Re: [controller-dev] FW: [Opendaylight-users] Clustering in ODL(Boron-SR3)

[Tom Pantelis]

for #1 - the traffic bandwidth between nodes depends on the datastore
transaction activity of the applications you're running.


On Thu, Jul 20, 2017 at 4:57 PM, Xingjun Chu  wrote:

> I found this question on user mailing list, kind of interesting .  Any
> comments from controller team.
>
>
>
> Regards
>
> Xingjun
>
>
>
> *From:* opendaylight-users-boun...@lists.opendaylight.org [mailto:
> opendaylight-users-boun...@lists.opendaylight.org] *On Behalf Of *QF Qin
> *Sent:* Thursday, May 25, 2017 1:44 PM
> *To:* opendaylight-us...@lists.opendaylight.org
> *Subject:* [Opendaylight-users] Clustering in ODL(Boron-SR3)
>
>
>
> Hi all,
>
>
>
> Recently I attempted to deploy a ODL cluster and encountered some
> questions.
>
>
>
> 1. Is there any estimation on the bandwidth of inter-controller traffic? I
> measured this bandwidth on port 2550 and found it rather large. For
> example, for a 3-node cluster over a linear topology of 50 switches, the
> leader sends traffic as much as 10 Mbps to any follower. Is it common?
>
>
>
> 2. Is it possible to assign controllers to switches in a partitioning way?
> For example, I have 6 switches and 3 controllers. And I want to:
>
> Assign switch 1, 2 *only* to controller A;
>
> Assign switch 3, 4 *only* to controller B;
>
> Assign switch 5, 6 *only* to controller C.
>
> Does ODL cluster support such setting? When I tried this, the L2 Switch
> did not work well.
>
>
>
> Thanks,
>
> Qiaofeng
>
> ___
> controller-dev mailing list
> controller-dev@lists.opendaylight.org
> https://lists.opendaylight.org/mailman/listinfo/controller-dev
>
>
___
controller-dev mailing list
controller-dev@lists.opendaylight.org
https://lists.opendaylight.org/mailman/listinfo/controller-dev

Re: [controller-dev] FW: [Opendaylight-users] Clustering in ODL(Boron-SR3)

[Robert Varga]

On 20/07/17 22:57, Xingjun Chu wrote:
> I found this question on user mailing list, kind of interesting .  Any
> comments from controller team.
> 
>  
> 
> Regards
> 
> Xingjun
> 
>  
> 
> *From:*opendaylight-users-boun...@lists.opendaylight.org
> [mailto:opendaylight-users-boun...@lists.opendaylight.org] *On Behalf Of
> *QF Qin
> *Sent:* Thursday, May 25, 2017 1:44 PM
> *To:* opendaylight-us...@lists.opendaylight.org
> *Subject:* [Opendaylight-users] Clustering in ODL(Boron-SR3)
> 
>  
> 
> Hi all,
> 
>  
> 
> Recently I attempted to deploy a ODL cluster and encountered some questions.

[+openflowplugin-dev, +l2switch-dev]

> 1. Is there any estimation on the bandwidth of inter-controller traffic?
> I measured this bandwidth on port 2550 and found it rather large. For
> example, for a 3-node cluster over a linear topology of 50 switches, the
> leader sends traffic as much as 10 Mbps to any follower. Is it common?

From the CDS side, it depends on the number and kind of transactions
executed. Also Carbon should have better Leader->Follower efficiency.

From the OF side, I think OFP gathers stats as fast as possible, hence
that traffic should be pretty much constant.

> 2. Is it possible to assign controllers to switches in a partitioning
> way? For example, I have 6 switches and 3 controllers. And I want to:
> 
> Assign switch 1, 2 *only* to controller A;
> 
> Assign switch 3, 4 *only* to controller B;
> 
> Assign switch 5, 6 *only* to controller C.
> 
> Does ODL cluster support such setting? When I tried this, the L2 Switch
> did not work well.

This is an question on OFP mostly. I am not sure what OFP allows for and
what l2switch is doing (and how).

From the perspective of spreading datastore load, Carbon CDS has
preliminary support for having per-switch shards. The OFP part is not
there and it needs to be tested -- which makes it probably Oxygen material.

Regards,
Robert



signature.asc
Description: OpenPGP digital signature
___
controller-dev mailing list
controller-dev@lists.opendaylight.org
https://lists.opendaylight.org/mailman/listinfo/controller-dev