|While the communications industryï¿½s doldrums
continue, Internet traffic is consistently growing ï¿½ that is, growing at a
rate thatï¿½s consistently incredible. As the world demands more
information, the pipes that both feed and get fed all that information must
keep expanding. This is great for the industry, since it means that sooner
or later, someone is going to have to buy something again.
What will they buy? Theyï¿½ll buy products that can both consume and
distribute these increased data flows. The products will require, among
other things, faster and more capable silicon, increased I/O throughput,
better cooling, increased availability, and faster internal routing. Since
central office space is finite, these systems will have to be ï¿½biggerï¿½
without actually being bigger. That is, the systems will need to step up and
handle the new requirements, at the same time conforming to some of the old
Can current systems handle these new requirements? In the short run, yes.
But the industry also sees that in the not-so-distant future, todayï¿½s
open, modular system hardware architectures will be limited, specifically if
the target environment is a central office (CO).
I think itï¿½s appropriate to confine our discussion to future
capabilities of open, modular systems. As we discussed in last monthï¿½s
column, I believe we can attribute much of the incredible innovation in the
Internet telephony industry ï¿½ innovation weï¿½d all like to see continue
ï¿½ to the use of open systems. With this in mind, the next leg of
innovation for our industry is to bring these open, modular components to
the CO using a standard hardware shelf targeted to meet expanding density
and I/O needs.
Are we in good shape to continue using open systems as requirements
become more demanding? Yes, and there is much progress in this area. One
such initiative, called the Advanced Telecom Computing Architecture (ATCA),
is being developed by the PCI Industrial Computer Manufacturerï¿½s Group (PCIMG),
the same group that brought us the CompactPCI* standard. ATCA has been
designed from the outset to meet next-generation communication requirements,
with the Version 1.0 specification release targeted for October 2002.
Letï¿½s take a sneak peek at some of the features and benefits of ATCA.
Since weï¿½re assuming these systems will have denser and faster silicon
components, there will inevitably be more heat to dissipate. How does ATCA
address this? First, there is a higher pitch between boards, which allows
both increased airflow and the use of on-board heat sinks.
What about increased density? Again, weï¿½ll assume new silicon will
handle increased throughput. But with ATCA, board size also increases,
allowing for more components on the board. While this may seem backwards, it
also allows for better cooling, since there is a larger surface area over
which the cooling air can flow. Since there will be more, denser, faster,
and hotter components, the folks developing ATCA also believe the systems
must be able to handle up to 200W in a single slot (as opposed to less than
70W in CompactPCI). In other words, the cooling had better be better.
What about increased availability and reliability? The power distribution
scheme defined in ATCA is ï¿½48VDC (redundant feeds) to accommodate central
office standards and central office battery backup. Also, as boards have
more varied silicon, each with different voltage and current requirements,
it becomes more difficult to supply each voltage required over the backplane.
Therefore, if the standard hardware shelf can distribute a single higher
voltage, locally converting it board by board, voltage distribution becomes
easier. Granted, a per-board converter takes more real estate. But it also
improves reliability by simplifying power supplies.
What about increased I/O throughput and faster internal routing, both
crucial for keeping down latency as traffic increases? While the proposed
ATCA 3.0 specification defines the mechanicals, power, and system management
we discussed, it also defines a generic backplane capable of supporting a
variety of switch fabrics with up to 5 Gbps signaling. The proposed 3.1,
3.2, and 3.3 specs detail different kinds of fabric support that can be
installed into 3.0 backplanes. The 3.1 spec details the Ethernet fabric
interconnects, similar to CompactPCI 2.16 but going beyond 1GbE. The 3.2
spec details InfiniBand* fabric interconnects. The 3.3 spec details
StarFabric* interconnects. These fabric definitions ensure compatibility
between the installed switch fabrics and processing boards that make up an
ATCA system. With these different kinds of fabric support, you donï¿½t need
a bus arbiter active at all times, which allows for increased throughput.
So how does this relate to Internet telephony? As an example, letï¿½s
look at the ability to create high-density VoIP gateways. ATCA would provide
the space and cooling to deliver the processing power to drive the large DSP
loads necessary for transcoding. With todayï¿½s DSP technology, you could
create gateway boards with densities of up to 8,000 ports.
There is one issue: There is no equivalent H.110 bus. But does this
matter? With such large board densities, there may be no need to move the
data off the board, right? Although that last comment was facetious, it is
possible, depending on the design. Even so, buses are not well suited for
the reliability and large port densities envisioned for ATCA. With ATCA, the
system designer has a very high-speed packet interface that is fully capable
of delivering voice over packet (VOP) and can scale well beyond the H.110
limitation as market conditions dictate. With a fabric backplane, the system
designer uses TDM in a different way. It moves from being propagated within
a system on a bus to being terminated on a card and carried over the serial
fabric. This addresses the performance, cost, and scalability issues
inherent with the H.110 bus.
If you like CompactPCI even a little bit, youï¿½ll love ATCA. If youï¿½re
on a proprietary hardware architecture but donï¿½t really want to be,
youï¿½ll love ATCA. And if youï¿½re a developer looking to continue
innovating, youï¿½ll love ATCA. Best of all, if youï¿½re a service provider
looking for innovative and cost-effective solutions that meet your
ever-increasing needs, youï¿½ll LOVE the companies that sell you ATCA
Jim Machi is director, Product Management for the Network Processing
Division of the Intel Communications Group. Intel, the worldï¿½s largest
chip maker, is also a leading manufacturer of computer, networking, and
communications products. For more information, visit www.intel.com.
*Other names and brands may be claimed as the
property of others.
To The September 2002 Table Of Contents ]