Practical study on appropriate usages for the zoned namespace storage

Abstract

This paper describes the interface and characteristics of ZNS, Zoned Namespace, next-generation storage device based on the Nand-Flash, using real SSD and ZNS samples. Generally, it’s easy to choose the Large Zone size which is normally aligned with the logical block size. Because it’s hard to get the required performance values without internal parallelism, even though ZNS products focus on the read performance rather than the write performance. This large zone ZNS needs high processing time for internal merge operations such as the Zone Garbage Collection. And it can take unexpected long latencies and host overhead. This paper suggests the small zone ZNS which has the small zone size aligned with one physical Nand-Flash. The small zone ZNS needs special operating methods because it cannot use internal parallelism with the small zone size compared to the large zone ZNS. In this sense, we solve this problem using the multi-zones concept. It means that we can control target performance values by managing the number of Zone used at a time. We’ve used 32 cores server for measuring performance values and found out that the sequential read throughput can be controlled up to 10.37 times compared to the minimum performance value by assigning multi-zones.