[c-nsp] RSP failover vs Chassis failover for switch/router clusters

[Mark Tinka]

On 5/25/17 1:12 PM, George Giannousopoulos wrote:

> Would you prefer a chassis or an RSP failover having in mind the highest
> possible service availability?

So the only chassis-based clustering we've done is switch stacking.
We've stayed away from clustering routers, or even running things like
MC-LAG.

Our switch clusters are discrete units running as several members that
form a cluster. Not much you can do there, as one member is main control
plane, and another the backup.

For all our chassis-based routers/switches, we've chosen to have dual
control planes if they are supported. The incremental cost is marginal,
and looking at how often they have failed (thankfully, one at a time,
and not both together), well worth it.

In the old days, when we ran Cisco 6500 for our core switching, we
initially run a single RP. These days, we are running Arista
chassis-based switches for high-capacity core switching, and these come
with 2 control planes each, as a matter of course.

Mark.