SSD Architecture and Benchmarking: Performance Differences by Cell Type and Reliable Testing Methods

NAND flash memory exhibits clear differences in bit storage capacity, lifespan (P/E cycles), and latency depending on the SLC, MLC, and TLC types. SLC stores 1 bit, offering a long lifespan of 100,000 P/E cycles and low latency, while TLC stores 3 bits, resulting in a lower unit cost but a shorter lifespan (5,000 P/E) and higher latency. Most consumer SSDs use MLC or TLC, whereas enterprise-grade SSDs use SLC.

The minimum read/write unit for an SSD is the Page (2KB~16KB), while deletion is possible only at the Block level (128~256 Pages). Performance figures from manufacturer datasheets may lack reliability during benchmarking due to short test durations. SSDs require 'Pre-conditioning' through sustained random writes (30 minutes to 4 hours) to observe performance degradation (Sustaining Mode) caused by contention between Garbage Collection and user requests.

Therefore, Pre-conditioning is necessary for reliable comparison, but worst-case performance does not guarantee optimal performance in actual production environments. It is important to develop in-house benchmarking tools tailored to the specific workloads of each system to comprehensively analyze various metrics such as Throughput, IOPS, and Latency.