It'd really be good if someone could share iperf3 performance testings/differencies on them all.
I think, if I recall correctly, cubic is the fastest one, at least when it comes to web servers..
Metering performance of networking from CC protocols point of view ONLY may sense ON REAL LOADING IN PRODUCTION or on production infrastructure (or its ideal 2-nd copy) by hardware or software like TRex packet generators GUDED BY SCRIPTS.
Due to its nature iperf3 (and it’s predecessor iperf) even in multi-flow mode - GREAT ONLY FOR TESTING HRADWARE THROUGHPUT ON NODE LEVEL (mean how CPU, PCI bus, NIC, memory and OS drivers and IP stack ABLE to keep hi-loading packet flow).
On each other level OUTSIDE the server’s case - You need MUCH MORE COMPLEX testing that IN FACT MAY IMITATE REAL TRAFFIC and TROUGHTPUT.
This not only mean MUXING IP’s PROTOCOLS in different proportions but VARIABLE PACKETS FLOW and OVERALL LOADING.
Because of enterprise-level price tag for even aftermarket hardware packet generators (and a much less flexibility, or the same high-price tag on license for mixing types of traffic) are not in budget of most here on forum, better to using
TRex (or similar) SOFTWARE PACKET GENERATORS.
Only ONE LIMITATION that Youhave in this case (like any other software packets generators in this world in 2024, even written on C++ or Go): TOP BANDWIDTH NOT POSSIBLE MORE 200Gb/s,- because all this generator’s code executed in user space (some packet generators have some modules that works in kernel space, so You need custom-compiled kernel, but this is only to be able working on non-top-servers-hardware, to be more accessible for small companies and developers/enthusiasts).
So, go to the
TRex web, read the
docs,
, study, and MORE PRACTICE ON EXAMPLES (
Source 1 and
Source 2, and
Source 3).
And of course, not forget to search solutions in
TRex Google User Forum!
