HopsFS is based around using an in-memory database as the metadata store for a filesystem - https://www.usenix.org/conference/fast17/technical-sessions/present

HopsFS is based around using an in-memory database as the metadata store for a filesystem - https://www.usenix.org/conference/fast17/technical-sessions/presentation/niazi https://www.usenix.org/conference/fast17/technical-sessions/... HopsFS is a drop-in replacement for HDFS, and the above paper showed 16X throughput improvement for Spotify's Hadoop workload over Apache HDFS. Since then, HopsFS has added small files in the metadata layer, with over 60X throughput improvement and an order of magnitude lower latency for writing small files (ACM Middleware'18): https://www.logicalclocks.com/millions-and-millions-of-files-deep-learning-at-scale-with-hopsfs/ https://www.logicalclocks.com/millions-and-millions-of-files...