SI Units for Request Rate (2024)
17 points by noncrab
17 points by noncrab
Request rate is a clearly nonconforming application of either Hz or Bq (for reasons laid out in the post), so I think the obvious, standards compliant solution is 1/s (or s⁻¹).
req/s works just fine for me though.
Previously (thanks @fanf): https://lobste.rs/s/awkjch/si_units_for_request_rate#comments_story_01hngfpf38fvbbv5cxyy46tfxz
(another domain name, so dupe-checker didn't catch it).
Edit: entirely missed the satire tag and took this way too seriously. That'll teach me to pay more attention next time...
I don't really understand why the author feels the need to shoehorn the SI special named units into their use case. There seems to be a misconception that the named units are more "proper" somehow, but their purpose is only to simplify the use of some common derived units (e.g. the ohm, which is rather annoying to write out as kg m^2 s^−3 A^−2) or add semantic precision (as in the case of the becquerel or the sievert). It's not a "silly mistake" that the becquerel is defined specifically for radioactivity, and using it for the completely unrelated purpose of measuring request rates is not only incorrect and confusing to everyone else, but it frankly makes you look like a bit of a dick. As another commenter wrote, the obviously correct thing to do here is to use 1/s.
Somehow I think the Erlang would fit in here but I really don't know enough about queueing theory to tell why.
But echoing @fleebee, "our request rate is X per second" is both unambigious and not that hard to write. Saving one character by introducing B(eqcuerel) is weird.
Erlangs are request arrival rate multiplied by request handling time, so they are dimensionless. (previously)