[c-nsp] data center/mpls/vpls

[Andrew Miehs]

On Tue, Apr 16, 2013 at 12:19 AM, Aaron <aaron1 at gvtc.com> wrote:

> Q1 - Why do you want MPLS on the top of rack switch?
> A1 - L2VPN's...L3VPN's...  Yes, 10 feet away is my Cisco ASR9006 MPLS PE
> edge box, so yes I could simply do 802.1q from tor switch to that 9k and
> then jump into mpls LxVPN's there.....perhaps this would be a way for me to
> keep the cost down on the TOR switch.
>

Cost will be your biggest issue - 5 with fully capable L3/ MPLS switches -
with 5x 10G interfaces.
The cheapest Cisco that can do this would be a 6500. Not too sure about the
HP or Juniper boxes as they didn't have large enough routing tables for use
to keep looking at them. And at $75K you will just be able to afford one of
these.


> Q2 - Do you need 1G, or 10G interfaces, and how many of them?  Why do you
> need 5? Would something like a pair of 5Ks and FEXs not do what you are
> after?
> A2 - 10G and about 5 of them per switch.  (2) Data Centers, with (2)
> cabinets each.  So (4) TOR switches....and we spare one of everything we
> have for quick disaster recovery incase of outage.... so 5
>

You should be able to get close to 5 x Nexus 5K with only the basic L2
licenses for that money.


Q3 - IIRC, the HP 5900s only support 16K routes - which could be a problem
> on an MPLS network.
> A3 - My MPLS network is just that, mine.  I am a small ISP.  I own the
> whole
> mpls network....currently 209 routes....it will grow, but not to thousands.
> And if I do ospf area's and route summarization at area boundaries down the
> road, I could probably still keep route table small.
>
> 3600#sh ip route summ | in Route|Total
> Route Source    Networks    Subnets     Replicates  Overhead    Memory
> (bytes)
> Total           4           209         0           12900       47784
>

I would imagine that this number would sky-rocket if you ever started
providing VPNs for your corporate customer, as many of these will
not summarise routes.
My experience is Campus MPLS networks so my experience will be different to
that of ISPs. Also don't forget to double this number to include IPv6
eventually.



> Q3.1 - Can't you just trunk the various VLANs back to a PE/ pair of PEs?
> A3.1 - yes, I could.... should I?  Would you/you all prefer mpls to the ToR
> edge to jump into and out of mpls LxVPN's right there?  Or is it not worth
> paying for?  Seems reminiscent of ATM....and the powers realized by getting
> atm closer and closer to the edge...but then always wondering, perhaps I
> could simply just vlan into the atm lane elan/pvc/pvp and be done with it.
> Hmmmm
>

I don't know of any box capable of 5x 10G with MPLS which costs $75K / 5
...
I would rather trunk back to your ASR9000 and home directly into the
correct MPLS VPN compared to VRF lite on a L3 switch, and THEN back to the
AS9000...
Saves you a L3 hop, and a lot of mucking about with VRF lite - for no gain
that I can see....



> Q4 - I think the QFX will be too expensive for your budget. IIRC, the
> EX4500s also have the not enough routes supported issue....
> A4 - 209 routes currently.  I see EX4550 does mpls.  Don't know how much it
> costs though...gonna talk to Juniper this morning in webex.  Is ex4550 a
> good box?
>

I haven't played with any of the Juniper kit so I don't know how reliable
it now is.

Good luck.


Andrew