You can do what Prometheus does and pre-aggregate; You size it at 1s buckets or .5s buckets, store the values of those buckets, and increment them as data point

You can do what Prometheus does and pre-aggregate; You size it at 1s buckets or .5s buckets, store the values of those buckets, and increment them as data points come in. You can store the data-points individually too, and regenerate the histogram if you really need to, but it's far more efficient to store the aggregations of a histogram.