RE: [EFM] RE: Relative OSP Costs of PON vs. P2P
You have raised many interesting points, I will try to walk through them one
I would first like to say that it is pointless pitting PON vs. P2P
endlessly. They both have the merits and a right to live, EPON has benefits
in high density deployments, while P2P in dedicated high-bandwidth
prelude) I do not believe a POP will be connected to the backbone through 16
ducts * 4 inner ducts * 800 fibers * terabits/s. This is not an economic
statement, and is the opposite of the switch/concentrator architecture of
routers existing today.
Although the fiber might exist, POPs (DSLAMs) today are mostly fed through
155Mbit pipes that will grow with technology and demand to 10G and 40G later
on. Some WDM will also be present, but not enough to justify only small
1) APON systems today, as well as the EPON spectrum planned, leave the
1550nm window open for future services. Systems built using this concept
allow upgradeability without any equipment replacement of non-affected
customers. Hybrid systems, in which the PON splitter sits in the CO and not
in the field exhibit even more flexibility.
2) No single equipment terminates an entire POP. Single termination is
possible only in scenarios where small numbers of subscribers exist. In
truth, access networks from the port devices to the metro uplink are
comprised of many switching elements in a Benes, Banyan, Clos or any other
type switching/concentrating network topology. Full mesh connectivity does
not scale well. Therefore, any network, be it with PON front-ends or P2P
front end, does not use the mythical 1Tbit non-blocking switch element.
Furthermore, all access devices utilizing these networks have the need to
reserve bandwidth along the entire switching path to setup an SLA, thus they
all have software of comparable complexity.
I do not know of single box solutions for terminating 1000 fibers, however a
100 port PON device is quite feasible, terminating 3200 subscribers.
Therefore access networks, when using P2P technology, will usually have
multiple boxes with backend routers before leaving the POP. Probably, PON
POPs will have much less equipment and thus be easier to manage...
3) The statement regarding large numbers and switching efficiency is true in
broad terms. However, due to the network aspect of the system architecture,
and the fact that Ethernet switching elements utilize spanning trees in one
form or another, only a single path connecting two points in the access
network exists. This means that the switching capacity is not 1T, but more
like 1~10G depending on the switching elements passed through, due to the
inability to utilize unused portions of the switch fabric. A side-effect is
the influence of two otherwise unrelated ports on each other. This is
solvable when routing is performed, a function that leaves the Ethernet
domain, and is much much more expensive. The conclusion is that the
concentration is not achieved in a single stage rather in multiple stages,
EPONs simply exhibit a first stage concentrating factor of 1:32 (other rates
down to 1:1 are possible).
4) In all respects, due to that fact that P2P solutions require more
equipment at the POP than PON, the operational costs are higher for P2P...
In short when analyzing an access network, the only difference between EPON
and P2P is the fact that the front end element of the switching network is
limited to 1G for PON (10G in the future), while a similar element for P2P
might be limited to 10G now. However most of the performance is dictated by
the architecture of the second and third layers of the switching network and
the bandwidth of the interconnects, all of which are limited in performance
From: pasi.vaananen@xxxxxxxxx [mailto:pasi.vaananen@xxxxxxxxx]
Sent: Tuesday, June 12, 2001 1:36 PM
To: ariel.maislos@xxxxxxxxxxx; pat.kelly@xxxxxxxxx;
Subject: RE: [EFM] RE: Relative OSP Costs of PON vs. P2P
>I believe there is a misconception we are locked in regarding P2P networks.
>The thread so far shows a concentration factor of 1:3 (32x1G in, 10G out).
>This is a very hefty number that does not scale well in medium to large
>A neighborhood switch serving 1000 customers will have a 300G uplink! While
>a typical CO with 10000 customers will have a 3T uplink. Backbones can't
>handle so much traffic, especially when it is not necessary. A large POP
>will have a 10G uplink, and 40G in the far future, this is a concentration
>factor of 1:1000.
I believe there is a miscalculation here - your neighborhood switch @ CO
would have 300M of bandwith per subscriber. With the 1G PON, 1:32 split
ratio and no any oversubscription, the number is a bit over 30G - and
according to above 1:3 ratio this would be about 10G, which is a very
reasonable number cost wise.
Backbones can handle the capacity, single fiber can carry over terabit/s
NOW, with the equipment commercially available from multiple sources
although the cost is still high. Typical backbone has ~16 ducts, with
four inner ducts a piece, and single innerduct can handle ~800 fibers
(conservatively). Only thing keeping us from low concentration ratios
at a moment is cost and lack of need for such high sustained bandwiths
due to bursty nature of traffic.
The concentration discussion in pt-pt vs. PON context is missing a few
1.) pt-pt systems have virtually unlimited capacity from OSP plant
perspective, and the per subscriber bandwith can be set anywhere system
operator wants, without affecting any other users in the system. Thus
the future upgrades are much easier than in PON (no modifications to
all subscribers required for speed upgrade, no downtime or plant
modifications for WDM upgrades & no differences in customer equipment
for the different wavelengths like in WDM PON). When the concentration
is done in the switch in pt-pt system, the only things that need
to be changed for any bandwith point are affected customers equipment,
and associated CO equipment.
2.) the oversubscription in the access portion (in PONs) requires
careful resource management (especially if anything better than
best effort is required). This adds to cost & management overhead.
3.) aggregates of traffic from large user populations (1000-10000:s
vs 10:s) behave better with respect to burstiness - this suggests that
it might actually help to have the concentration done in switch.
4.) discussion is omitting the operational cost differences, where
pt-pt would have advantage over PONs, although these are hard to
quantify them are important part of the system life time cost.
>Hence, although theoretically EPONs provide lower bandwidths
>due to sharing of the medium, there is no difference what so
>ever in practice between P2P and EPON networks.
This is not true, in P2P the operator can match the access speed
to any point, with the backbone concentration ratio (done in
switch) that satisfies the required QoS constraints with respects
to loss rates, delay, etc. PONs cannot simply do this for higher
bandwith subscribers (or essentially "degrade" to pt-pt system).
Peoples are really doing this in practice for various types of
users and SLAs right now.
PONs definitely have their place, which is for the operators who have
some fiber already installed (most), and in situations where the
services do not (ever) require the high bandwith available in pt-pt