[c-nsp] Downsides of combining P and PE functions into a single box

[Mark Tinka]

On Thursday, October 20, 2011 05:32:40 PM Jared Mauch wrote:

> Somewhat similar to the intro to the movie "The Fifth
> Element" where they ask "what have you got that's bigger
> than x" and the response is nothing. When your customers
> have n*10g access, where do you place them? When they
> have n*100ge where do those go?

Being a relatively younger network, we have the benefit of 
not having too much legacy kit running around. It also means 
we're building at a time when vendors are reasonably putting 
out dense, modern kit. Right place, right time.

I'll admit that this is a harder problem to solve unless 
you're looking at current generation Ethernet edge boxes. 
But then again, if you aren't looking at these, chances are 
you don't have this problem anyway.

> It is certainly not ideal to place them on an edge device
> then buy another pair or two to just uplink/passthrough
> to the core to passthrough to a peering edge.

The core will have been there long before you customers 
start ordering 10Gbps circuits. Provided you have the right 
platform in the core, scaling up will neither be an issue 
nor a problem when you have downstreams buying dozens of 
10Gbps circuits from you.

For folks whose core is aggregating non-Ethernet circuits 
too, it's not trivial to throw such links into a P/PE device 
optimized for Ethernet. This may not necessarily justify 
having an independent core layer, but if you're big enough, 
maybe it is, among other reasons.

There is a reason Cisco and Juniper built multi-chassis 
systems, even if I don't necessarily subscribe to the idea 
even today :-).

> We may be networking but just on different scales. When
> you have 50+ 10g circuits on a path for traffic it makes
> you look at the costs differently than when you have one
> or two.

Agree, and working out the cost is not just a function of 
whether you can do more with less. There's also avenues such 
as special discount levels or relationships with vendors to 
exploit. These are the bits, unfortunately, that won't make 
it into the monthly industry journal :-).

> It may also be the market in your part of the world can
> bear these higher costs.

Like I said, cost isn't necessarily the only hand you have 
to play. At these networking levels, it's more than a "Hi, 
bye" between you and the vendors. Every ISP networking at 
these kinds of bandwidth levels knows that.

> I've been hearing of sub-$1/meg
> contracts.

Me too.

> Not sure what your price per meg is with all
> those layers.

It's competitive - we don't rely only on the kit to pay for 
itself :-). But I digress...

> I'm seeing one player come back at nearly
> $100/meg in the us. They obviously will not be selected
> for transit.

US$100/Mbps/month? In the U.S.? Is that via satellite, 'coz 
that ain't right!

Mark.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part.
URL: <https://puck.nether.net/pipermail/cisco-nsp/attachments/20111020/6080fa22/attachment.pgp>