Upon realizing that Edgestore, their product metadata database, could not continue to scale as Dropbox grew and released new tools and platforms, Jonathan Lee, Stas Ilinskiy, William Ehlhardt and their colleagues decided to design Alki, a petabyte-scale metadata store that could be used for infrequently accessed audit logs and other cold data. Since audit logs are ingested at high volume and frequently read randomly following ingestion, Dropbox decided to build a two-tier system that would enable data to be stored in hot storage (DynamoDB) when written but then transitioned to cold storage (S3) as time passed. Alki is based on LSM trees (which are also used in Cassandra, HBase, and BigTable) and leverages Blackbird (Dropbox’s batch execution system) and StepFunctions for offload and compaction processes.