---------------------------------------------------------------------------------
Figure 2.11 A 3.5-inch hard disk with 10 megebyte capacity. Note the arm
supporting the head.(Courtesy of Hewlett-Packard Co.)
---------------------------------------------------------------------------------
move the heads very precisely in a minimum of time. A hard disk with the head
supported by the arm is shown in Figure 2.10.
The heads are mounted on the arm so that they are aligned vertically with
one another. Thus if the head on the first surface is positioned over, say, track
479, all of the other heads will be positioned over track 479 on their respective
surfaces. All of the tracks that are under the heads at the same time are called a
cylinder. The sectors of a disk pack are ordered so that all in one cylinder have
consecutive addresses. Thus, when processing data sequentially, all sectors or
tracks in a cylinder can be accessed before the arm must be moved. This is im-
portant, because moving the arm takes far longer than any other aspect of ac-
cessing disk data. Only one head is selected at a time for reading or writing.
Switching between heads is done electronically and is there fore very fast.
Figure 2.12 shows a removabel disk pack.
Disks are frequently referred to as random access devices because, unlike
tape, any part of a disk pack may be reached in a small fraction of a second.
However, a true ramdom access device is one in which all addresses are equally
accessible. This is clearly not true of a disk. The most accessible sectors are
those which will pass next under the heads. The next most accessible are those
on the same cylinder. Those on distant cylinders will require the most time to
reach because of the time needed to move the arm. Thus it is more appropriate
to refer to disks as pseudorandom access devices.
The total time required to transfer data to or from a disk gas three com-
ponents. First is the time required to reposition the arm, called seek time. This
usally varies from a few milliseconds for a one-track move to scores of mil-
liseconds or more for a move from one end of the disk to the other. Most drives
-------------------------------------------------------------------------------------
Figure 2.12 Removale disk pack(inside cover) with capacity of 404 mega-
bytes. (Courtesy of Hewlett-Packard Co.)
-------------------------------------------------------------------------------------
give a specification for an average, or random, move, which should be used in
calculations.
The secound component is the time from when the arm is positioned to
when the first sector of the transfer moves inder a head. This time is called
latency and depends on the speed of rotation of the disk. In the best case there
would be no wait at all, while in the worst case where the beginning of the sec-
tor had just been missed, the wait would be the time required for a full revolu-
tion. On the average, the latency will be equal to the time required for the disk
to turn half a revolution. For example, at 3600 rpm the disk turns 60 revolutions
every second. Thus a full revolution requires 16.7 milliseconds. The average
latency is the time required for half a revolution, or 8.3 milliseconds.
The third component is the time actually required to transfer the data. If
we assume that the transfer is done within the bounds of one cylinder, which
will usually be the case, the time will be that required for the sectors involved
to pass, one at a time, beneath the head or heads. For example, a disk spinning
at 3600 rpm, with 40 sectors of 256 bytes each per track, will require(1 / 60)
* (5 / 40)=0.00209 second or 2.09 miliseconds to transfer a block of five
sectors wich contains 1280 bytes. Average transfer rates should alwats be cal-
culated on the basis of the time required for the data to pass the head. The
specification given for transfer rate on disk drives is frequently the peak or burst
rate. This is alwats higher than the average rate due to the gaps between sec-
tors.
Seek time dominates the total time required for a disk access, typically fol-
lowed by the latency time, and finally the data transfer time. When it is possible
to do a transfer without moving the arm, there is a great savings in the total
time, and there are some schemes that attempt to minimize the arm travel for
this reason. Figure 2.13 shows a cabinet containing two disk drives.
Disk operations are simple, because the disk driver and controller usually
take care of automatically moving the arm to the required cylinder, selecting the
appropriate head, checking the sector addresses, and handling any errors. The
basic operations are thus reduced to reading or writing a specified number of
sectors, beginning at a given sector address. Data transfers must always be for
an integral number of sectors. It is not usually possible to transfer a fraction of
a sector. Thus the