Topics / Benchmarks / Options
| No | Topic | Benchmarks | Options | Result |
|---|---|---|---|---|
| 1 | WAL/Manifest | –disable_wal –wal_bytes_per_sync | fillseq/random | PPT |
| 2 | Memtable | –write_buffer_size –max_file_size | fillseq/random readrandom | PPT |
| 3 | Compaction | –base_background_compactions –compaction_style | fillseq/random readseq/random seekrandom | PPT |
| 4 | SSTable | –write_buffer_size –max_file_size –block_size | fillseq/random readseq/random seekrandom | PPT |
| 5 | Bloom Filter | –bloom_bits | readhot/random seekrandom | PPT |
| 6 | Cache | –cache_size –block_size | readhot/random seekrandom | PPT |
5 Steps of Experiment
Hypothesis
- What changes will be happen internally, if option changes?
- How will internal changes affect the metrics?
- What result and graph do you expect?
Design
- Do the simplest and smallest experiment that can test your hypothesis.
- Do not experiment with multiple independent variables at once from the beginning.
- Do not let uncontrolled variables ruin your experiment.
- Variables
- Independent
- Options, Benchmarks, # of KV pairs
- Dependent
- Metrics (Throughput, Latency, WAF/SAF/RAF)
- Controlled
- Enviornment, Page Cache, Compile Options, Existing DB, Compression Ratio, Bloom Filter Bits
- Independent
Run Experiment
- Please use shell/python script.
- echo, redirection, pyplot, …
- Do not change experiment enviornment.
- Please use shell/python script.
Result and Discussion
- Draw graphs and figures that explain experiments.
- Verify your idea and hypothesis with result.
- Explain why your hypothesis is correct or not.
Presentation
- Present your experiments in 10 minutes.
- Write a document that explains your experiment.
- git-book