[AusNOG] VPLS Router Support

[Ben Dale]

On 20/04/2010, at 10:54 AM, Ryan Benson wrote:

> Yes the topology is more of a daisy chain than a mesh layout with a
> few resilient links thrown in here and there for good measure. Using
> STP between Metro sites with a secondary link is where I am feeling
> the pain at the moment.
> 
> Most documents I have seem to have the standard Site A Company A, --->
> Layer 2 ---> Site B Company B type scenario, where I will be pulling
> Vlans up to reseller interconnects and letting them do what they like
> over it. This is where I am unsure if MPLS will be sufficient to do
> the job.
> 

So at each site you're adding unique VLANs to an ever-growing trunk and terminating them all centrally?  Or are you taking some VLANs a variety of sites?
Martini circuits will fail-over along redundant paths quite nicely using fast-re-route, but they'll only work point-to-point.

> Based on the fact the network is more of a daisy chain than a mesh,
> would there be any advantages between going with LDP or BGP? (Not to
> say that in the future it wont be fully meshed)

It's really a scaling question - LDP is much simpler to configure initially, but as you add sites, BGP-based auto-discovery really comes into it's own.  BGP-VPLS also has a nice HA mechanism for multi-homing to diverse PEs that doesn't require spanning-tree, whereas LDP will require an underlying spanning-tree to avoid this causing a loop.

The downside of BGP-based used to be vendor support in that only Juniper supported it, but a quick google suggests that this has changed: Cisco and Mikrotik seem to be capable now.  If your number of sites isn't too large and fairly static, the question becomes: do you have the in-house skills to run BGP internally?  If not, go with LDP.

> Thanks again,
> 
> - R
> 
> On 19 April 2010 17:59, Ben Dale <bdale at comlinx.com.au> wrote:
>> Hi Ryan,
>> 
>>> At the present moment I am trying to find out how best to transport
>>> customer Vlans across my network, if this can achieved with MPLS with
>>> some degree of success, or if best to go all the way and move into a
>>> VPLS based solution. Spanning tree is my worse nightmare at the
>>> moment!
>>> 
>>> Cheers, - R
>> 
>> It will depend a bit your topology - if you just need to take a bunch of site-specific VLANS from different locations and bring them all back to a core site, you could achieve this fairly simply using EoMPLS/martini tunnels.  The nice thing about martini tunnels is no mac-learning - just in one end, out the other.  Simple.  You can terminate their VLANs on your interface and drop each one into a separate tunnel, or pass whatever tags they throw at you straight through in a single one.  If they run spanning-tree, it will just pass right on through (the tunnel) without touching your network.
>> 
>> If you need the same L2 domain to appear at more than 2 sites, then VPLS is a better way to go.  You will still have to learn MACs and it's just as susceptible to L2 issues caused by loops joined outside of your PE, but if your customer runs their own spanning-tree, then this isn't your problem (or your gear's).   If you can, police hard to mitigate fall-out from any broadcast storms that do happen (if nothing else, it will shield other customers) and you should be fine.  Be aware that multicast in a VPLS environment is also flooded though.
>> 
>> Good luck!
>> (and death to spanning tree - bring on TRILL)
>> 
>> 
>>> On 19 April 2010 10:25, Ryan Benson <mail at ryanbenson.net> wrote:
>>>> Hi All,
>>>> 
>>>> Has anyone got any recommendations on good kit that supports VPLS? I
>>>> can see I will have to use 7600's and upwards to support this on CISCO
>>>> but does anyone have any views on any other vendors, Huawei, Juniper,
>>>> Ericsson...?
>>>> 
>>>> Cheers
>>>> 
>>>> -R
>>>> 
>>> _______________________________________________
>>> AusNOG mailing list
>>> AusNOG at lists.ausnog.net
>>> http://lists.ausnog.net/mailman/listinfo/ausnog
>>> 
>> 
>> 
>