SSD Optimization Guide for Programmers: From Page Alignment to Garbage Collection

This document is the final summary of the 'Coding for SSD' series, outlining 29 key tips that programmers and system engineers need to know to understand SSD characteristics and maximize performance. The main contents are as follows.

1. Understanding Physical Characteristics: SSDs read and write in page units and can only be erased in block units. Since existing data cannot be overwritten, the Read-Modify-Write method is used, which causes Write Amplification. The FTL (Flash Translation Layer) maps logical addresses to physical addresses and converts random writes into sequential writes.

2. Optimizing Access Patterns:

   • Writing: Align data to multiples of the page size (typically 4KB–16KB) when writing, and buffer small data in memory for batch processing.
   • Separating Reads and Writes: Processing large volumes sequentially rather than mixing reads and writes is advantageous for utilizing internal caching and Readahead mechanisms.
   • Thread Strategy: Processing large data with a single thread may be advantageous for both latency and throughput, but frequent writes of small data may yield better throughput with multi-threading.
   • Data Separation: Store frequently changing Hot data and unchanging Cold data separately to improve the efficiency of Garbage Collection.

3. System Configuration:

   • Interface: PCI Express and SAS are faster than SATA.
   • Over-provisioning: Allocate 10–25% of the total capacity as over-provisioning space to improve Wear Leveling and Garbage Collection performance.
   • Enabling TRIM: Configure the OS and file system to support TRIM commands to pass information about deleted blocks to the SSD controller.
   • Partition Alignment: Align partitions to the NAND flash page size when creating them.