- Distributed Databases
- Disaggregate Storage and Compute to Improve Throughput and Utilization
- Hailstorm, disaggregation, compute, storage, distributed, database, key-value store, skew, compaction offloading, RocksDB, MongoDB, TiKV, TiDB, YCSB, TPC-C, TPC-E
Hailstorm is a drop-in replacement filesystem for existing distributed LSM-based databases that disaggregates storage from processing, thereby improving throughput and utilization. Hailstorm pools all storage devices at rack-level to reduce load imbance without resharding by allowing individual storage engines to utilize spare storage capacity and bandwidth. The system also addresses CPU load imbalance by allowing expensive compactions tasks to be outsourced to less utilized storage engines.