Why StarRocks (WIP)

[alberttwong]

Here are some additional questions that get asked about Why StarRocks not documented anywhere else.

Do SIMD optimizations have an effect on JOIN performance?

Yes, if the SIMD optimizations are done on the JOIN operators. However, this optimization can be done on any operator, so it is not really JOIN specific. This is why we are faster than Java based DBs, like Druid and Presto. (or actually why CH is faster than Druid)

Why hasn't ClickHouse implemented data shuffling?

Too big of a change, too much work. And JOIN is a process of iterative optimizations, having shuffling is not enough, they need to go through what we went thru in the last 3 years. This issue has been open for three years: Shuffle / Redistribution based join / aggregation · Issue #12616 · ClickHouse/ClickHouse