|
The Right Tool: Next-Gen
Data Transport Architectures
BY BRIAN CARR [
Go To Sidebar: An HA Multi-Service Platform ]
There’s an old adage that says: always use the right tool for the
job. It’s just common sense — you get the job done more quickly and
with better results, and it’s often the difference between a job done,
for example, by a professional builder and the same job done by an
enthusiastic homeowner. Many column inches have already been written about
the next-generation converged network and the need for network and service
to evolve to embrace packet protocols. The dominant protocol — in the
access, edge, and core portions of the network — will eventually be the
Internet protocol (IP). However, where voice communications is concerned,
IP still has a way to go. There are still significant quality of service (QoS)
issues to be ironed out, issues that are already addressed and mostly
solved by ATM. This gives ATM a prominence in existing implementations,
particularly in the mobile communications sector, which guarantees ATM can
expect to be an important technology for a long time yet.
Open standards are able to help accelerate the development of converged
network equipment, and CompactPCI has quickly established itself as a
leading open-standards choice for multiple-card telecom applications. The
PCI Industrial Computer Manufacturers Group (PICMG), which manages the
standard, has moved quickly to establish advanced inter-card data
transport mechanisms as required to help support the evolution of
applications.
This article examines how these various open standards apply to the
real demands of next-generation equipment and considers whether they are
sufficient in themselves to help meet the needs: to be the right tools for
the job.
The Computer Telephony Bus Extension (H.110)
The H.110 computer telephony bus extension is designed to carry voice
calls between cards in a chassis in the same 64 Kbps
time-division-multiplexed (TDM or timeslot) form that they are received
from the public telephone network (PSTN).
The concept is simple and scaleable. Voice calls are terminated on
dedicated line termination cards and are routed via the H.110 bus to a
range of additional voice processing or DSP resource cards that perform
specific functions under the control of a service application.
When voice over IP applications such as VoIP trunking and access
gateways began to be considered, it was natural to use the same basic data
flow architecture. However, the CompactPCI bus is not suited for
aggregating packet voice flows. Including the packet processing and packet
network interface on the DSP resource card solves this. Each
such packet voice resource board then has a separate redundant 100 Mbps
Ethernet link for the packet data stream that, for larger installations,
is aggregated upstream in an external Ethernet switch.
The limitation of H.110 bus timeslots remains a serious issue for
scalability — particularly for high-density packet voice gateways that
are intended to support many thousands of users. The H.110 bus is limited
to 4,096 timeslots, supporting only 2,048 users. One way to address this
limitation is to remove the need for H.110 entirely by provisioning each
packet voice resource board with a suitable line termination and
sufficient DSP and packet processing resources to process all channels.
This processor blade approach solves the problem without resorting to the
extra expense and backplane designs for an H.110 segmentation approach,
but does mean that each card should do everything required by the
application.
Packet Switched Backplane Extension
Having addressed the limitation on scalability for IP connected
systems, larger and larger VoIP gateways could be constructed by using
more and more cards. With each card requiring Ethernet links to both
redundant external Ethernet switches, the cabling and space required for
the external Ethernet switches has come under scrutiny. The result was
simplicity itself — to bring the multiple redundant Ethernet links and
aggregation switches inside the chassis. This is the premise behind the
CompactPCI Switched Backplane (CPSB), now standardized as PICMG 2.16.
CPSB has two dedicated fabric switch cards in a shelf. Each peripheral
node is linked directly to each fabric using traces that can support
10/100/1000 BaseT Ethernet. Each switch fabric is designed to perform
aggregation and switch/routing for the entire chassis and features gigabit
Ethernet connections to a LAN/WAN.
CPSB is complementary to H.110 — it means that both H.110
architecture systems and all-in-one packet voice resource boards work
equally well. It also means that multiple processor clusters
are capable of being easily implemented, supporting applications like
database servers and cellular home and visitor location registers (HLR/VLR).
Furthermore, the native Ethernet link to each card slot has another
advantage that is expected to become more important in the next year or
so: that of improving high availability by replacing the CompactPCI bus as
the primary means of application control.
Considerations For HA
Despite the CompactPCI bus standard beginning the drive towards open
standards-based HA systems, it is actually the bus itself and the
accompanying driver architectures that are now perceived as barriers to
achieving higher availability. Any single bus must be considered as a
single point of failure in a system and so must either be duplicated or
eliminated. Duplication of the PCI bus across a large chassis with
multiple bridging is much more complex than elimination, but until
recently, standards to replace the capability of PCI to manage both
low-level hardware and higher level software did not exist.
A standard for low-level hardware management evolved from efforts by
Intel in the desktop and server PCI space to develop the IPMI management
bus. This was brought into the CompactPCI arena as the PICMG 2.9 standard.
Across the management bus, each card slot is able to be interrogated and
controlled independently without recourse to the PCI bus. When implemented
as a star topology with a link to each card, the management bus itself is
not a single point of failure.
The missing link was application level control. However, with the
advent of PICMG 2.16, each card is now connected to a redundant Ethernet
network, and so high-level management and command and control can be
implemented using a secure IP transport mechanism. The way is clear to
remove the actual CompactPCI bus and eliminate this single point of
failure — and this mode of operation was specifically supported within
PICMG 2.16.
ATM Transport — The Challenge
ATM is required for advanced applications such as voice over DSL
gateways and is used heavily in UMTS (3G) wireless elements such as Node B
base stations, radio network controllers (RNC), and media gateways (MGW).
However, most recent emphasis in the standards arena has focused on
improving the data flow architecture for IP communications while keeping
the basic telephony handling capabilities of H.110. Nothing yet has
addressed the needs of ATM.
This omission may be traced back to the roots of the open standards
industry. Until now, most applications for open standards have been in the
access network — feature-rich service platforms with relatively few
channels. The core switching network elements were predominantly
proprietary, and it is in the core network or the network edge that ATM
has been most evident. This situation is now being changed first by the
network evolution towards all-packet networks (where ATM has already
solved all the quality of service issues confronting the all-IP
proponents) and second by the growth in the outsourcing business model
that leads even core network equipment manufacturers to look to open
standards for their next-generation developments. Both aspects are now
bringing ATM requirements to the open standards industry.
Currently, the situation is evolving in the same way that CPSB evolved.
Rather than aggregating and distributing ATM traffic within a chassis, ATM
connections are being taken directly from the cards to external ATM
switches. Unfortunately, this is problematic not only from a cabling view.
The latest ATM/AAL2 standards allow up to 248 channels to be embedded
within a single virtual circuit, but many ATM switches can only make
connections to the resolution of a single virtual circuit and so cannot
perform individual call switching.
This means that the industry needs to develop a new architecture with
the capacity and cell structure of ATM plus the redundancy of CPSB and the
per-channel resolution of H.110.
Innovating New Standards
The needs of communications equipment are evolving quickly, and the open
standards toolbox is evolving to keep step. Advanced network elements must
be capable of packet communications as well as the more traditional
timeslot-based voice communications. The growth in the outsourcing
business model has meant core network reliability requirements being
placed on open standards equipment. New tools are needed.
At a hardware and interface level, this has meant the definition of
standards that build on the successful base of CompactPCI to offer
timeslot traffic distribution and Ethernet IP packet distribution within a
chassis. The new management bus, together with the all-slot IP control
that CPSB brings, allows the CompactPCI bus itself to be eliminated. These
tools, especially CPSB, are certainly effective for extending traditional
computer telephony style applications into the next-generation converged
network. However, H.110 has some severe limitations, and there is one
crucial tool missing — the one that helps support ATM applications.
For ATM there is no standard underway yet that meets in-chassis
aggregation and distribution requirements. Thus it is left to individual
innovation to fill the gap, and to provide the right tool for the job.
Brian Carr is telecom product manager, Motorola Computer Group. For
more information about Motorola Computer Group, please visit www.mcg.mot.com.
[
Return
To The December 2001 Table Of Contents ]
|