[j-nsp] MX Virtual Chassis?

[Darius Jahandarie]

On Thu, Jan 10, 2013 at 11:44 AM, Saku Ytti <saku at ytti.fi> wrote:
> On (2013-01-10 14:53 +0000), OBrien, Will wrote:
>
>> I'm curious if anyone has been using MX's in a VC config. It's supported on the new MPC blades, but supposedly not with the older DPCs.
>> I haven't done any testing yet, just minimal research.
>>
>> Why would I want to? Well, I'm after redundancy with my services blades. Specifically, MS-DPCs. I've got a pair of 480s, each with a single MS-DPC.
>> I'd be hoping to configure rsp interfaces on the VC.
>
> I firmly believe on average all of these software shared chassis solution
> offer worse availability than single box.
>
> Outage is most commonly caused by
>
> 1. operator
> 2. broken software
> 3. broken hardware

Although I agree in general (especially in the core of a network), I
don't agree so much at the server edge. Servers generally do not speak
the STP, OSPF, BFD, MPLS, etc. that would be necessary to reduce
single points of failure. The best you can do is run LACP with your
immediate uplink to protect against port or wire failures.

With multi-chassis technology you can gain something which was not
possible without it, namely the ability to protect against total
hardware failure of your direct uplink switch/router by using an
MC-LAG to your server.

One still needs to consider broken software versus broken hardware,
but it is at least no longer a question of 'highly likely broken
software' (multi-chassis tech) versus 'maybe broken software'
(standard protocols).

-- 
Darius Jahandarie