Muni fiber: L1 or L2?

Scott Helms khelms at zcorum.com
Thu Feb 7 15:18:58 UTC 2013


On Thu, Feb 7, 2013 at 3:51 AM, Mikael Abrahamsson <swmike at swm.pp.se> wrote:

> On Wed, 6 Feb 2013, Scott Helms wrote:
>
>  The cost difference in a single interface card to carry an OC-3/12 isn't
>> significantly more than a Gig-E card.  Now, as I said there is no advantage
>> to doing ATM, but the real cost savings in a single interface are not
>> significant.
>>
>
> There has always been a substantial price difference for ATM/POS compared
> to ethernet.
>

Yes this is true, which is why I specifically limited the scope to a single
ATM interface.

>
> But when designing ETTH networks, the cost saving is in the use of very
> simple devices. L2/L3 switches all the way. No tunneling, no fancy
> encap/decap Q-in-Q etc. Enough intelligence to do the BCP38 stuff to
> prevent spoofing, MitM-attacks, nothing more, but still deliver needed
> services over unicast and multicast.
>
> So as soon as the design contains any of the words L2TP, PPPoE/A, ATM,
> POS, OC-whatever, xPON or anything like it, you're incurring unneccessary
> cost, especially for high bw services.
>

That really depends on how the technology is used, what is already in place
especially on the WAN side, and what OSS the operator already has in place.
 Now, in general for greenfield builds I'd agree except for PON, which is
in many cases cheaper than an Ethernet build.  Cost of the physical pieces
of the network are only one part of the cost of owning and running a
network and over the long run its actually one of the smaller pieces.  Now,
I wouldn't build a PPPoE based network today UNLESS there were significant
reasons that it would be be cheaper for that operator.  The same is true of
ATM, but I'll give you a concrete example of why it sometimes makes sense.
 In areas (and this is usually a rural challenge) there are a limited
number of operators you can buy WAN connectivity from as a local ISP
yourself.  I have customers in Montana and Wyoming especially that have
this challenge where they can either choose to pay for an ATM capable OC12
(622 mbps minus overhead) for a given price per month or a Gig-E connection
for nearly twice the amount of MRC.  In that case it makes much more sense
to pay a 5-6 thousand more for the ATM interface once than to pay ~$1,500
per month more.  This also takes into consideration that their current
bandwidth requirements are around 300 mbps.



>
> The most inexpensive device to L3-terminate 10GE worth of traffic from a
> few thousand customers is in the few thousand dollar range, what's the cost
> if you want to do the same using L2TP or PPPoE ? What about ATM? I don't
> even know if ATM on OC192/STM64 is even widely available. My guess is
> anyhow that you're not looking at a device that costs at least 5-10x the
> cost.
>

There are not generally available OC192 SAR engines.  At the 10 Gig scale
its certainly true that you'll have challenges.


>
> Designing a fiber plant very much like the traditional copper plant, ie
> aggregating thousands of households in a single pop, and letting "anyone"
> terminate that fiber, is a very future proof and scalable approach. The
> fiber can be lit up using any technology (active p-t-p ethernet, or PON, or
> whatever is desired), this doesn't have to be chosen at time of actually
> drawing the fiber. Yes, it's a high initial cost but I firmly believe that
> over tens of years of lifetime of the fiber, this cost is lower than other
> solutions.


That has not been demonstrated in the market.  There are lots of people who
say this, generally they're involved in building fiber plants, but in the
US and Canada I've not seen a single report of an actual network where this
was true.  Do you have any documentation to this effect?  I will also
acknowledge that we don't have a large sample size in the US of plants
built this way.


>
>
> --
> Mikael Abrahamsson    email: swmike at swm.pp.se
>
>


-- 
Scott Helms
Vice President of Technology
ZCorum
(678) 507-5000
--------------------------------
http://twitter.com/kscotthelms
--------------------------------



More information about the NANOG mailing list