Compression. For our app, we do multilevel caches, where warm is arrow (in-memory, inflight, streaming, ...) and cold (persistent/scaleout) is parquet (ssd, s3,

Compression. For our app, we do multilevel caches, where warm is arrow (in-memory, inflight, streaming, ...) and cold (persistent/scaleout) is parquet (ssd, s3, ...).