Cassandra-stress uses this library/format as well.
There are 2 big advantages of hdr histogram:
it doesn't record a latency histogram for the whole run, but you can configure it to record a latency histogram for e.g. every second in a run. This way we can track latency over time.
when stored to disk, we can use other hdr tools to merge the histograms of different scylla-bench instances for example and do other magic like pruning, filtering etc etc.
Add support for HdrHistogram for recording the latency distribution.
https://github.com/HdrHistogram/hdrhistogram-go
Cassandra-stress uses this library/format as well.
There are 2 big advantages of hdr histogram: