A Study of Database Performance Sensitivity to Experiment Settings
Yang Wang, Miao Yu, Yujie Hui, Fang Zhou, Yuyang Huang, Rui Zhu, Xueyuan Ren, Tianxi Li, and Xiaoyi Lu.
Proceedings of the VLDB Endowment, Vol. 15, No. 7
see hardware_config
Please refer to the README in each folder to see how to setup environment, build and run each work.
You can find the correlated raw data in the plot script. (All the plot script produced by gnuplot)
- 1a: Silo (32 worker threads)
- 1b: Aria (12 worker threads and default cross-warehouse rate)
- 1c: Janus (default cross-warehouse rate).
- 1d: Cicada (default cross-warehouse rate)
- 1e: Star (12 worker threads per node)
- 1f: GAM (4 worker threads per node)
- 3: Throughput of Janus and 2PL with extra network latencies (6 warehouses and default cross-warehouse rate)
- 4: Throughputs of H2 with interactive transactions and stored procedure
- 5a : Silo (#warehouse=#worker and no cross-warehouse transactions).
- 5b: Cicada (#warehouse=#worker and default cross-warehouse rate)
- 5c: Janus (6 warehouses and default cross-warehouse rate).
- 6a: Cicada with 50% reads and 10M KVs
- 6b: Cicada with 50% reads and 40M KVs
- 6c: Cicada with 95% reads and 10M KVs
- 6d: HERD with uniform distribution
- 6e: HERD with Zipfian (0.99)
- 6f: TAPIR with 50% read and Zipfian (0.99)