electronic Noise is caused by random fluctuations typically in the first stage amplifier in the reproducing circuit.
    8. Head Noise:The magnetic flux in both write and read heads is subject to thermally induced fluctuations in time.

    Error rate is controlled through the use of Error Control Codes.

  6. Reliability:

    Disks fail because of a variety of reasons. Modern disks can monitor the error rate to predict failure, allowing the system to automatically move data to other disks or at least to send out an error message to the human administrator / operator.

  7. Power Use:Operation of a disk is a major source of battery drain in mobile systems.
  8. Interface to the system.
© 2003 Thomas Schwarz, S.J., COEN, SCU                                                                              SCU            COEN            T. Schwarz            COEN 180