Interesting! We did something similar for distributed query engines like presto and Spark, instead of indexes we use partitions and buckets, the benefits are less compute, time and money.