Re: [openstack-dev] [Neutron] High bandwidth routers

2014-06-24 Thread Mark McClain

On Jun 23, 2014, at 9:21 AM, CARVER, PAUL 
pc2...@att.commailto:pc2...@att.com wrote:

Is anyone using Neutron for high bandwidth workloads? (for sake of discussion 
let’s “high” = “50Gbps or greater”)

With routers being implemented as network namespaces within x86 servers it 
seems like Neutron networks would be pretty bandwidth constrained relative to 
“real” routers.

As we start migrating the physical connections on our physical routers from 
multiple of 10G to multiples of 100G, I’m wondering if Neutron has a clear 
roadmap towards networks where the bandwidth requirements exceed what an x86 
box can do.

Is the thinking that x86 boxes will soon be capable of 100G and multi-100G 
throughput? Or does DVR take care of this by spreading the routing function 
over a large number of compute nodes so that we don’t need to channel 
multi-100G flows through single network nodes?

I’m mostly thinking about WAN connectivity here, video and big data 
applications moving huge amounts of traffic into and out of OpenStack based 
datacenters.


There are few internal implementations of the l3 plugin that are backed by 
dedicated hardware vs commodity+network namespaces.  Of those few, all are site 
specific (due to limited feature support and not likely to upstreamed).

mark

___
OpenStack-dev mailing list
OpenStack-dev@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev


[openstack-dev] [Neutron] High bandwidth routers

2014-06-23 Thread CARVER, PAUL
Is anyone using Neutron for high bandwidth workloads? (for sake of discussion 
let's high = 50Gbps or greater)

With routers being implemented as network namespaces within x86 servers it 
seems like Neutron networks would be pretty bandwidth constrained relative to 
real routers.

As we start migrating the physical connections on our physical routers from 
multiple of 10G to multiples of 100G, I'm wondering if Neutron has a clear 
roadmap towards networks where the bandwidth requirements exceed what an x86 
box can do.

Is the thinking that x86 boxes will soon be capable of 100G and multi-100G 
throughput? Or does DVR take care of this by spreading the routing function 
over a large number of compute nodes so that we don't need to channel 
multi-100G flows through single network nodes?

I'm mostly thinking about WAN connectivity here, video and big data 
applications moving huge amounts of traffic into and out of OpenStack based 
datacenters.

___
OpenStack-dev mailing list
OpenStack-dev@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev


Re: [openstack-dev] [Neutron] High bandwidth routers

2014-06-23 Thread A, Keshava
Hi,
I think there is no much consideration of  L3 forwarding capacity,  of the 
order of 100G in Network-Node(NN).
Not sure current software queues in NNare capable of handling 100G times of 
packet rate.
(Of course for compute node there will  SRIOV to speedup these)

Instead you can consider  of having multiple Network Nodes deployed, so that L3 
forwarding will be distributed across multiple NN.
Make sure you will have separate public  IP for each of these NNs, so that any 
session related issue will not have issues in NN.


Regards,
Keshava.A.K.


From: CARVER, PAUL [mailto:pc2...@att.com]
Sent: Monday, June 23, 2014 6:51 PM
To: OpenStack-dev@lists.openstack.org
Subject: [openstack-dev] [Neutron] High bandwidth routers

Is anyone using Neutron for high bandwidth workloads? (for sake of discussion 
let's high = 50Gbps or greater)

With routers being implemented as network namespaces within x86 servers it 
seems like Neutron networks would be pretty bandwidth constrained relative to 
real routers.

As we start migrating the physical connections on our physical routers from 
multiple of 10G to multiples of 100G, I'm wondering if Neutron has a clear 
roadmap towards networks where the bandwidth requirements exceed what an x86 
box can do.

Is the thinking that x86 boxes will soon be capable of 100G and multi-100G 
throughput? Or does DVR take care of this by spreading the routing function 
over a large number of compute nodes so that we don't need to channel 
multi-100G flows through single network nodes?

I'm mostly thinking about WAN connectivity here, video and big data 
applications moving huge amounts of traffic into and out of OpenStack based 
datacenters.

___
OpenStack-dev mailing list
OpenStack-dev@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev


Re: [openstack-dev] [Neutron] High bandwidth routers

2014-06-23 Thread Collins, Sean
You can use the provider networking API extension for Neutron, in order
to utilize non-openstack L3 hardware.

-- 
Sean M. Collins
___
OpenStack-dev mailing list
OpenStack-dev@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev