neal young / Amjad16Design

  • We empirically evaluate various stack-based LSM merge-compaction policies, including the default policy for Google Bigtable, Cassandra's size-tiered policy, HBase's exploring policy, and new policies proposed in "Online Bigtable merge compaction". We consider in depth the setting where, to keep read-response time low, the policy must manage compactions to keep the stack size below a specified parameter K. We evaluate the policies by implementing them within the AsterixDB NoSQL database. We also empirically partially validate the theoretical model for stack-based LSM merge-compaction proposed in "Online Bigtable merge compaction".

© Copyrights are reserved by the publishers.
Download for personal and limited academic use only.