Weighting ratio

[MortenBroerup]

In the RPM weighting, you put 50 % weight on object transfers on new connections, and 50 % weight on object transfers on existing HTTP connections.
Is this representative of common internet traffic usage patterns? I would guess that much more than half of all objects are transferred on existing HTTP connections, which - if true - would put much less weight on the handshake cost.

Furthermore, HTTP/3 has fewer (TCP and TLS) handshake round trips than HTTP/2, so I fear that the first parameters in the RPM equation are on track to become obsolete.

For long term relevancy, you might want to consider object transfer on existing HTTP connections the only parameter for the result, and show the speed of the handshakes as additional information. Or generate two outputs: RPM for existing connections (focus on this), and RPM for new connections (as additional information).

[cpaasch]

I do like the idea of reconsidering the weighting of the different values or at least, make it a parameter of the test.

[cpaasch]

After thinking more about this, I actually think that the current weighting is fine.

Sure, connection-reuse is important and encouraged. But, typical Internet use often does not hit the same hostname. Resources on a webpage typically come from a diverse set of hostnames for which connection-reuse is thus not an option.

So, IMO the weighting is fine.