Authors: Nikhil Ghosh, Peter McNeely, Matthew Roger Nelson
ntran/
│
├── main.go (Entry point, handles CLI)
│ ┌──────────────────┐
├── policy/ │ Policy Interface │
│ │ └──────────────────┘
│ ├── policy.go (Defines interface for all implementations)
│ │
│ ├── Implementations ──────────────────────┐
│ │ ├── serialclient.go (PostgreSQL) │
│ │ ├── duckdbparallelclient.go (DuckDB) │ Share common test cases
│ │ ├── duckdbserialclient.go (DuckDB) │ and benchmarking logic
│ │ ├── coldneondbclient.go (NeonDB) │
│ │ └── prewarmneondbclient.go (NeonDB) ┘
│ │
│ ├── queries.go (Shared SQL test cases)
│ ├── benchmark.go (Performance measurement)
│ └── experiment.go (Results collection)
│
├── schemas/ (Database schemas)
│ ├── schema.sql (Initial setup)
│ └── rollback.sql (Cleanup)
│
└── results/ (Benchmark outputs)
└── *.csv (Raw timing data)
Flow:
┌─────────┐ ┌──────────┐ ┌───────────────┐ ┌──────────┐
│ Setup │ -> │ Execute │ -> │ Random Winner │ -> │ Cleanup │
└─────────┘ └──────────┘ └───────────────┘ └──────────┘
│ │ │ │
schema.sql N parallel Pick 1 result rollback.sql
or serial to commit
transactions
ntran depends on golang 1.23.2, so be sure to get that (https://go.dev/dl/). Then navigate to the ntran directory (i.e. cd ntran).
To build the executable, simply run go build. This will generate an ntran executable.
You can run the ntran executable built above, or from within the ntran directory, you can run ntran by executing the command go run ..
This section will assume you have built the executable (but if you simply just replace ./ntran with go run ., it should work equivalently).
To run an experiment, you must specify the policy to execute. For instance, to run the serial-snapshot experiment, run ./ntran -policy serial-snapshot. Run ./ntran --help to view the full usage and entire list of supported policies.
This policy executes N transactions sequentially under one parent transaction on a postgres database. After each sub-transaction has performed its command, the sub-transaction is rolled back.
This policy expects to connect to a Postgres database whose connection string can be found under the environment variable SERIAL_DATABASE_URL.
This policy executes N transactions on N instances of DuckDB in parallel with one another.
This policy executes N transactions on N instances of DuckDB in sequence with one another.
This policy executes N transactions on N instances of NeonDB in parallel with one another. The "cold" in cold-neondb is a reference to the fact that compute nodes are spun up right before a given transaction is executed.
This policy depends on the neon cli tool for branch management commands. To download, follow the instructions here: https://neon.tech/docs/reference/neon-cli. Don't call the CLI neonctl like the docs suggest. We expect it to be called neon.
The NeonDB project used by the authors is https://console.neon.tech/app/projects/patient-hall-76729406.
This policy executes N transactions on N instances of NeonDB in parallel with one another. The "prewarm" in prewarm-neondb is a reference to the fact that N compute nodes are spun up prior to any transaction being executed.
This policy depends on the neon cli tool for branch management commands. To download, follow the instructions here: https://neon.tech/docs/reference/neon-cli. Don't call the CLI neonctl like the docs suggest. We expect it to be called neon.
The NeonDB project used by the authors is https://console.neon.tech/app/projects/patient-hall-76729406.
ntran provies a way to analyze the results of its experiment runs. This analyzer depends on a Poetry installation, so be sure to get that (https://python-poetry.org/docs/). Once poetry is installed, run poetry install to install its dependencies. Then, to analyze the results and generate .pngs (in the figures/ directory), run poetry run python ntran/analyze.py.