[EFM] P12 Local Operator Perspective - EFM General Requirements
EFM General Requirements
1. Overview
CTBC Telecom is looking at new technologies to deploy high speed
interactive services to its customers. We are tracking the progress of the
EFM standard because we believe that Ethernet-based architectures are
positioned to become the most practical way to deploy the kind of services
that we want to provide.
This document outlines the general requirements for the EFM technology,
focusing on its applications. In-depth technical requirements will be
discussed later, in the scope of each one of the subgroups. Some of the
requirements are not directly focused on the scope of the EFM work, but
have immediate consequences that may impact the requirement analysis.
This document reflects CTBC Telecom's understanding of the issues
surrouding EFM. We have discussed some of the issues with other carriers.
We tried to highlight some points that resulted from these discussions.
However, this document should not be taken as the view of any other
company, be it related or not to CTBC Telecom.
2. Network architecture
CTBC Telecom uses a distributed approach for its current POTS network
architecture with excellent results. The short length of the final drop to
the customer allows for lower costs and easier maintenance. We intend to
keep the same architecture for the optical access network. The typical
length of the drop is under 500 m (1500 ft). The typical last drop
concentration point is housed in cabinets deployed on the field, with 500
to 1000 accesses per cabinet. There are very few exceptions, mainly on
rural or other sparsely populated areas.
2.1. Copper x Fiber
2.1.1. CTBC Telecom's network is comprised of conventional copper lines,
many of them deployed over the past few years. The high quality of the
cable and the short loop length can be used to leverage the opportunity of
copper-based EFM technologies. However, there are still some important
areas served by older cable that may not be able to offer the quality
required for this service.
2.1.2. It is important to state that, even considering the huge installed
base of copper, CTBC Telecom is seriously considering the deployment of
all-new fiber networks to support broadband services. There are some trends
pointing in this direction. Even at his best, we feel that the current
copper technology will not have enough capacity for the growth that we are
predicting. The cost of the fiber has dropped in the past years, and it is
now very well positioned as an alternative to the copper. The final
decision depends on cost analysis.
2.1.3. It is also highly possible to make a decision to deploy different
technologies for specific parts of the network. In this sense, some level
of interoperability between EPON, EP2P and copper-based technologies is
welcome and should be encouraged given the practical limits.
2.2. EPON x EP2P
2.2.1. We recognize the benefits of the passive optical network design, as
there are no active components inside the distribution network to be
powered or managed. However, we also realize that in order to achieve these
benefits, we have to make some tradeoffs in the complexity of the edge
components, in particular on the network termination units. Also, the lack
of manageable components in the distribution network may make
troubleshooting more complicated, which may negate the advantages of the
passive design.
2.2.2. On the other, we also recognize that the advances in the technology
of the active components allows the practical deployment of reliable units
on the field, diminishing the importance of the maintenance issues in the
P2P design. The availability of low-powered, easy to manage units may bring
the issue to an even lower importance level. The practical benefits and
cost of the active units must be attractive enough to overcome the
disadvantages.
2.2.3. In short, the question of PON x P2P design has to be analyzed in a
broad sense, taking into account all the differences and inherent
advantages of each proposal.
2.3. Layer 2 x Layer 3 network model
The network must support a good Layer 2 model, in opposition to a Layer 3
design. The reason is clear: to isolate the physical network structure from
the Layer 3 network topology. Proposals that bind the provisioning of
access points to IP addresses are in effect bypassing the layer 2, and may
make the network topology much harder to manage. The practical effect of
this requirement is the need to support a solid VLAN implementation, in
opposition to the 'flat' access network model with IP-based encryption and
service selection.
3. Summary of the criteria for requirement analysis
All the items below are of high importance. The order should not be
understood as a hint of prioritization. Tradeoffs will be different for
every carrier.
3.1. Bandwidth
3.2. Support for services
3.3. Cost
3.4. Security
3.5. Interoperability
3.6. Maintainability
3.7. Scalability
3.8. Reach
3.1. Bandwidth
The bandwidth criteria will help to decide between copper and fiber
solutions. The path for high bandwidth is clearly over the optical
networks. Considering the huge installed base of copper, it only makes
sense to build a new network infrastructure if it can leverage the inherent
capacity of the fiber to provide high speed services. In particular, the
amount of bandwidth available in the downstream direction is of utmost
importance to allow the network to deliver a wide selection of content to
every customer.
3.1.1. Copper based solutions may be able to offer a more limited set of
services. This has to be counterbalanced by the cost of the solution. The
copper solutions must use the existing cable, and not ask for new copper to
be deployed, as this may make the balance change to fiber side.
3.1.2. Fiber-based solutions will be able to offer a much wider set of
services. However, it makes no sense deploy fiber-based solutions with
speeds (measured per user) close to what can be attained over the existing
copper. Other important issue is that it may take a long time to implement
all proposed services over the network, which makes the cost analysis much
harder.
3.2. Support for services
Having bandwidth is not enough if the network does not provide adequate
support for the services. Ethernet is a proven technology with a lot of
track record in this respect. Our goal is to implement an integrated access
network, offering all the traditional services over the same structure,
while allowing for the implementation of new services. The following base
services have to be supported by the network, each one with some particular
requirements:
3.2.1. Data services, with focus on IP Internet/VPN access. This service
needs a secure implementation, keeping the privacy of one customer's data
stream. There are some specific requirements for PON implementations:
a) In a PON, the same data stream travels to every customer in the
downstream direction. There shall be specific provisions against the
possibility of sniffing/snooping data packets directed to other customers.
b) In a PON, the media access control for the upstream channel needs to be
the adaptable to the bursty nature of data traffic. The specification of
the DBA (Dynamic Bandwidth Allocation) mechanism must be part of the
standard.
3.2.2. Voice. The transport of voice may use one of the basic approaches:
a) Voice over TDM. The physical implementation of the protocol provides
direct support for TDM-compatible 64 kbps channels for every voice endpoint
in the network. The TDM signal is carried independently of the main
packet-based signal (using sidebands, special encoding, or a different
wavelength).
b) Voice over packet, with three options:
b.1) Voice over TDM emulation, transparent mode. The TDM emulation layer
works directly over layer 2 Ethernet packets, delivering the equivalent of
the 64 kbps channel between the voice port on the CPE and the PSTN
interface on the network core. QoS mechanisms must be implemented over the
network to insure bounded latency and minimum packet loss for voice
packets. The interconnection with the PSTN is done through direct
connection of E1/T1 interfaces.
b.2) Voice over TDM emulation, advanced signaling. Similar to the (b.1)
approach, but the interface with the PSTN uses advanced signaling, such as:
GR303, V5.2, or similar standards.
b.3) Voice over IP, delivering the calls to a soft switch using open
standard protocols. In this case, there should be possible to map IP-level
QoS requirements to layer 2 specific QoS mechanisms, similar to the ones
required for the TDM emulation layer.
In any of the cases, support for european standards and their brazilian
variations is of utmost importance. Although this is not a direct
requirement for EFM, we believe that it may affect some of the design
decisions. For instance, the need to develop advanced, and sometimes
proprietary, signaling functions for the (b.2) approach may make it more
expensive to develop than either (a) or (b.1), which are transparent to the
signaling issues; or (b.3), which use open IP based standards. As such, we
believe that the choice of the method to carry voice calls will impact the
requirement analysis for the EFM standard.
3.2.3. Video, with the ability to efficiently deliver high-bandwidth video
content to the users. Some services, such as true Video-on-demand and
Personal Video Recorder, lead to a direct implementation over the data
network. Other services, in particular the transport of broadcast video,
have special requirements that have to be mapped on the network design.
Services based on video broadcast (or CATV-emulation service) may be
delivered in either of two ways:
a) Analog video overlay, using a WDM approach. A separate wavelength is
used to carry video in the downstream direction, allowing for the
transparent delivery of video signal.
b) Digital video over packet. This is the preferred approach. The
requirements are:
b.1) Video must be encoded using open standards, such as MPEG-2 (or MPEG-4)
video over IP. The use of proprietary codecs must be discouraged.
b.2) In consequence of the previous requirement, we have to reserve about
6 Mbps of bandwidth for a single video channel. Considering a typical
lineup of 100 channels, we will need approximately 600 Mbps to broadcast
them all. This leads to the requirement for at least 1 Gbps aggregated
bandwidth in the downstream.
b.3) A lot of bandwidth can be saved by selectively transmitting only a few
channels at a time, depending on the customer selection. However, this
depends on whether the control is done at layer 3 or layer 2. Layer 3
control mechanisms, such as IP multicast, must be mapped to equivalent
layer-2 broadcast domains, which in turn requires the implementation of
VLANs.
b.4) In consequence of all facts stated above, we believe that the EFM
standard must define the implementation of VLANs through the network,
allowing for the flexible assignment of VLANs at the customer level for
easy service selection.
b.5) As an extension, the question of how to distribute the same video
signal inside the customer premises (at his home or office) may also
require flexible VLAN assignments. We have studied this problem, and we
believe that the amount of broacast data that is sent to some units (such
as set top boxes) must be limited to avoid congestion. The implementation
of features for dynamic provisioning of 802.1q compatible VLANs through the
home network is one of the possibilities that we have analyzed, but there
are other alternatives. We're available to discuss the issue with anyone
interested.
3.2.4. SLAs
The implementation of SLAs for the services described above is subject to a
lot of debate. We consider two different strategies for the implementation
of SLAs:
3.2.4.1. Conventional, ATM-style, fine grained QoS. This allows for the
enforcement of precise SLAs. The problem here is the scalability and the
cost, both in terms of switching equipment and also in terms of management
of the network.
3.2.4.2. Give the user as much bandwidth as needed. By overdimensioning the
trunks, the SP may be able to offer very high levels of quality without
much regard to the fine-grained QoS control. Proponents of this solution
say that, while you have overdimensioned trunks and a not-so-precise SLA,
the capital and operational costs are potentially lower.
With regards to the actual SLA implementation, we decided to take a careful
approach. We understand that fine-grained QoS controls are in high demand,
but are so expensive today as to make their deployment for the mass market
impossible. This is a decision where we have clearly not reached a
consensus. Every carrier will do it differently, and both cost and
scalability will be key for the decision.
3.3. Cost
The focus of our work is on the FTTH market, in particular for the
residential customer. This market is highly cost-sensitive, but it is the
only market where we have the scale and the relative concentration of users
to effectively deploy the new technology. Any approach that focus on the
requirements of the business market will have difficult scaling to the
residential market, because:
- at least in CTBC Telecom's case, the demand in the business market is not
as concentrated as in the residential market. The strategy for the
residential market allows for the deployment of the new service
infrastructure in selected areas, growing slowly to include more and more
customers. In the business market, the pressure to deploy the service for
scattered customers over a big area is much more difficult to handle.
- the business market is less cost sensitive. There will be less pressure
to increase scale or reduce the costs of the technology.
- the residential market asks for a more practical and transparent approach
to provisioning. We have historical evidence that solutions developed for
the business market have trouble scaling to the residential market.
- the technical features asked for by the business customer tends to add
complexity, thus increasing the cost of the solution. One such example is
the common requirement of fine grained SLAs. It is pretty hard to reduce
this complexity thereafter.
We believe that all effort must be done to focus the standard work on the
residential market. The base standard must as simple as possible, with
careful reservation for future growth. This growth must be possible in
incremental steps, allowing for graceful migration to more advanced
technologies. We believe that this is the best strategy to leverage the
huge scale of the residential market.
3.4. Security
Security-related concerns, such as privacy, isolation of the communications
between customers, are of very high importance. In some countries,
law-enforced taps may be mandatory (this is not the case of Brazil, as far
as we are concerned now). This is a particular issue with public service
providers, such as telephone companies. Other types of service providers
may not be subject to the same security concerns (at least not at the same
level).
3.5. Interoperability
All the units must be based on open standards, and as such, all the parts
of the solution must be interoperable. In particular, the network
termination unit must be interoperable between different vendors. This
interoperability should preferably be as simple as connecting a new network
termination unit with automatic recognition by the access network platform.
We would like to point out that the development of a working business model
for the units to be sold to the residential market is also a very important
part of the strategy that should not be relegated. The customer must have
choice of interoperable units offered directly to him in the open market,
and not only through the network owner. In order to develop the market,
CTBC Telecom will be providing network termination units to the initial
users of the service as part of the service package, either through a lease
or other financial arrangement. However, we intend to leave the purchase of
the network termination box open to the customer choice as soon as
possible. This step will allow for more competition in the market, leading
not only to lower costs, but also to the development of new units that best
fit the needs of the customer. We have historical evidence that shows that
this is the best way for quick development of the market.
3.6. Maintanability
The system must be easy and as inexpensive as possible to maintain. To
reduce the operational costs associated with operating crews, the system
must include solid remote provisioning and troubleshooting tools. Related
to the maintenance problems, we can detail some specific topics:
- The physical design of the network has to take into account the needs of
the maintenance crew.
- Testing points must be provided to allow for easy testing of the system.
- The use of connectors must be evaluated, as a way to make maintenance
easier by quickly changing parts of the cabling. This is particularly true
at the drop level. We understand that the use of connectors does have a
impact on reliability, and also introduces higher losses on the optical
path; however, the practical issues addressed by the easier maintenance may
make it much more practical.
- There must be provisions for remote management of the field units, even
in the case of the passive ones. In a PON network, the use of optical
systems to detect situations such as open splitter boxes must be thorougly
understood. In the P2P, there must be MIBs in the remote units for
environmental and physical security management.
3.7. Scalability
The network must allow for easy growth by incrementally adding new accesses
with automated provisioning, while keeping the service levels within the
bounds defined by the contract. The base requirements are:
a) The addition of a new network termination unit must not:
a.1) disrupt the service to the other users in the same segment;
a.2) force a slow renegotiation of 802.1d or similar protocols. The
renegotiation has to be either completely avoided, or at least optimized
for the constraints of the application;
a.4) significantly affect the latency per user;
a.5) affect the availability of some particular service, such as video
broadcast, due to insufficient resources to honor all the QoS demands. In
this case, the new unit must not be activated.
b) The removal of a network termination unit must not:
a.1) disrupt the service to the other users in the same segment;
a.2) force a slow renegotiation of 802.1d or similar protocols. The
renegotiation has to be either completely avoided, or at least optimized
for the constraints of the application.
c) There should be mechanisms to activate new services after the
installation of the ONU by means of automated provisioning. Services based
on STB (set top boxes) installed at the customer premises should be easily
controlled at the management center. We assume that the STB mentioned above
are 802.3-compatible equipments, connected by means of internal Ethernet
cabling inside the customer premisses, and thus, may be connected to the
core network by bridging. This may or not require specific modifications at
the 802.3ah layer.
3.7.1. Wavelength allocation plan
Of similar importance, is the potential of using the same optical
infrastructure to deliver advanced services using several wavelengths, by a
CWDM or DWDM design. CTBC Telecom proposes that the EFM working group
should define a plan for wavelength allocation which will allow the
coexistence of the first generation of services, using a single wavelength
approach; and the future, 'xWDM', layers of service, using the same fiber
infrastructure. The initial allocations may reserve wider bands for the
first generation, in order to allow for the use of customer-grade
components of less precision. Future allocations will be done using
narrower bands, allowing for a progressively denser use of the available
spectrum on the fiber. Older allocations may be potentially revoked (or
'deferred') in the future, allowing for a smooth transition.
As for the possibility of using the current ITU standards for wavelength
allocation: we understand the importance of adhering to international
standards. However, as it is laid out now, we feel the current ITU
allocation plan does not give much room for the EFM standard. As a result,
we may end up needing to implement more expensive components. It may be
possible to propose an alternative plan, best suited for the needs of the
access networks (in opposition to the DWDM core/metro networks, from where
the current plans seem the be derived). This question has to be analyzed by
the PMD subgroups.
3.8. Reach
Given our distributed architecture, reach is not our ultimate concern. In a
tradeoff, reach would be considered less important than some of the other
issues outlined here. There is one important exception in the case of
remote customers, such as in rural areas. In this case, we believe that the
technology used is a completely different case, with a different set of
requirements. Any attempt to develop a single standard for both the
high-density/short reach market and for the low density/longer reach market
has the risk of being too complex and expensive for both cases. A common
layer of interoperability in the upper layers is enough to keep both
platforms integrated; however, the physical demands are so different that
it may be better to have separate solutions.
Carlos Ribeiro
CTBC Telecom