Crystal

Cloud analytical databases employ a disaggregated storage model, where the elastic compute layer accesses data persisted on remote cloud in block-oriented columnar formats. Given high latency and low bandwidth to limited size of fast local storage, caching at node is important has resulted renewed interest for analytics. Today, each DBMS builds its own solution, usually based file-or block-level LRU. In this paper, we advocate new architecture smart cache system called Crystal , that co-located with compute. Crystal's clients are DBMS-specific "data sources" push-down predicates. Similar spirit DBMS, incorporates query processing optimization components focusing efficient serving single-table hyper-rectangles regions. Results show Crystal, small source connector, can significantly improve latencies unmodified Spark Greenplum while also saving from storage.