VMs on shared VLAN mesoscale-openflow are not communicating

[lnevers@bbn.com]

Created a sliver named "2shared" at the InstaGENI aggregate which requested two VM on the shared VLAN "mesoscale-openflow". The create sliver was successful and was able to login to each of the nodes in the sliver. But, the nodes in the sliver cannot ping each other. Nodes in the sliver that show this problem:

  HostName pc5.utah.geniracks.net Port 32314
 
  HostName pc5.utah.geniracks.net Port 32315

Able to duplicate problem. Attaching the RSpec used to create the sliver.

[lnevers@bbn.com]

Capturing Leigh response in the ticket:

On 10/2/12 1:36 PM, Leigh Stoller wrote:

> #54: VMs on shared VLAN mesoscale-openflow are not communicating 

This is a problem I have seen before, but not very often. In this case the
two VMs on pc5 were mapped (by assign) using a different physical
interface.  So one of them is using eth1 for the 1750 vlan and the other is
using eth5. All of this set up properly and if you wanted to, I bet you
could probably contact other nodes on vlan 1750 as long as they were on a
different physical host.

In this case though, the kernel is not handing back arp replies since it
does not think it has to; the request came from itself. But in fact it
needs to cause the virtual ethernet interfaces each have their own macs.

This came up once before, and I do not remember finding a solution that
time around either.

I am pondering but if anyone has a brilliant idea, feel free to speak up!

Lbs

[lnevers@bbn.com]

Capturing additional comment on this ticket:

On 10/2/12 1:58 PM, Nicholas Bastin wrote:

On Tue, Oct 2, 2012 at 11:44 AM, GENI: instageni <​trac@geni.net> wrote:

#54: VMs on shared VLAN mesoscale-openflow are not communicating

FlowVisor? was dead.

-- Nick

On 10/2/12 2:08 PM, Nicholas Bastin wrote:

On Tue, Oct 2, 2012 at 1:36 PM, Leigh Stoller <​lbstoller@gmail.com> wrote:

In this case though, the kernel is not handing back arp replies since it does not think it has to; the request came from itself. But in fact it needs to cause the virtual ethernet interfaces each have their own macs.

This came up once before, and I do not remember finding a solution that time around either.

Erm, there's an arp sysctl for this….

<rustles around an OpenVZ ndoe>

net.ipv4.conf.all.arp_ignore=1 net.ipv4.conf.all.arp_announce=2

may fix it.

-- Nick

Since the problem was reported,have been able to request shared VLAN VMs and have them communicate, closing ticket.