[AusNOG] Router on a stick for a production environment

[Chris Kawchuk]

No Caveats that I'm aware of, other than the obvious "if that one interface goes down" parable.

We used to do this with M7i's on a 1G link, with a Foundry NetIron switch underneath it to "fan out" all the VLANs to physical ports. (yeah, this was a long time ago when 1 Gbps/sec was a lot of bandwidth). Worked very well. Naturally, as bandwidth increased, and wanted some type of failover/ospf dynamic routing, we added more interfaces between the routers to shuttle traffic between them if the underlying singly-connected-switch went down.

One interposing note here -- Many BRAS/BNG farms all single-end into a single switch, since your traffic is always IN+OUT, so it makes sense to reuse the same interface (usually a 10G LAG/LACP Group) in order to get better efficiencies of your links. Whats nice here, is that if the BNG fails (and you have other BNGs), then both the northbound and southbound traffic both fail, and subscribers simply re-temrinate to another BNG/BRAS. (via L2TP failover, PPPoE reauth after PPP-LCP fails, etc..)

Basically, the effect here is you've "fate shared" both the Internet and subscriber-facing sides of your router... this alleviates the problem whereas your southbound (PPPoE/L2TP/Subscruber DHCP facing interfaces) somehow "stay up", but the Northbound (Internet/Core facing/BGP/OSPF/ISIS) routing interfaces go down. That situation results in people staying connected to your BNG farm, yet have no 'real' internet connectivity. (L2TP/PPPoE is "up", but once the subscriber is terminated on the BNG, they can't "get anywhere useful". Works well.

- CK.



On 29/04/2015, at 9:44 AM, James Mcintosh <james.mcintosh at rocketmail.com> wrote:

> Hi Noggers,
> 
> Are any of you out there running "router on a stick" in your production environments?
> 
> Traditionally this was only set up in lab/test environments but given how expensive 10Gbps+ adapters are from some vendors, and that additional adapter capability often forces you up to their next most expensive router models is there any reason not to run it in production?
> 
> Most ISP's already run hundreds or even thousands of sub-interfaces per physical interface so is there any tangible downside to to just using a single physical interface for all the in/out connectivity to your router?

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ausnog.net/pipermail/ausnog/attachments/20150430/dc8ad2f0/attachment.html>