Database Reference
In-Depth Information
of 5 megabytes. Whereas modern disk technology can pack 150 billion bits
per mm
2
, the RAMAC supported a storage density of approximately 4 bits
per mm
2
(100 bits per inch on each track, with 20 tracks per inch). In order
to improve the signal-to-noise ratio for the recordings, early disk technology
attempted to reduce the flying height of the head to the disk surface; this fea-
ture made its debut in the 1960s. Approximately 15 years after the RAMAC,
IBM introduced the 3340 using what was termed “Winchester” disk technol-
ogy, which is considered the ancestor of modern disk technology. The 3340
introduced a thin-film inductive head that flew just 18
u
m from the surface
on a cushion of air. The bernoulli effect enabled the head to fly that close
to the disk surface. Modern hard drives continue to employ this same basic
flying-head approach of these early washing machine-sized disk units. The fly-
ing height for modern disks is now on the order of 15 nm (just 15 atoms of
air between the head and the disk surface) with the medium flying at 50-150
MPH beneath the head for a typical 7,500-15,000 RPM rotational rate.
Disks continue to be the preferred technology for secondary storage on com-
puting devices from desktop computers to the largest-scale supercomputing
systems. In the following subsections, we will discuss the organization of disk
storage devices, technology trends, and their ramifications on design consid-
erations for storage devices for scientific computing systems.
1.3.1 Fundamentals
Ever since the very first RAMAC disk device, disks have been organized as
a spinning magnetic media on a platter — with multiple platters in the disk
often referred to as a
spindle
. On the platter are concentric
tracks
of recorded
data that are often referred to as
cylinders
for tracks located at the same radius
on different platters on the same spindle. The tracks are in turn subdivided
into “sectors” that contain a fixed number of bytes comprising a disk
block
.
The most common block size is 512 bytes. Consequently, disks are typically
referred to as
block storage devices
due to this aspect of their organization.
Because the circumference of tracks on the inner diameter of the disk unit
are much smaller than for the outer tracks,
zonal
recording packs more sectors
on the outer tracks of the disk unit in order to maintain uniform bit density on
all tracks. Zonal recording maximizes the storage density of the device, but the
sustained bandwidth of data transfers from outer tracks is much higher than
that of the inner tracks of the device. Consequently, algorithms for disk file
systems preferentially pack data on the outermost tracks in order to sustain
maximum performance. Read and write operations that are smaller than the
native block size of the devices will waste bandwidth because the device works
with data only at the granularity of blocks. For write operations, it is necessary
to read an entire block, update a subset of that block, and then write the
complete aligned block back to the disk subsystem. Consequently, unaligned
accesses can suffer, and write operations consume both the read and the write
bandwidth of the disk devices. Composite devices, such as RAID, also have
