[ovs-discuss] Topology discovery in ovs

[Peter Phaal]

Hi All,

The original question was about L2 topology discovery. One of the differences between sFlow and NetFlow is that sFlow operates at layer 2 (reporting layer 2 interfaces and MAC addresses) and NetFlow typically operates at layer 3 (reporting router (VLAN) interfaces and IP addresses). Consequently, inferring L2 topology from sFlow data is much easier than using NetFlow.

As Robin pointed out, topology inference is a function of the sFlow/NetFlow collector. Inferring L2 topology from sFlow involves examining the sampled packet headers and forwarding paths reported by all the switches in the network, see:

http://blog.sflow.com/2009/05/packet-paths.html

Installing Host sFlow on the hypervisor (http://host-sflow.sourceforge.net/) helps topology discovery by exporting virtual machine information (UUID, hostname, MAC addresses) that can be joined with sFlow data from OVS to identify which vNICs are associated with each VM.

Linking the virtual and physical topology topology gets interesting if you use tunneling (GRE, VxLAN etc), but it is possible for an sFlow analyzer to tie the layers together based on sFlow data:

http://blog.sflow.com/2012/05/tunnels.html

There is an inherent limitation in using traffic data to map topology - you can only see links that carry traffic. However, even backup links carry some traffic (STP, LLDP etc.) so it generally isn't much of a problem.

Regards,
Peter

On Jun 4, 2012, at 7:50 PM, Motonori Shindo wrote:

> Kim and Robin,
> 
> NetFlow and sFlow can give you some idea how your network topology looks like, but it is difficult to draw a complete picture of the network with NetFlow/sFlow because they are not designed to do such a thing. 
> 
> Theoretically NetFlow and sFlow can tell on which interface a particular flow traverses in the network and hence allows us to deduce the network topology from there. In order to do so, however, NetFlow/sFlow collector would have to correlate one flow record exported by one device to another flow record exported by another device. This is extremely difficult to do when there's a massive amount of flows in the network (and it is often the case). In some case (but it is not the case for OVS though), flow is "sampled" (meaning that not all packet are collected to generate flow records) and it makes topology discovery by NetFlow/sFlow even more difficult. 
> 
> Regards,
> 
> ---
> Motonori Shindo
> 
> On 2012/06/05, at 2:41, Robin Wang wrote:
> 
>> Hi Kim,
>> 
>> I found an email that indicates OVS doesn't have built-in support. But it was a year ago, not sure whether it's supported now.
>> http://openvswitch.org/pipermail/discuss/2011-July/005400.html 
>> 
>> For sFlow, to me, it is similar to NetFlow, and usually used to track dynamic traffic in a network with the help of flow collector. There're lots of different collectors. What you can get through sFlow depends on what features the collector provides, not OVS. Not sure if there's a collector provides topology things. Maybe someone who is familiar with collector could give us some advice. Thanks.
>> 
>> Best,
>> Robin
>> 
>> 2012/6/5 Kim, Hyojoon <joonk at gatech.edu>
>> Hi all,
>> 
>> 
>> 
>> What is the most common way to do layer-2 network topology discovery in OVS? Some searching gives me a hint that people install lldpd, and query that table once it’s populated. Is this the most common way?
>> 
>> 
>> 
>> I figured out there is no SNMP support in OVS, but there is sFlow. Is sFlow in OVS capable of doing topology discovery? It gives me a feeling that it is more about traffic monitoring.
>> 
>> 
>> 
>> Thanks,
>> 
>> Joon
>> 
>> 
>> _______________________________________________
>> discuss mailing list
>> discuss at openvswitch.org
>> http://openvswitch.org/mailman/listinfo/discuss
>> 
>> 
>> _______________________________________________
>> discuss mailing list
>> discuss at openvswitch.org
>> http://openvswitch.org/mailman/listinfo/discuss
> 
> 
> 
> _______________________________________________
> discuss mailing list
> discuss at openvswitch.org
> http://openvswitch.org/mailman/listinfo/discuss

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20120604/a741fd9c/attachment.html>