PLIS: Persistent Learned Index for Strings

Persistent memory, emerging as a potential replacement for the next generation of main memory, is gradually gaining prominence. Presently, index structures based on persistent memory primarily focus on B+ trees, hash tables, and indexes, with significant performance improvements observed in recent research on learned indexes. However, most efforts concentrate on reducing the update costs of learned indexes, leaving inadequate support for string key types. Therefore, this paper aims to create a persistent memory string learned index structure capable of handling variable-length string keys, reducing write amplification, and ensuring crash consistency. We evaluate SLIP cost effectiveness ratio using real and synthetic datasets, results show outperforming state-of-the-art string learned indexes SIndex and SLIPP across various workloads.