>
> Make sure you do measure routing performance (packet source and sink on
> different machines) and not local stack performance
> (tcpbench/iperf/whatever
> on the machine itself).
>
One lesson I learned the hard way -- make sure your packet source
and sink machines are _faster_ than the system under test.
There's little point testing 2022-era CPU and NIC with 2010-era
desktops or laptops. It proves nothing. And confuses everyone.
(For your own definition of "_faster_", whether using UDP or TCP,
single machines or a cluster.)
John
No comments:
Post a Comment