[Users] ovirt 3.2 - high multicast packet loss

Thu Mar 21 15:48:53 UTC 2013

Hi,

"Plain KVM", that is: Fedora 18 + KVM with VirtManager is suffering the 
same:

On my KVM host:

# cat /sys/class/net/virbr0/bridge/multicast_snooping
1

Now, on 2 virtual (Centos 6.4) nodes:

will give huge packet loss:

92.168.122.151 :   unicast, xmt/rcv/%loss = 300/300/0%, 
min/avg/max/std-dev = 0.258/0.706/1.170/0.097
192.168.122.151 : multicast, xmt/rcv/%loss = 300/261/13% (seq>=2 12%), 
min/avg/max/std-dev = 0.357/0.861/1.944/0.198

Increasing omping to -c 500; packet loss will be about 47%.

Now, on the KVM host:

# echo 0 > /sys/class/net/virbr0/bridge/multicast_snooping
cat /sys/class/net/virbr0/bridge/multicast_snooping
0

Giving it several tries, packet loss is 0%!

I'll give it a try on oVirt tonight.

Winfried

Op 21-03-13 12:17, Antoni Segura Puimedon schreef:
> Michael Tsirkin (Thanks!) proposes to try the following:
>
> try disabling multicast snooping in the bridge
>
> Could you give it a shot?
>
> ----- Original Message -----
>> From: "Winfried de Heiden" <wdh at dds.nl>
>> To: users at ovirt.org
>> Sent: Thursday, March 21, 2013 10:14:04 AM
>> Subject: Re: [Users] ovirt 3.2 - high multicast packet loss
>>
>> So far no reactie about the multicast packet loss......
>>
>> I bumbed into this bug:
>> https://bugzilla.redhat.com/show_bug.cgi?id=880035
>>
>> This looks the same as the problems I am suffering using oVirt 3.2:
>> heavy multicast packet loss after some time.
>>
>> This the bug affect oVirt 3.2 ovirt-node (2.6.1-20120228.fc18)?
>> Can anyone reproduce the problem (omping between 3 virtual nodes)?
>>
>> Winfried
>>
>>
>> Op 18-03-13 16:58, Winfried de Heiden schreef:
>>> Same for Debian 6 (x86_64); 47% packet loss:
>>>
>>> ssmping -c 500 192.168.1.234
>>>
>>> --- 192.168.1.234 statistics ---
>>> 500 packets transmitted, time 500001 ms
>>> unicast:
>>>     500 packets received, 0% packet loss
>>>     rtt min/avg/max/std-dev = 0.352/0.675/0.863/0.072 ms
>>> multicast:
>>>     265 packets received, 47% packet loss since first mc packet (seq
>>>     1)
>>> recvd
>>>     rtt min/avg/max/std-dev = 0.414/0.703/0.885/0.086 ms
>>>
>>>
>>> Winfried
>>>
>>> Hi all,
>>>> Playing around with Red Hat Clustering, it turns out I have a
>>>> hughe
>>>> multicast packet loss: (Centos 6.4 - x86_64  with all updates)
>>>>
>>>> omping 192.168.1.211 192.168.1.212 -c500 (node1)
>>>> omping 192.168.1.212 192.168.1.211 -c500(node2)
>>>>
>>>> will give almost 50% loss!
>>>>
>>>> 192.168.1.211 :   unicast, xmt/rcv/%loss = 500/500/0%,
>>>> min/avg/max/std-dev = 0.330/0.610/0.789/0.064
>>>> 192.168.1.211 : multicast, xmt/rcv/%loss = 500/268/46%,
>>>> min/avg/max/std-dev = 0.416/0.635/0.921/0.066
>>>>
>>>> 192.168.1.212 :   unicast, xmt/rcv/%loss = 500/500/0%,
>>>> min/avg/max/std-dev = 0.388/0.653/0.863/0.069
>>>> 192.168.1.212 : multicast, xmt/rcv/%loss = 500/263/47%,
>>>> min/avg/max/std-dev = 0.396/0.670/1.080/0.074
>>>>
>>>> OK, I am using simple hardware, but this hardware is virtually
>>>> doing
>>>> nothing...
>>>>
>>>> As mentioned on
>>>> https://access.redhat.com/knowledge/sites/default/files/attachments/rhel_cluster_vm_multicast.pdf,
>>>> I set the txqueelen to 500, same result?
>>>>
>>>> I 'm still guessing whether this is an oVirt, virtio or Red
>>>> Hat/Centos issue? Problems only happend after some time; that is
>>>> 200
>>>> mo-pings shows everything is fine.
>>>>
>>>> Anyone?
>>>>
>>>> Winfried
>>>>
>>>>
>>>> _______________________________________________
>>>> Users mailing list
>>>> Users at ovirt.org
>>>> http://lists.ovirt.org/mailman/listinfo/users
>>> _______________________________________________
>>> Users mailing list
>>> Users at ovirt.org
>>> http://lists.ovirt.org/mailman/listinfo/users
>> _______________________________________________
>> Users mailing list
>> Users at ovirt.org
>> http://lists.ovirt.org/mailman/listinfo/users
>>
> _______________________________________________
> Users mailing list
> Users at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/users