If your total data size is 638GB, you have an average shard size of less than 300MB, which is very small and inefficient. The guideline on shard count relative heap size is a maximum aimed at reduce the risk for oversharding, which is one of the most common problems when dealing with time series data. The general rule of thumb is that you should aim to have a maximum of 20 shards per GB of heap, which in your case would mean a maximum of 600 shards across the cluster. Note that the limit is not a recommended number of shards nor a level any cluster is guaranteed to be able to handle. In a properly configured cluster I would expect the shard count to be considerably smaller than the maximum as the recommended shard size is often between 10GB and 50GB when dealing with time series data.