On 23/04/2009 8:51 PM, Anup Patil wrote:
On Tue, 2009-04-21 at 21:59 +0200, Manuel Bouyer wrote:
>>
What network test ? who is sending the datagram and who is receiving them ? Did you check that the dom0 can handle the network load ? What OS is the dom0 running ?Dom0 is linux-2.6.18.8, built from the xen sources. The sender and receiver is iperf ( http://sourceforge.net/projects/iperf/) Receiver is connected to the dom0 by a 100 Mb/s switch, and cpu load in dom0 seems to be all right as other domU's (linux 2.18.8 ) perform differently.
So the scenario is a NetBSD domu sending out via a linux dom0 bridge? Receiver is _another box_ connected via 100Mbps?
And the linux domu operate faultlessly with this test? That would be a crucial piece of info to isolate the issue.
Have you performed the same test from dom0?It's highly possible it's how NetBSD operates under a linux dom0 but more specific info is needed. Have you tried current? It's possible it's been fixed and if it has, this would be a definitive test.
This is something I'm sure they'd like to fix in 5-RC if it's not working, especially if it's already been fixed :) ... A release with incompatible domu kernels for the current supported version of xen isn't ideal.
Sarton