Hi Dave, re: SONET stats, next gen alternatives, etc.
At first I began replying to you assuming that the author was
writing from the perspective of the enterprise user organization.
Then I re-read Raj Shanmugaraj's synopsis after a spell, again, and
realized he was viewing the matter from a next gen fiber network
carrier's perspective. Oh, well.. some triage, and some
marrying of the two perspectives in the following unified reply.
--------------
From the article, you cite:
re<< Transparent LAN service will require the evenly spread
model of bandwidth distribution.>>
Then you note, >>Is this really true, or just a "taking the easy way
out" view of what really is a far more complex problem than can be
described by a linear/uniform distribution?<<
He's being overly general here. But before I address this point, let's
see what he had to say before that. He stated:
"Next-generation carriers build their networks as they acquire
customers on a "Just-in-Time" basis. The result is that metropolitan
fiber topologies in new carriers tend toward natural meshes. The
result is the ring topology that has become the norm under SONET
is an impediment to the " Just-in-Time" business models needed by
next-generation carriers."
Some timely and good ideas, but he gets ahead of himself and
doesn't explain what layer of infrastructure is already in place. Is he
talking about the physical pulling of cable here? Or, simply the
addition of virtual links onto paths that are already established?
Surely, as most fiber carriers of any appreciable size will attest,
when they decide to penetrate a market one of the first things they
tend to is securing rights of way and pre-laying cable, or pre- conduiting,
in circular (rings, in effect) physical routes. These need not necessarily
be topologically ring based, such as SONET would imply, but they
are almost always topographically circular in nature, or perimeterized
in an interlocking fashion, so as to leverage adjacencies and
maximize the yield per route.
Whether intentionally, or otherwise, this will also (almost invariably)
aid in the fail-safeness of future topological options and schemes, as well.
There is no such thing as point and click in this game, where creating
new physical routes is concerned, and a bunch of homework and planning
needs to be done in order to cope with franchise issues, not to mention
coordinating digging and construction schedules with other operators.
So, I would dispel the JIT argument if he is truly talking about how
to approach a market from scratch. Unless, again, he is referring to
the creation of higher layer customer links onto physical paths which
already exist.
----
Getting back to your question, since he is taking LANs into
account, as well he should, then I can't dismiss the realities of the
user's LAN environment. I have to ask, in turn, what kinds of
networking topography(ies) or domains is he matching up, or
referring to?
-WAN/Internet
-MAN/CAN - Metro and Campus Area Networks
-carrier provided loop section, or campus links
-building entrance facilities
-LAN concentration/aggregation points
-riser backbones
-distribution closets
-horizontal links to end points
Those are only the physically defined domains of interest, and they
do not address the various networking topologies and protocols
that are possible.
The end user's in-building riser bandwidth (i.e., their backbone
fabric) should be so plentiful that enough 'head room' exists within it
to absorb just about anything that user traffic presents when both
end points which are communicating are on the same LAN.
If not, then simply sink some more capital on a one-time basis into
more parallel riser backbone or campus backbone pipes, or
distribution links to the end points. This is a relatively inexpensive
point solution when compared to the ongoing costs of similar
amounts of bandwidth in the greater WAN.
LANs/CANs are very friendly this way, and a one time sunk cost
can get you near enough bandwidth so as to be never fully usable.
This is the major distinction between local and wide area networking
which has historically, up until recently, been the cause of the
"great divide" between LANs and WANs.
On the local loop it gets proportionately tighter if you are using T1,
T3, or OCn, etc. tariffed services (but less so with metro dwdm,
one would assume) and in the WAN segment, tighter still.
If the WAN, in fact, is the Internet (as it is increasingly being
referred to these days, even by networkologists), then the costs of
Internet port sizing will come into play and govern just how
evenly spread, end to end, bandwidth could be. Not to
mention, I should add, the vagaries of b-w supplies at various
points along the 'net, itself.
I am referring to the T1/T3/OC3/etc. port costs, which for T3 ports
on the 'net, say, could be as high as $50,000 to $65,000 per month, or
more, depending on the SP and features agreed on.
That's an incremental annualized expense of some ~$700,000 per
major location plus the cost of loop charges (another $40,000/annum
for T3s, on average, if close proximity exists between the user and
the ISP), per major geographic location. Lower amounts would apply
to branch offices, and smaller locations, of course, since their port
sizes would be correspondingly less, as well.
If the WAN is a private enterprise network, then ditto in spades, as
the economics of a traditional private WAN will restrict the port
sizes even further, since the same economies of scale and statistical
advantages do not exist in this case.
How "evenly spread" across all of these domains, then, becomes a
spending decision for the user to determine on the basis of
cost-benefit.
One thing is for sure, tho, and that is that the costs will not be
evenly spread, despite shrinking b-w costs and the new availability
of fiber/lambdas in some still privileged areas where fiber carriers
dare to buck the incumbents and their regs.
The proposition offered by the author is an ideal one, but IMO, too
broad to be taken as a single rule of thumb for most practical
situations which are broad based, still.
----
When GbE from the enterprise building meets a tariffed WAN or
local loop, there is almost always a step-down/step-up process
taking place through store and forward bridging as protocols are
converted from IEEE to ANSI T (although, wire speed cut through in
certain cases when the carrier permits same, or where metro dwdm is
employed).
Which means that where these dissimilar domains meet a bottleneck
potential exists requiring buffering and the ability to do flow control.
Alternately, dropped packets, if adequate traffic management
measures don't exist, or if buffers are not sized properly.
"In fact, the statement in the paragraph that follows the above:
<<The next generation carrier environment will demand an
inter-office transmission system that is a more agile manager of
bandwidth...>>
seems to imply a non-linear model.
Agreed, it seems he is taking a more practical approach at that point.
re:<<About 70 percent of the SONET rings operate at
OC12-or-less line rates. >>
Does that seem accurate to you?
Yes, thus far, I would think, but I don't have any accurate statistics
on this. Some reasons for this may not be obvious, and there may be
an offsetting argument to his observation.
One must consider that the next step up from an OC12 is OC48, a
non-trivial step up for some carriers who have legacy 12's in place.
Also, there is what I call a Times Four pattern at play here: OC3,
OC12, OC48, OC192, and now an emerging OC768. Times Four. This
doesn't quite fit the "powers of ten" definition, but the same
principles of gain are at work.
The offset? When some systems are brought up as OC12, they are
actually early add ons to infrastructure occupying the shelf space,
and a single STM module (OC12), in an OC48 system. Again, these
may be during periods of early growth. Through modular expansion,
these 12's can be incremented, at will, to the next b-w requirement
through the addition of optical line card modules.
Also, multiple [up to four] OC12s can be derived from the same
OC48 add drop mux, or ADM, which makes his observations kind of
moot in this respect, when you think about it. How many different
ways do you want to slice an apple?
<<Bandwidth scalability and the ability to overlay existing
SONET rings can be accomplished through the use of DWDM >>
Does this provide the fine level of granularity the author speaks
of?..
I think that this is a relative assessment, hence a subjective one. For
a carrier, yes, these increments in b-w might be considered "fine"
ones, but for an enterprise who must pay the freight across vast
distances, they might be seen as monumentally coarse.
"It seems in one sense the author is saying the SONET ring
topology is just "getting in the way and should be dropped rather
than band-aided into oblivion," and in another sense the author
says next generation carriers need to consider compatibility as a
foremost concern:"
<<This must also work in rings for the purpose of overlaying
existing SONET fiber topology and for service providers that are
more acclimated to ring operation>>
"I'm not real sure what he's trying to say..."
Hard to tell, but I would bet that a great part of this probably has to
do with what he perceives (and is probably correct in) as work
habits, legacy thinking, all the things that are on the side of new
startups who do not carry the baggage on their backs that the
incumbents do. Also, the embedded infrastructures, both
administrative and operational, which established carriers have in
place need to be leveraged and fully used in order to maximize
investment potentials (so conventional wisdom dictates, in any
event), which newer frameworks of fiber-based networking are not
fully conducive to.
Next stop, self healing and failover schemes. Later.
Regards, Frank Coluccio |
