Drive Operations
Chapter 3
Technical Reference Manual
Page 21
Command Execution and Buffer Management
The drive dynamically adapts to queuing and non queuing depending on
whether a Queue Tag Message is presented with the command. A 512Kbyte
onboard segmented buffer is dynamically configured to adapt to the particular
drive configuration or operating environment. Look ahead reads are performed
by the drive when there are no queued commands awaiting execution. Look
ahead reads are not performed when there are commands in the queue since
another command will be waiting for execution as soon as the current command
completes and because the queue affords pre-knowledge of subsequent
commands instead of having to anticipate them.
Tagged Command Queuing
The drive when operating in a Tagged Command Queuing environment can
queue up to 32 commands. Commands in the queue are re-ordered to maximize
I/O throughput. There are three types of re-ordering that the drive may do:
•
Seek Ordering based on physical cylinder position in order to minimize
actuator motion
•
Rotational Position Ordering (RPO) to minimize rotational delays
•
Read/Write Coalescing to minimize overhead
Command re-ordering is controlled by Message Code (Simple, Head of Queue or
Ordered Queue) during the Message Phase. Re-ordering of seeks issued with a
Simple Queue Tag message may also be restricted using the Queue Algorithm
Modifier bits in MODE SELECT page 0A
H
. Refer to the Eleventh Generation
SCSI Interface Manual for additional information on Messages and MODE
SELECT page 0A
H
.
Seek Ordering
Commands in the queue which involve seeks are re-ordered to minimize actuator
motion using a C-SCAN (circular scan) method. Seeks already in the queue are
re-ordered so they can be executed sequentially on a sweep toward the inner
diameter (ID). Any new seeks entering the queue ahead of the sweep is ordered
for execution during the current sweep. Any new commands requiring seeks
entering the queue behind the current sweep are held for re-ordering during the
next OD to ID sweep. When there are no more commands in the queue ahead of
the current sweep, the drive executes the command with a target address closest
to the OD.
Rotational Ordering
RPO is done whenever there is more than one command in the queue waiting for
execution. The RPO algorithm accounts for both actuator seek time and
rotational latency time to determine which command in the queue will require
the least amount of time (from completion of the currently executing command)
to reach the target location.
