Results
https://blog.fivetran.com/warehouse-benchmark-dce9f4c529c1
Design
This is based on the TPC-DS benchmark, a standard data warehouse benchmark that uses lots of joins, aggregations and subqueries. The TPC-DS queries have been modified somewhat to improve portability across implementations, and eliminate the use of obscure SQL features like grouping-sets. There are two data configurations:
| Data size as uncompressed CSV | Largest fact table |
|---|---|
| 100 GB | 400 million rows |
| 1 TB | 4 billion rows |
There are two configurations for each warehouse:
| Data size | Warehouse | Nodes | Cost / Hour |
|---|---|---|---|
| 100 GB | Redshift | 8 × dc2.large | $2.00 |
| Snowflake | X-Small | $2.00 | |
| Presto | 4 × n1-standard-8 | $1.23 | |
| Azure | ? | ? | |
| BigQuery | - | - | |
| 1 TB | Redshift | 4 × dc2.8xlarge | $19.20 |
| Snowflake | Large | $16.00 | |
| Presto | 32 x n1-standard-8 | $9.84 | |
| Azure | ? | ? | |
| BigQuery | - | - |
Usage
These scripts are intended to be manually copy-pasted into various terminals. You can skip steps 1-4 since gs://fivetran-benchmark and s3://fivetran-benchmark are already populated.