For me, I want to be able to query data faster. SQL queries on S3 data are affected by read latencies. When you're doing huge aggregations and joins say on Par

For me, I want to be able to query data faster. SQL queries on S3 data are affected by read latencies. When you're doing huge aggregations and joins say on Parquet files, the engine has to read the Parquet metadata multiple times, and read the relevant portions of the data. There are multiple reads going on, and every read incurs latency. Implementing a caching layer really helps, but it's not out of the box.