By "two-armed router" I mean a router that is using two separate networks.
In the case of the tests in my Scenarios 4 and 5, I use the 40-Gbit network for inter-router traffic, and the 'normal' 1-Gbit network for communication between each router and its attached clients. The concept is that this might reduce congestion, cause a massive increase in throughput, and shower everyone involved with glory and honor. Unfortunately there is such a large disparity in speed between these two networks that I doubt this is a fair test of the idea. Showering of glory and honor postponed until further notice. Maybe someday I will find some extra high-speed NICs and cables lying about .... On Fri, Jun 25, 2021 at 8:45 AM Michael Goulish <mgoul...@redhat.com> wrote: > > The upshot: > * Worst-case results across the board have greatly improved. > * HTTP adapter has gotten slightly slower in requests per second. > > Complete results here > <https://docs.google.com/document/d/1dsSd6y90HwfI1YYo9z8uu4-K4mYe4SZIFkjRkzBlYI0/edit?usp=sharing> > . > > Next up -- throughput tests using iperf3. > > > The tests use 'hey' load generator on one machine, and 'nginx' server on > the other. > > The two machines are connected by a 40 Gbit/sec network. Receiving machine > has 32 cores, sender has 16. > > The tests are: { 1 sender, 10 senders } { TCP adapter, HTTP1 adapter } { > No Router, One router, Two routers, One 2-armed router, Two two-armed > routers } . > >
