Seek Splitting: Preempting ${\bf T_{seek}}$

The seek delay ($T_{seek}$) becomes the dominant component when the $T_{transfer}$ and $T_{rot}$ components are reduced drastically. A full stroke of the disk arm may require as much as $20$ ms in current-day disk drives. It may then be necessary to reduce the $T_{seek}$ component to further reduce the waiting time.

Definition 3.3: Seek-splitting breaks a long, non-preemptible seek of the disk arm into multiple smaller sub-seeks.

Benefits: The seek-splitting method reduces the $T_{seek}$ component of the waiting time. A long non-preemptible seek can be transformed into multiple shorter sub-seeks. A higher-priority request can now be serviced at the end of a sub-seek, instead of being delayed until the entire seek operation is finished. For example, suppose an IO request involves a seek of $20,000$ cylinders, requiring a $T_{seek}$ of $14$ ms. Using seek-splitting, this seek operation can be divided into two $9$ ms sub-seeks of $10,000$ cylinders each. Then the expected waiting time for a higher-priority request is reduced from $7$ ms to $4.5$ ms.

Overhead:

• 1. Due to the mechanics of the disk arm, the total time required to perform multiple sub-seeks is greater than that for a single seek of a given seek distance. Thus, the seek-splitting method can degrade disk throughput. Later in this section, we discuss this issue further.

• 2. Splitting the seek into multiple sub-seeks increases the number of disk head accelerations and decelerations, consequently increasing the power usage and noise.