net-kvm: connection is failing when using bonded host networking with arp monitoring about kvm-guest-drivers-windows

[YanVugenfirer]

Hi Maurits,

I will take a look at the problem later this week as currently I am out of
the office for public holidays.

Some questions as I never tested such configuration:

1. What are the offload settings of the NICs in the bonding? (ethtool -k)
2. Can you try and disable GRO ?

My suspicion is that exactly as with macvtap no one handles capability
mismatch between the guest (tap device) and the devices on the host. And
Windows guests until Windows server 2012 do not support GRO (we added a
patch recently for it, the feature is called RSC by Microsoft).

Best regards,
Yan.

On Mon, Mar 25, 2013 at 11:08 AM, MvdLande [email protected] wrote:

Hello Yan,

I'm testing a setup on CentOS 6.4 which uses arp to monitor a network bond
interface instead of miimon.
like:
...
IPADDR=172.16.93.6 # VMhost1f
BONDING_OPTS="mode=1 arp_interval=500 arp_validate=all
arp_ip_target=172.16.93.10,172.16.93.20,172.16.93.23,172.16.93.29"
...

Then vlans are added to bond0 which are each added to it's own bridge. The
network bridges are used for the virtual networking.

In the past I had used :
BONDING_OPTS="mode=1 miimon=100"

which is working fine but with arp monitor link failures beyond the
directly connected switch will be detected.

On the host machine networking appears fine and the cluster manager does
not detect any error's.

But the on the guest machine network starts with a lot of trouble. e.a. a
slow connection and often failing until after a couple of minutes the guest
networking fails completely.

Then after manual disabling the guest network interface and enabling it
again the network connection is temporary restored until it fails again.

I have tested several net-kvm driver versions and none are working with
this current setup.

Do you have a clue about what's happening?

Best regards,

Maurits van de Lande

—
Reply to this email directly or view it on GitHubhttps://github.com/YanVugenfirer/kvm-guest-drivers-windows/issues/10
.

---

Daynix Computing LTD
Yan Vugenfirer, CEO
Email: [email protected]
Phone: [phone]
Web: www.daynix.com

[MvdLande]

1. Can you try and disable GRO ?
2. What are the offload settings of the NICs in the bonding? (ethtool -k)

ethtool -k eth0

Features for eth0:
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp-segmentation-offload: on
udp-fragmentation-offload: off
generic-segmentation-offload: on
generic-receive-offload: on
large-receive-offload: off
rx-vlan-offload: on
tx-vlan-offload: on
ntuple-filters: off
receive-hashing: on

ethtool -k eth1

Features for eth1:
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp-segmentation-offload: on
udp-fragmentation-offload: off
generic-segmentation-offload: on
generic-receive-offload: on
large-receive-offload: off
rx-vlan-offload: on
tx-vlan-offload: on
ntuple-filters: off
receive-hashing: on

#2. Can you try and disable GRO ?
done, I'll test again

[MvdLande]

After disabling GRO on eth0 and eth1 the guest network is still not stable.

[YanVugenfirer]

What about the bond interface and the bridges? What properties do they
have?

Best regards,
Yan.

On Mon, Mar 25, 2013 at 12:33 PM, MvdLande [email protected] wrote:

After disabling GRO on eth0 and eth1 the guest network is still not stable.

—
Reply to this email directly or view it on GitHubhttps://github.com/YanVugenfirer/kvm-guest-drivers-windows/issues/10#issuecomment-15386236
.

---

Daynix Computing LTD
Yan Vugenfirer, CEO
Email: [email protected]
Phone: [phone]
Web: www.daynix.com

[MvdLande]

br110 is connected to the guest

ethtool -k bond0

Features for bond0:
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp-segmentation-offload: on
udp-fragmentation-offload: off
generic-segmentation-offload: off
generic-receive-offload: off
large-receive-offload: off
rx-vlan-offload: on
tx-vlan-offload: on
ntuple-filters: off
receive-hashing: off

ethtool -k br110

Features for br110:
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp-segmentation-offload: on
udp-fragmentation-offload: off
generic-segmentation-offload: on
generic-receive-offload: off
large-receive-offload: off
rx-vlan-offload: off
tx-vlan-offload: off
ntuple-filters: off
receive-hashing: off

Best regards,

Maurits

[MvdLande]

I have just tested the network connectivity without the net-kvm driver with an e1000 device. Now it is even worse.
No network connectivity at all!

So net-kvm is not the problem.

Apparently arp monitoring for network bonding does not currently work with libvirt/kvm.