Skip to content

Latest commit

 

History

History
66 lines (49 loc) · 2.63 KB

QUICKSTART.md

File metadata and controls

66 lines (49 loc) · 2.63 KB
Raw

Quick start

Data Ingest and Database Preparation

To ingest data for a TPC-H benchmark with scale factor 10 using this toolkit, run the following command:

./prepare_benchmark \
    --dsn postgresql://postgres@localhost:5432/swarm64_benchmark \
    --benchmark tpch \
    --scale-factor 10 \
    --schema s64da_performance

Adapt the values for the parameters --benchmark and --scale-factor for other benchmarks and scale factors accordingly. If the system you want to benchmark runs on a different host, you need to adapt the value for --dsn as well.

Benchmark Running

To run a bechmark that has been prepared with the prepare_benchmark script, execute the following command:

./run_benchmark \
    --dsn postgresql://postgres@localhost:5432/swarm64_benchmark \
    tpch

Change tpch to a different value if you ingested data for a different benchmark in the step above.

Please refer to README.md for a detailed description of all the available parameters and options.

Umbra Quick Start

Using the provided demo script

The provided demo-umbra-htap script automatically starts an Umbra server, ingests the initial HTAP database population, and finally runs the HTAP benchmark. The script should be invoked as follows:

UMBRA_PATH=<path-to-umbra-binaries> \
DB_PATH=<path-to-put-database-files> \
UMBRA_PORT=<port> \
./demo-umbra-htap \
    <scale-factor> \
    <oltp-workers> \
    <olap-workers>

For example, the following invocation

UMBRA_PATH=/home/xyz/umbra/bin DB_PATH=/raid/db ./demo-umbra-htap 1 16 2

will use the /home/xyz/umbra/bin/sql and /home/xyz/umbra/bin/server binaries to configure and start an Umbra instance which places its database files in the directory /raid/db/htap/. The script will then load the initial database population at scale factor 1 (corresponding by default to 20 warehouses, see benchmarks/htap/lib/helpers.py), and subsequently run the HTAP benchmark with 16 OLTP threads and 2 OLAP threads. For maximum performance, DB_PATH should point to a reasonably fast SSD drive (especially in terms of write bandwidth and response time). Alternatively one can also have DB_PATH point to directory into which a tmpfs is mounted. UMBRA_PORT can be omitted and defaults to 54321.

More fine-grained control over the benchmark is possible by manually adjusting the contents of the demo-umbra-htap script, which internally simply calls the prepare_benchmark and run_benchmark Python scripts with some sensible default parameters. Please note that both scripts require the --umbra flag to be set in order to ensure compatibility with Umbra.