>I've seen some claims online that, contrary to what Elastic recommends in their docs, a few machines with huge heaps (100+ gb) is the way to go, rather than ma

>I've seen some claims online that, contrary to what Elastic recommends in their docs, a few machines with huge heaps (100+ gb) is the way to go, rather than many machines with 20gb heaps. Usually the recommendation is less than 32GB - this link has some more discussion about it: From https://discuss.elastic.co/t/es-lucene-32gb-heap-myth-or-fact/22920/9 https://discuss.elastic.co/t/es-lucene-32gb-heap-myth-or-fac... It seems whether it's better or worse depends on your data set . But I would love to see tests of different kinds of workloads with large or smaller heaps.