From: David Jones <jones@mc...> - 2003-06-13 18:52:53
So has anyone truly seen better net performance between two UMLs when
using the switch daemon over a tap device? The UML homepage implies that
there is better performance when using the uml_switch.
I ran a basic iperf test between two UMLs on the same host and I got these
~5.18 MB/s via tap
~3.56 MB/s via uml_switch
My iperf parameters were 'iperf -s' on the server end and 'iperf -c
uml-1-2 -f M' on the client side.
On Fri, Jun 13, 2003 at 01:52:43PM -0500, David Jones wrote:
> I ran a basic iperf test between two UMLs on the same host and I got these
> ~5.18 MB/s via tap
> ~3.56 MB/s via uml_switch
Is this using SKAS mode? This would be an area where using SKAS will
improve things a lot (as there are a lot of system calls).
It would be interesting to find out the inherent loss as well. I think
perhaps the pchar app would be useful for that.