Set -Xms and -Xmx to be the same. More heap means Elasticsearch can keep more data in memory for faster access. But more heap also means that when the Java heap is close to full, the JVM's garbage collector will run a full garbage collection. At that point, all other processing within the Elasticsearch node experiences a pause. So, the larger the heap, the longer the pauses will be. The maximum heap size that you can configure is around 32 GB. Another recommendation to keep in mind is that we should allocate no more than 50% of the total available RAM on the machine to the Elasticsearch JVM. The reason is that the system needs enough memory for the filesystem cache for Apache Lucene. Ultimately, all the data stored on the Elasticsearch node is managed as Apache Lucene indexes, which needs RAM for fast access to the files.

So, if you are planning to store huge amounts of data in Elasticsearch, there is no point in having one single node with more than 64 GB RAM (50% of which is 32 GB, the maximum heap size). Instead, add more nodes if you want to scale.