| 1 |
On Monday, December 08, 2014 11:17:26 PM lee wrote: |
| 2 |
> "J. Roeleveld" <joost@××××××××.org> writes: |
| 3 |
> > create 1 bridge per physical network port |
| 4 |
> > add the physical ports to the respective bridges |
| 5 |
> |
| 6 |
> That tends to make the ports disappear, i. e. become unusable, because |
| 7 |
> the bridge swallows them. |
| 8 |
|
| 9 |
What do you mean with "unusable"? |
| 10 |
|
| 11 |
> > pass virtual NICs to the VMs which are part of the bridges. |
| 12 |
> |
| 13 |
> Doesn't that create more CPU load than passing the port? |
| 14 |
|
| 15 |
Do you have an IOMMU on the host? |
| 16 |
I don't notice any significant increase in CPU-usage caused by the network |
| 17 |
layer. |
| 18 |
|
| 19 |
> And at some |
| 20 |
> point, you may saturate the bandwidth of the port. |
| 21 |
|
| 22 |
And how is this different from assigning the network interface directly? |
| 23 |
My switch supports bonding, which means I have a total of 4Gbit/s between the |
| 24 |
server and switch for all networks. (using VLANs) |
| 25 |
|
| 26 |
> > But it's your server, you decide on the complexity. |
| 27 |
> > |
| 28 |
> > I stopped passing physical NICs when I was encountering issues with newer |
| 29 |
> > cards. |
| 30 |
> > They are now resolved, but passing virtual interfaces is simpler and more |
| 31 |
> > reliable. |
| 32 |
> |
| 33 |
> The only issue I have with passing the port is that the kernel module |
| 34 |
> must not be loaded from the initrd image. So I don't see how fighting |
| 35 |
> with the bridges would make things easier. |
| 36 |
|
| 37 |
Unless you are forced to use some really weird configuration utility for the |
| 38 |
network, configuring a bridge and assiging the bridge in the xen-domain config |
| 39 |
file is simpler then assigning physical network interfaces. |
| 40 |
|
| 41 |
-- |
| 42 |
Joost |
Archives