[c-nsp] Real life and worst-case performance of Cisco and Juniper?

[Mark Tinka]

On Friday 27 February 2009 10:58:32 pm E. Versaevel wrote:

> We have been in a similar situation last year.
> Originaly we had 7206 VXR NPE-G1's as both access routers
> and core routers, however maintaining a full bgp table on
> the core routers became a bit to much for the 7206's...

We try not to get too religious about it, but forwarding on 
labels in the core is a good way to keep BGP (at least for 
IPv4) out, i.e., MPLS.

Again, I'm not recommending that MPLS is a solution to any 
kind of problem. I'm just saying that because of it, we've 
kept a couple of NPE-G2's and NPE-G1's running in smaller 
PoP's as core routers since we don't have the CPU and memory 
dealing with BGP. I may add that we've seen an NPE-G2 push 
as much as 950Mbps, aggregated, in this role.

> In the end we went for the 7606/RSP720 as a core
> router(s) and moved the `core` 7206's to the access
> network .

The problem with hardware memory, e.g., CAM, is that bad 
things start to happen when it's used up. The RSP720 has 
sufficient hardware memory (anyone moving from a SUP2/32 to 
a SUP/RSP720 knows this, painfully) for a long time to come, 
but keep that in mind as the Internet routing table 
continues to grow.

The advantage with software routers is that they will hold a 
ton of routes in memory, and use them for forwarding as 
well. Too bad their FIB is handled in software though :-(. 
That said, they do make great route reflectors. We just need 
to pester vendors to ensure larger memory sizes are 
supported in software routers for this purpose. We're not 
keen on using hardware platforms as route reflectors... but 
I digress.

> First of all because our entire network is Cisco which
> means our support staff wouldn't have to learn a new
> router OS.

We believe in a little diversity, but not too much - it's 
good for business and competition :-).

> Next to that we've tested a M7i which
> performed flawlessly and I personally like the JunOS
> config style (tested mpls/ldp/ospf/bgp only, no ip
> vpn/virtual routers/BBA etc just plain packet
> forwarding). However we have a policy that every access
> device needs 2 direct connections to the core routers...

Which is good.

> which means we need quite a few interfaces.

The M7i is fairly small, but with redundant connectivity 
into the core, you can get a couple of Gig-E interfaces in 
there to trunk to customers over 802.1Q.

Depending on the number, you might even have room to 
terminate a couple of leased lines too. This will vary, of 
course, as each network has its own requirements.

> Interface
> pricing on the Junipers is ridicules imho, 18k$ for a
> single gigabit ethernet connection, for a fraction of
> that you would get a 24x SFP module for the 7600
> series...

Vote with your wallet - get your account team (with 
whichever vendor) so see "cost/benefit" your way :-).

> (add to that that we have a few STM-1
> connections which are even more expensive)

I think STM-1, or generally, sub-Gig-E ports on the 7600 
just don't make any sense. A 7200, ASR1000 or M7i might make 
more sense here.

But since you moved your ex-7200 core routers into the edge, 
this isn't a problem you're facing.

Cheers,

Mark.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: This is a digitally signed message part.
URL: <https://puck.nether.net/pipermail/cisco-nsp/attachments/20090227/9171c4ba/attachment.bin>