2.6 Summary of Techniques and Models

In this section, we have explored how by scaling the number of disks in a storage system we can 1) reduce seek distance, 2) reduce rotational delay, 3) reduce overall latency by combining these techniques in a balanced manner, and 4) improve throughput. To achieve these goals, the storage system needs to be configured based on a number of parameters. We have developed simple models that capture the following parameters that influence the configuration decisions:

$\bullet$

disk characteristics in the form of seek and rotational characteristics (S and R),

$\bullet$

read/write ratio (p),

$\bullet$

busyness of the system (q), and

$\bullet$

seek locality (L).

We note that there does not exist a single ``perfect'' SR-Array configuration; instead, there may exist one ``right'' configuration for every workload and cost/performance specification. As we increase the number of disks, and if we properly configure the storage system, under the right conditions, the various models of this section suggest the following rule of thumb: By using D disks, we can improve the overhead-independent part of response time by a factor of $\sqrt{D}$.