Very large databases always need fast and reliable operations. Existing SQL engine and object-relational mapping (ORM) implementations take advantage of efficient in-memory caching mechanisms, which can really make the difference in terms of efficiency. However, RAM is limited by cost and hardware, compared to the increasing availability and low cost of flash SSD storage. This paper presents a strategy to use such storage as an extension to the RAM buffer of a SQL engine.

Using SSD storage as a caching mechanism is a common trend in the market. Major software and hardware vendors are providing caching mechanisms that fully exploit the availability of SSD storage. However, such mechanisms act at different layers, namely pure cache at the operating system level (for example, Google’s BCache) or database bufferpool extension (for example, IBM DB2’s TAC or Oracle’s Exadata Smart Flash Cache).

The framework proposed, flash as a cache extension (FaCE), has a twofold contribution. First, according to the authors, “it provides flash-aware strategies for managing the flash cache that can be designed and implemented independently from the RAM buffer management policy.” FaCE also optimizes write operations in the flash memory: this is important due to the fact that random writes in SSD devices are five to ten times slower than sequential writes. Second, FaCE accelerates database system recovery from a failure. Thanks to the nonvolatility of flash memories, data pages in the cache are utilized to minimize recovery overhead.

FaCE has been implemented in a PostgreSQL server engine and evaluated and compared with other products using a relatively small database (50 G) in a data center. The data type used was unknown, and it has been impossible to obtain the software to replicate the experiments. However, the study promotes encouraging results compared to off-the-shelf products, which raises my interest in contacting the authors to further clarify the possibility of exploiting their framework in a production environment.