Re:Show HN: A benchmark for analytical databases (Snowflake,...
by singhrac on Wednesday July 13, 2022 @01:38PM
(#32086097)
A small suggestion: since the machines aren’t always comparable (eg Redshift) maybe it makes sense to allow the user to make a plot of ($/hr) vs query time? I realize this adds to the complexity, but most people in this space are making a trade off between (maintainability/ease-of-use, long-running costs, performance), and plotting the latter two would probably highlight the Pareto frontier.
[ Parent
| on HN ]
Re:Show HN: A benchmark for analytical databases (Snowflake,...
by zX41ZdbW on Wednesday July 13, 2022 @02:05PM
(#32086499)
Yes, this is what I want to include - the cost metric.
Although it will be tricky for systems with consumption-based pricing: BigQuery and Athena.
And Snowflake has almost consumption based pricing - while you pay for warehouse time, it can quickly spin it up and down based on load.
Maybe use some assumption like - the price for running queries on consumption based systems is equalized to one hour of run time of other systems...
Re:Show HN: A benchmark for analytical databases (Snowflake,...
by ruw1090 on Wednesday July 13, 2022 @04:30PM
(#32088361)
Yea, Redshift looks like it was run on an ra3.xl (4 cores) and is comparing against clickhouse on a c6a.4xl (16 cores). I suspect if this were normalized Redshift would have an edge.
Re:Show HN: A benchmark for analytical databases (Snowflake,...
by zX41ZdbW on Wednesday July 13, 2022 @04:48PM
(#32088562)
Redshift configuration was selected to get 16 cores in total, so it is 4 nodes of ra3.xplus 4vCPU to get 16vCPU in total.
https://github.com/ClickHouse/ClickBench/tree/main/redshift
We can add more results on different configurations easily.
There is also Redshift Serverless, which does not have any hardware configuration to tune.