Well, this is also somewhat theoretical (as it has not been tested yet :o)), 
but in [1] I implemented select groups as a weighted round-robin of the 
buckets, without any hashing (around [2]). Anyway, if anyone is interested in 
using this feature, I'll promise to fix any problems that might arise.

Regards,
Zoltan.

[1] https://github.com/TrafficLab/of11softswitch
[2] 
https://github.com/TrafficLab/of11softswitch/blob/master/udatapath/group_entry.c#L413

 

> -----Original Message-----
> From: [email protected] 
> [mailto:[email protected]] On 
> Behalf Of Rob Sherwood
> Sent: Tuesday, August 16, 2011 5:15 PM
> To: Opeflow Mailing List
> Subject: [openflow-discuss] Splitting a flow
> 
> On Tue, Aug 16, 2011 at 8:00 AM, Ben Pfaff <[email protected]> wrote:
> >
> > OpenFlow 1.0 doesn't support this behavior.  I can't say 
> for OpenFlow 
> > 1.1, but I doubt it.
> 
> [expanding a bit on what Zoltan just said]
> 
> In theory, OpenFlow 1.1 has support for multi-path routing, 
> where a single flow entry can be hashed across multiple 
> ports, such that the flow is divided among the ports.
> 
> The expected use case for this would be taking a flow entry 
> that aggregated multiple tcp flows (e.g., "all traffic dst to
> 192.168.1.0/24") and using, for example, a hash on the 
> packet's five tupple (src and dst ip + src and dst l4 port + 
> ip protocol) to map individual tcp flows to a port, so 
> depending on what you mean by "flow", this use case might be 
> what you're asking.  The other side of it is that openflow 
> 1.1 does not define the hash function used, so in theory you 
> could use this mechanism to map packets from the same tcp 
> flow went out different ports.
> 
> Unfortunately, all this is theoretical as I have not seen an deployed
> 1.1 implementations or implementations that implement this feature.
> If others know of some, I (as would this list :-) would be 
> very interested in knowing about them.
> 
> Hope that helps,
> 
> - Rob
> .
> _______________________________________________
> openflow-discuss mailing list
> [email protected]
> https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
> 
_______________________________________________
openflow-discuss mailing list
[email protected]
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss

Reply via email to