On 10/25/17 10:20 PM, Tony Sarendal wrote:
Configure the interfaces into separate rdomains.
/T
Yup that works. For sure I know the packets are traveling along the wire
using rdomains now. But I wonder how much of the RTT on each packet is
due to the kernel/driver/rdomain code now. And also due to this test
platform being a virtual machine in Linux/XEN environment.
The ping response I get is, across the wire, more than what I got
without the rdomains: round-trip min/avg/max/std-dev =
0.415/0.530/0.890/0.165 ms, not bad but not what I'd like to see from
10GbE. A large std-dev, relatively, and not quite as low as I'd expect
for 10GbE. As this test instance currently is a Xen guest, the interface
is an xnf driver; the 10GbE is bridged on a Linux Xen server's device.
But based on this working, I may be able to justify some hardware
allocations happening so I can build a platform natively OpenBSD.
Thanks, Tony, for the good tip.
CP
