RE: [EFM] Active Architectures
Hugh, thanks for the comments. My comments in-line prefaced with [dh].
From: Hugh Barrass [mailto:hbarrass@xxxxxxxxx]
Sent: Tuesday, August 21, 2001 7:57 AM
To: Horne, David M
Cc: 'Sukanta ganguly'; ramu_raskan@xxxxxxxxxxxxx;
Subject: Re: [EFM] Active Architectures
That's a fascinating idea - I like the way you think laterally. A few
Firstly, I don't see how this fits into the IEEE 802 model of LAN/MAN
[dh]I may not be getting your comment, but if you were to collapse the
distribution fiber from PoP to node down to zero so that the switch was in
the PoP, it becomes a P2P proxy at the topology level.
The concept of 3+ unidirectional links is pretty scary. Why would
you not go the next step and make it a simple curbside switch with a GE
and copper edge ports?
[dh]I guess the scariness would have to be tempered with the cost savings,
performance advantage, and time to market advantage. Fear dissipates with
familiarity also, and right now this is appears like a new concept. With a
little time pondering it, it is really not so new. On the second part of
your comment, it just couldn't compete in cost compared to the 850nm
100BaseSX solution. Oh, copper is what you are asking about. That is a
possibility and Francois mentioned extension to that in an earlier message.
To me that is only a practical alternative if conduit is used, and fiber can
be installed easily at a later date. New copper would have to be laid since
current in-place copper is not laid at this node-size granularity. No
matter which of the half-dozen bandwidth consumption trends you look at, the
life of a copper solution that is in the tens of megabits is probably less
than its depreciation life. We're all in trouble if these bandwidth growth
trends don't continue so I consider this to be in the category of "We hold
these truths to be self-evident."
Secondly, it is not very flexible in topology. An EPON can be configured
single 1:16 split point, as a tree with (say) 4 layers of 1:2 / 2:4 / 4:8 /
or even as a broken ring with 16 serial 1:2. It really depends on whose
way you are deploying your service into.
[dh]Yes, it would get a big red "X" in our architecture trade study matrix,
as far as morphing into a distributed architecture like PON could do. This
would be an important consideration for a sparse rural deployment. It's
definitely not architectural nirvana, but that does not exist with any of
the architectures. I think the scenario it shines in is dense to
medium-density urban deployments. These will make up the majority of the
deployed endpoints so it does have broad applicability.
Lastly, after thinking about this, it could fit into the definition of EPONs
they stand. If the upstream (and the downstream if you want) is simply an
electrical-optical translation then the overall behavior will be the same as
EPON. I haven't thought deeply about it, but I would guess that a smart
vendor could make such a device interoperate with standard EPON head ends
standard EoVDSL CPEs.This could be a great idea - have you patented it?
[dh] Nope, no patent, I didn't feel it was all that unique.
I think the bottom line is that (according to my first reading) this idea
doesn't impact the work that we are doing on the 3 media as defined here.
[dh]Hard to say at this point, since things are still in flux except at the
highest levels. If extending the distance is important (between ONU and
node) it would require a modification to 100BaseSX for 850nm VCSELs, as well
as OA&M additions to 100BaseSX. This would increase cost, but still be well
below a 13x0 >10km solution. For the ONU, to optimize it functionally it
would need to strip out 1000BaseLX receive and 100BaseSX transmit and
combine them into a unified function that becomes a new media definition.
This isn't mandatory though; it just results in idle silicon functions if it
isn't combined into 1 (while also integrating in OA&M). So I think it would
be a new media type, but not a very disruptive one.
"Horne, David M" wrote:
> On the topic of active architectures, one that seems attractive in
> comparison to PON for some deployment scenarios could be called a
> "half-PON." Below I give a proposal for this. Comments appreciated.
> The half-PON is a hybrid active/passive architecture that preserves the
> attractive features of PON and P2P while eliminating many of the
> unattractive features of each.
> Downstream, it would be the same as the PON downstream (15x0nm GbE from
> to splitter; dedicated fibers from splitter to each of 16 (for example)
> And upstream would be in two pieces:
> The first would be from the ONUs to the active node (hardened Ethernet
> switch in node; node also contains the downstream splitter) and has a
> dedicated fiber per ONU, each to a separate port on the Ethernet switch in
> the node enclosure (more details in a second).
> The second part of the upstream would be a single-fiber 13x0nm GbE between
> the node and the PoP (wavelength-muxed into the same fiber as the
> so only one fiber between node and PoP).
> ADVANTAGES AND SAVINGS:
> The key savings come from the use of low power 850nm transmitters between
> the ONUs and the switch, using 100BaseSX (TIA/EIA 785) as the baseline. So
> instead of 16 high power (in comparison) 13x0 laser sources (one per ONU)
> for PON upstream, each capable of >10km reach thru a 16:1 splitter, you'd
> have 16 low power, low cost 850nm sources (one per ONU) that only have to
> reach about 300m to the active node, and 1 low power 13x0 (low because no
> splitter to go thru) source from the switch back up to the PoP. There are
> couple options on the fiber between the node and the ONU but I'll leave
> discussion for later.
> Perhaps more importantly though, in comparison to PON you lose the need
> developing and agreeing upon a TDMA protocol, a ranging protocol, a sync
> protocol, and a contention protocol for requests. You also get uniform
> transmit power back to the PoP since there is only 1 source rather than 16
> time-multiplexed from different ONUs at different distances, and no
> about run-away transmitter operation bringing down the node. You also get
> management-capable intermediate network point which will allow greater
> diagnostic, provisioning, and demarcation capabilities.
> So, ONUs would be significantly cheaper. On this point there is no
> For reference, a 100BaseFX (13x0 laser, 15km reach) to 100BaseTX media
> converter goes for about 3-4x the cost of a 100BaseSX (850nm, 300m) to
> media converter. Essentially, this is the ONU's function for any of the
> architectures. Multiply that cost differential by 16 then subtract off a
> 10km lower power (no splitter to go thru) 1000BaseLX , then add back in
> powering costs. It seems to be an attractive option as a first order
> estimate (i.e. thousands of dollars in savings per node branch, serving 16
> ONUs), and powered nodes are a mature technology with lots of cost
> tradeoffs that can be made.
> By the way, I am not presenting this as a replacement for PON, but as an
> alternative that has not been discussed, and may be more attractive in
> certain situations. It would also be much less development effort since
> the constituent pieces have a defined Ethernet heritage, which means
> time to market. Arguably it is more future proof than PON since it starts
> out with 100Mb ports per user, which PON cannot achieve due to TDMA
> and guardbands. It also appears cheaper overall than PON, per user (at
> in the early going anyway).
> In any event it is an alternative FTTH architecture that is decoupled from
> the development time required for PON. The primary development effort
> involve adding OA&M to 100Base SX (but there are a couple other
> decisions to make). The rest of the architecture uses EFM P2P elements as
> envision it.
> Any interest?
> --Dave Horne
> -----Original Message-----
> From: Sukanta ganguly [mailto:sganguly@xxxxxxxxx]
> Sent: Sunday, August 19, 2001 9:06 AM
> To: email@example.com; firstname.lastname@example.org
> Subject: Re: [EFM] EFM Requirements
> A all Fiber architecture will be simpler in the
> overall aspect but would be a extremely expensive
> deployment. So wo should have some sort of E/O deliver
> in the last/first half mile.
> Rich, Vladimir,
> An active architecture within the field isn general
> is not a good idea, but actual scenario may digress
> from generality. Some switching is required, no doubt.
> I haven't read any deeper discussion on the switching
> complexity (Maybe I missed some of the discussions in
> the past.) I also did not understand the assumption of
> the last/first 0.5 mile being more scalable than PON.
> What kind of scalability is required at the last/first
> mile. If you are talking about 1000baseX being
> distributed to the real consumer or to then the amount
> of bandwidth that is to be switched is fairly small
> (i.e. in comparison to the 10G portion). The switching
> components as far as scaleability and complexity is
> concerned will be very different and hence, atleast
> from my point of view is a very different discussion.
> I think, and you folks may help me understand it
> better, it is not a good idea to come up with a
> generic one switching logic which can be applied to
> 100base-X/1000base-X and 10G network.
> I would be extremely interested in discussing the
> active component network on different scale of
> operation as opposed to a PON as far as cost/ease of
> deployment and maintainence, complexity etc is
> My two cents.
> (Sukanta Ganguly, An independent view)
> --- ramu <ramu_raskan@xxxxxxxxxxxxx> wrote:
> > Rich, not sure I understand the logic behind some of
> > your points, but the simple answer appears to be
> > that it requires new trenching for the new copper
> > drops, which will never fly. If I misunderstand,
> > please elaborate.
> > I don't quite get you conclusion: 'since E/O is
> > required, 1KBaseX is significantly more cost
> > effective, scalable, and simpler than PON.'
> > If E/O is required in the field it can't be PON.
> > Whether it is significantly more cost effective
> > cannot be judged without detailed designs of each.
> > Whether it is more scalable I guess depends on your
> > definition. Simpler is hard to imagine, but again
> > may depend on definition. If you could elaborate
> > your view I would be interested.
> > Since you are suggesting an active architecture with
> > new electronics in the field, I for one would
> > welcome a discussion of such a network architecture
> > that had fiber for the last thousand feet. No one
> > has addressed that at all to my recollection. None
> > of the architectues is perfect in all respects so an
> > all-fiber active architecture undoubtedly has some
> > advantages.
> > Vladimir,
> > It would seem that the most cost effective approach
> > for a 10 mile EFM
> > solution would be to use standard point-to-point
> > 1000BASE-X or 10GBASE-X
> > for the first 9.5 miles and then a 0.5 mile copper
> > tail for the
> > first/last half mile. Since E/O conversion is
> > required at the 9.5 mile
> > mark, standard 1000BASE-X or 10GBASE-X technology
> > would be significantly
> > more cost effective, scalable and simpler that PON
> > at that point. I
> > expect that there will be switching equipment
> > located at the upstream
> > (10 miles away) side, negating any benefit of a PON
> > split at that point.
> > Please tell me what's wrong with this picture?
> > P.S. I understand that this does not address the
> > rural market portrayed
> > by Frank Miller in this thread, but neither does
> > PON.
> > --
> > Best Regards,
> > Rich
> > Get 250 color business cards for FREE!
> > http://businesscards.lycos.com/vp/fastpath/
> Do You Yahoo!?
> Make international calls for as low as $.04/minute with Yahoo! Messenger