[c-nsp] CSR1000V and CPU usage

[Pshem Kowalczyk]

On Fri, 14 Aug 2015 at 02:36 Roland Dobbins <rdobbins at arbor.net> wrote:

> On 13 Aug 2015, at 20:05, Mark Tinka wrote:
>
> > That's what we do. Works like a charm, over 12x months now.
>
> Yes, that's a perfect application for it.
>
>
Horses for courses :-)

We have found some applications for it as well:
1. Simplification of the interface between the 'compute' and network -
instead of terminating all of the L3 on physical PEs we terminate that on
the CSR1000v  (as PE) - this way the L2 doesn't have to go far - in most
cases remains within the physical host that runs the VMs and the CSR1kv
(admittedly the L2 for east-west  traffic still spills outside of the
hosts).

2. Small bandwidth NFV - we have a bunch of low bandwidth (and sometime
quite transient) applications for things like L2TP termination or IPSEC
that couldn't justify a physical device on their own (and for various
reasons we don't want to put all the configs on a single 'shared'
devices).  A 10Mb/s licence is perfect for them. Apparently another more
dynamic licensing model is coming as well, where the cost is directly
related to the bandwidth used.

At this stage we're looking at automating the provisioning of some of those
functions (using pre-generated configs and some REST calls) and also
plugging CSR1kv into openstack/kvm environments (to do similar things we
currently do on ESX).

kind regards
Pshem