Beyond RAM: A Technical Deep Dive into Dragonfly SSD Data Tiering

Scaling an in-memory workload past your RAM limit shouldn't mean accepting worse latency, a bigger cluster, or a lock-in to one cloud provider's instance family. In this session, Dragonfly engineers Roman Gershman and Vlad Oleshko walk through how Dragonfly SSD Data Tiering actually works — from the io_uring-based async I/O pipeline and the three-state data model (hot, cold, cooled) to DASH table indexing, page cache bypass via O_DIRECT, and small-value compaction. They'll contrast this directly with how ElastiCache and Valkey handle tiering, explain where single-threaded architectures hit a wall under disk I/O, and share benchmark results and real production workload data. If you're managing a dataset that's outgrown your DRAM budget, this is the session to attend.