New class which records the number of samples by bucket. E.g. if we want to record messages sizes, we could define a number of buckets (of bytes):

• 0 .. 10
• 11 .. 100
• 101 .. 1000
• 1001 .. 10000

This would allow us to find out the distribution of message sizes in a finer grained way, better than histograms.
Possibly add a parameter "key" when adding a value, e.g. "request", "response" etc, which associates a counter with a key.
Inside a given bucket, we could then maintain a hashmap of keys and counters, so we'd get something like

0 .. 100: "request": 1'250, "response": 5'300