ATCA: A Practical Perspective

Advanced Telecommunications Com-puting Architecture, known as ATCA, is a new system
form-factor defined by the PCI Industrial Computers Manufacturers Group (PICMG)
and was created with a number of objectives in mind. The main objectives of the
ATCA standard are to enable building carrier-grade convergent systems—systems
that include computing and communications products for myriad applications.

Specifically, it is aimed at converging telecom access and edge equipment functions
with data center and storage equipment functions in a modular fashion. This modular
approach enables both equipment producers and users to employ the same chassis/backplane
for multiple types of products, using different modules, as explained later. As
such, the ATCA standard was developed with the following attributes, among others,
as the main guiding principals:

  • Scalable Capacity of up to 2.5 Tbits/s (per chassis) with a centralized
    switching hub interconnected to all module slots in a star (radial) or a full
    mesh configuration
  • Redundancy throughout the system configuration to achieve over
    99.999% availability (Carrier Grade), while allowing less demanding applications
    to utilize a non-redundant configuration for lower cost
  • Modularity and configurability to enable multiple modules with
    various interfaces and different technologies, such as DSPs, NPUs, CPUs and
    storage media to be mixed and matched for diverse applications in the same
  • Specifications that support strict regulatory requirements such
    as NEBS with long life components and advanced power distribution and cooling
  • Specifications that support multiple types of Switching Fabrics
    (the core of the platform), such as Ethernet (GbE), PCI Express and others

The result is a standard that offers definitive advantages over its predecessors
such as Compact PCI. ATCA supports regulatory requirements such as NEBS, as
well as high-availability enabling requirements. The architecture’s higher
power allowance and more advanced power distribution mean designers can pack
more functionality and higher performance per shelf. ATCA provides higher capacity
and the ability to mix various types of modules and technologies to enable the
convergence of multiple types of equipment and multiple applications in one
platform. It also allows rear-mounted modules permitting added functionality
per slot or rear interface cabling. Figure 1 depicts an outline of the physical
attributes of the platform.

Additional software—or middleware—specifications to support high-availability
platform infrastructure are under development by organizations such as the Service
Availability Forum. Such high-availability infrastructure middleware enables
building robust ATCAs, as well as other platforms such as CompactPCI, which
provide a standard foundation for high availability, distributed computing and
system management capabilities across the industry. A distinct advantage of
such standards is the ability to build multivendor embedded systems that simplify
building higher-layer applications by NEPs (Network Equipment Providers) in
a shorter time-to-market.

Three Avenues for Differentiation in ATCA

There are three fundamental areas in which the system architecture based
on an ATCA platform could be different.

Difference in Switching Fabrics Configuration

The switching fabric is the heart of the system in that it transports
all the traffic between the different modules of the system, both within
a chassis and between chassis. One architecture could use a single switching
fabric to handle all traffic—user information as well as control
and management information—in some cases even storage modules may
be interconnected through the same fabric. Another architecture may use
separate switching fabrics for each different class of data, such as control
data, user data and storage data. A hybrid approach could separate storage
from user and control flows via two separate fabrics.

Difference in Switching Fabrics Technologies

There are many switching fabric technologies that can be used as an
ATCA switching fabric, such as 10/100 bT Ethernet, GbE, Fibre Channel,
PCI Express, Infiniband and Cell-Based fabrics to name a few. Furthermore,
they can be mixed as base and extended fabrics, where the base is used
for one purpose and the extended for another in the mixed fabric architectural
model discussed in the last paragraph. This leaves multiple choices from
which to select the appropriate fabric for the appropriate application
as well as a path of evolution as fabrics evolve. While changing a fabric
will require new modules to support the interface to such fabric, the
main chassis and backplane remain intact and capable of simultaneously
supporting two fabrics at a time in any given chassis. This provides an
added advantage of utilizing an ATCA platform without a “forklift”
approach to evolution.

Difference in Processing System and Shelf Management

There are architectures that combine the processors that handle the
system and shelf management with the switching fabric in one hardware
module. Others use a separate CPU module for management. In this case,
the former architecture offers definite economic and simplified implementation
advantages over the latter.

Multiple Product Possibilities

The capabilities built into the ATCA specifications enable a compliant platform
to support a number of products for multiple applications following the building
blocks concept discussed below. To create multiple products utilizing the same
platform, it is necessary to understand which basic common building blocks are
shared among the products, create modules that meet the requirements for such
building blocks and use the proper mix of the building blocks to create each
specific product.

Consider for example three telecom products: a blade server, an SGSN and a Media

Blade Server. A blade server is a modular computing platform composed of multiple
blades/modules that can span one or more chassis depending on the capacity required
and the type of performance needed. A blade server can also use embedded storage
media as part of its own infrastructure or external network-based storage, both
of which are supported by the ATCA platform.

Note that a blade server has many applications such as softswitches, HLRs, SCPs
and network management/OSSs, not to mention data center processing equipment
including data bases and web servers. A basic blade server can be achieved by
building a switching fabric (as the heart of the platform inter-module and inter-chassis
communication), different types of computing modules—which can be modular
themselves—and I/O modules such as GbE interfaces. Add storage modules
and a versatile blade server platform is born

SGSN. The Serving GPRS Support Node is the network element in a mobile wireless
network that supports user data communication functions within the network.
Depending on the generation of the network (2.5 or 3 G), the data is separated
from the voice by the Base Station Controller (BSC) or the Radio Network Controller
(RNC) and forwarded to the SGSN to perform data packet processing such as authentication,
forwarding, SS7 signaling and user plane protocols such as tunneling over ATM.
An SGSN platform requires computing capability and network interface modules
to handle the protocol processing and interface to the other network elements.

Clearly the computing and switching fabric elements are common with the blade
server. However, different interface modules are required, e.g., a blade server
may only require Ethernet interfaces, while an SGSN requires Ethernet in addition
to standard network interfaces such as E1/DS1 and SONET/SDH interfaces (OC-3c/STM-1).
This brief description illustrates how the SGSN utilizes certain modules/building
blocks common with the blade server while requiring another set of different
modules/building blocks.

Media Gateway. Media Gateways (MGs) are widely used in service provider and
enterprise networking applications. An MG is a good example of how an ATCA platform
may be utilized in more than one market segment and a variety of applications
with satisfactory economical and technical results. An MG is also a good example
of employing the building block concept to achieve re-use of modules/building
blocks common with other equipment. Of the many applications of a media gateway,
multiple protocol conversions for voice services can be used as an example.

Because of the variety of service interfaces between TDM and Packet Voice services,
it is necessary to provide interworking functionality between the different
interfaces to create seamless services among the different subscribers. Of course
the media gateway is simultaneously required to support signaling, which could
also be multiple types of signaling at the same time.

To achieve that, the media gateway requires the use of CPUs (for signaling and/or
packet processing) as well as specialized processors to deal with the interworking
functionality, which is typically between voice over TDM and packet voice over
ATM or IP, but could also be between ATM and IP. This indicates that an MG can
use the switching fabric and CPUs utilized by the blade server (for inter-module
and inter-chassis communications, management and signaling/control functions)
while requiring specialized modules for interworking and voice packet processing.

The latter could be DSPs or NPUs depending on the system architecture and
design as well as the requirements. For example, if compression/decompression/silence
suppression are required, it would be best if DSPs are employed. Otherwise,
NPUs will do the job for interworking.

As the above description illustrates, it should be clear that there are common
elements between all three types of products and some unique elements in each.
Of course the higher layer applications may be quite different, but those are
achieved via software that is integrated with the embedded system platform based
on ATCA. Figure 2 illustrates, in color code, the similarities and differences
in building blocks among the three product examples just discussed.

Technology and Products Status

Currently there are a few platforms available from different sources, including
RadiSys. Some suppliers offer one element only, such as a chassis or a processing
module. Others offer a more complete solution including the chassis and processors.
Mass Storage media (other than a few Gbytes on the processors) will be available
in early 2004. The products available today are not at a production released,
GA, level but are expected to reach that state in the first quarter of 2004.
Additionally those are products that are more suitable for blade servers. Products
appropriate for other network elements (mentioned above and shown in Figure
2) are expected to be available for lab trials in the first half of 2004, and
GA, one to two quarters later depending on performance and trial results.

ATCA is a clear winner as it will provide economies of scale coupled with scalability
and high performance for multiple critical network elements in both wireless
and wire-based networks. Early availability products are currently undergoing
lab trials by early adopters, and multiple interoperability workshops are held
every year by the PICMG members to ensure multivendor compatibility, another
winning proposition for the industry.

Hillsboro, OR.
(503) 615-1100