The response time of solid state drives (SSDs) has<br />
dramatically reduced according to the spread of non-volatile<br />
memory express (NVMe) devices. These devices have response<br />
times of less than 100 micro seconds on average. The response<br />
time of all-flash-array systems has also drastically reduced<br />
through the use of NVMe SSDs. However, there are applications,<br />
particularly, virtual desktop infrastructure and in-memory<br />
database systems, that require storage systems with even shorter<br />
response time. Their workloads were found to contain many<br />
input-output (IO) concentrations. We define IO concentration by<br />
using a declarative style. Input-output (IO) concentrations are<br />
aggregations of IO accesses. They appear in narrow regions of<br />
the storage volume and continue for periods of up to about an<br />
hour. These narrow regions occupy a few percent of the logical<br />
unit number capacity, include most IO accesses, and appear at<br />
unpredictable logical block addresses. To drastically reduce the<br />
response time of these workloads, we developed automated tiered<br />
storage system called “automated tiered storage with fast memory<br />
and slow flash storage” (ATSMF). The memory component of<br />
ATSMF is a memory with a non-volatile feature. The system<br />
predicts the remaining duration of IO concentration, calculates<br />
the response-time increase during migration and response-time<br />
decrease after migration, and migrates the IO concentrations<br />
if the response-time decrease after migration surpasses the<br />
response-time increase during migration. Experimental results<br />
indicate that ATSMF is at least 20% faster than flash storage<br />
only and its memory access ratio is more than 50%.