ATCA's at a Fork in the Road

The Advanced Telecom Computing Architecture (AdvancedTCA) is gathering momentum among telecom gear makers, but there are two paths to take regarding switched interconnect standards (see AdvancedTCA Makes Headway).

Two groups have come up with their own proposed standard ways to interconnect processors, memory controllers, and other in-system devices for carrier-grade networking environments. The RapidIO Trade Association (RIO) and Advanced Switching Interconnect SIG (ASI SIG) are each aiming to handle the demands of multi-gigabit backplane applications that will push the limits of tomorrow's telecom systems.

The significance of ATCA, which is based on the PCI Industrial Computer Manufacturers Group (PICMG) 3.0 group of standards, can't be overstated. The standards push seeks to assure that telecom components will become more like PC parts in that they'll be commodity items. This will result in lower overall prices for telecom equipment, because previously proprietary components will be off-the-shelf. And it will result in lower operations expenses for carriers, because ATCA gear will interoperate with other devices from different manufacturers.

But there are several speed bumps on the road to one set of standards for all telecom gear -- and the latest appears to involve different interconnection techniques.

"ATCA has become the evaluation platform of choice,” explains Dan Bouvier, PowerPC architecture manager for Freescale Semiconductor Inc. and chairman of RIO. “Every OEM’s lab has ATCA chassis in it. And that’s where they are evaluating all of the new backplane technologies, protocols, next-generation system architectures, and associated topologies” like those for RIO and ASI SIG.

Is Ethernet the answer for these connections? Not really. Ethernet was a natural fit for PICMG 3.1 as it’s pervasive, and a huge ecosystem of board and component makers, standards, and software stands behind it. But there are technical limitations to using Ethernet in high-availability communications equipment with multi-gigabit-per-second throughputs, according to OEMs, especially the computing power required to terminate the TCP/IP protocol stack in an Ethernet environment.

From that knowledge grew the two distinct approaches to a switched interconnect: RIO and ASI. ASI is now PICMG AdvancedTCA 3.4 and RIO is PICMG AdvancedTCA 3.5.

Before highlighting the two groups' differences, it helps to know what a “switched interconnect" is in the first place. Lucent Technologies Inc.'s (NYSE: LU) Dave Wickliff, who is vice chair of the RIO Steering Committee, explains the concept this way:

“In the classic PCI bus implementation, you can have up to eight boards or devices, all sitting on the same electrical bus or electrical signal path. You have a basic physical limitation in your system as to how fast you could drive that with all the different devices hanging off the same wire.

“With a switched interconnect, we say, ‘Let’s put two nodes on a wire -- a sender and receiver -- and we can drive that little wire much faster.' " Once the sender and receiver are established, Wickliff says, a switch can be used to stitch together several point-to-point links and "recreate the interconnect that you had with the bus."

All that said, in a world where packet-based applications are bringing the computing and communications worlds together, RapidIO (communications) and ASI (computing) are running a dead heat.

RapidIO was originally put forth by Freescale with an eye toward the embedded computing, real-time operating system (RTOS) world that houses Freescale’s PowerQUICC control plane CPUs and wireless base-station DSPs.

The RIO trade group comprises these same sorts of embedded players: Ericsson AB (Nasdaq: ERICY), Alcatel SA (NYSE: ALA; Paris: CGEP:PA), Texas Instruments Inc. (NYSE: TXN), Mercury Computer Systems Inc., and so on.

ASI grew from Intel Corp.'s (Nasdaq: INTC) interest in a scheme to connect PCs, servers, and storage at multi-gigabit speeds with very high availability. Today the group includes storage powerhouse EMC, Huawei Technologies Co. Ltd., and Siemens AG (NYSE: SI; Frankfurt: SIE), among many others. ASI has more silicon design engineering and semiconductor IP providers than does RIO, mainly because these folks are in the business of pleasing Intel.

The bigger goal for each group is to pull in the hundreds of other hardware and software concerns that could ensure its proposed standard's commercial success.

Who will win as these two groups advance their cause? In the short term, the programmable logic players, including companies such as Xilinx Inc. (Nasdaq: XLNX) and Altera Corp. (Nasdaq: ALTR). Until chip A can talk to chip B in the same language via an interconnect standard, equipment OEMs will turn to programmable devices to bridge them together. Engineers often call these parts “glue logic.”

So the two standards are going to coexist for a while. And, until silicon built to RapidIO and ASI appears, around mid-2005 and mid-2006 respectively, programmable logic will be continue to provide the glue that systems vendors need to meet carrier demands (see ATCA to Be Worth $3.7B in 2007). Unlike the stock market, the programmable chips market likes a little uncertainty.

— Gale Morrison, special to Light Reading

edgecore 12/5/2012 | 1:07:12 AM
re: ATCA's at a Fork in the Road ..oops, hit the button too fast :-(

A word on RIO vs ASI, for ATCA where the philosophy is to ride the economies of scale and the cost curve from the desktoparound

What I meant to say was that RIO has no chance whatsoever...its all about reducing the costs of the chips (that only the desktop volume can amortize)...so even if Motorola owns telcoms, only ASI can survive, driven by Intel...

Want proof, look at the next gen 8641D from Motorola, it has both PCI Express and RIO interfaces...because PCI Express = Intel = cheaper chips and economies of scale...and Yes I know that PCI Express does not equal ASI...but it is all Intel!


Does ATCA have a chance, Yes it does, but it will take a long time and it will only be across a certain set of network elements.

Am I thinking along the same lines as some readers or am I way out in left field?

edgecore 12/5/2012 | 1:07:12 AM
re: ATCA's at a Fork in the Road The tier 1 OEMs have grown up on custom HW and Motorola embedded processors...allright Cisco did grow up on Mips, but that is moving to PowerPC, slowly but surely...

Back to the topic:

VME came along, Compact PCI came along....they garnered about 1% of all HW developed by tier 1's, mostly in telephony centric applications.

Most of the network elements out there are obvioulsy custom designs; the custom hw still is much better suited for the job, requirements for much bigger line cards, better thermals than ATAC can provide...combined with legacy frames upon frames of carrier owned custom HW make the intersection point with ATCA difficult to imagine.

Where ATCA makes sense, and where I assume most Tier 1-2-3's are planning to use it is as a base platform (a set of defined shelves for compute intensive applications where the performance enveloppe of a fully loaded ATCA shelf can easily be defined, measured and sold...load it up with 7457's or BCM 14xx or your favorite processors...). Aside from one base platform per OEM, what will be the critical turning point that will make everfyne want to embrace ATCA vs the bigger custom systems that are allready in place and working fine?

A word on RIO vs ASI, for ATCA where the philosophy is to ride the economies of scale and the cost curve from the desktoparound

As Intel silicon evolves on the desktop, and as Intel chips gets hotter and hotter, the only way for Intel to chase the dream of one day playing in bigtime telco systems is with a version of Compact PCI that can actually handle the thermals...hence the ATCA dream...plus nice things that have limited acceptance such as NPU's and SAF sw fit well into a modular story crafted by Intel around ATCA.

ragho 12/5/2012 | 1:07:11 AM
re: ATCA's at a Fork in the Road
You make some interesting points, but as a close watcher of ATCA in the past few years, I have some comments.

RIO at first was geared toward system interconnects within the board. I have considered it in the past but preferred StarFabric (ATCA 3.3) for various reasons. I've been promised ASI for quite some time now but yet to see anything of practical use.

PCI Express (as in original ATCA 3.4) will not cut the mustard when it comes to transport and core network applications; perhaps low end access and server applications. ASI and RIO have both a chance at this, but I actually believe RIO has a better shot. One, IMHO, it looks like it will scale very well. Two, it already leverages established signaling and routing concepts where the COST factor will come into play.

Your comments are probably very meaningful in specific application segments. The areas I'm particularly interested in are edge/core, transport/network applications. VME and PICMG 2.x never made heavy inroads into this area simply because they were shared architectures that never appealed to the multi-gigabit OEMs.

PICMG 2.16/2.17 has addressed some of this, with many switched gigabit (meshed and non-meshed) fabrics available today. On the low-end aggregation (5-15 Gbps), this is an excellent choice in terms of cost and compliance (NEBS, ETSI, FCC...). This may not sound like much, but many carriers using edge boxes for BRAS/DSL/FTTx are happy to get 20-40G in a 7' rack today (my experience).

Which is what keeps my interest and some faith in ATCA. Apart from the obvious benefits, the key wish for me is to be able to deliver 100-200 Gbps in a single 7' rack. And I believe that both ASI and RIO can deliver this in the next two years. Stargen already has 2.17 StarFabric switch that delivers ~100G, I'd like to see that with ATCA 3.3 with some solid node/hub boards.

When that happens, I'll be sure to go get some VC money ;-)
BobJG 12/5/2012 | 1:07:02 AM
re: ATCA's at a Fork in the Road re:
What I meant to say was that RIO has no chance whatsoever...its all about reducing the costs of the chips (that only the desktop volume can amortize)...so even if Motorola owns telcoms, only ASI can survive, driven by Intel...

While Intel+desktop often means success, this is NOT true for ASI. It is technology very far afield from the desktop, and RIO can be just as friendly with PCI/PCIe as ASI claims to be.

The real battle is between RIO and Ethernet, and in most cases, Ethernet is just too dumb.

Re: ATCA - the real benefits might be in the standardization of the subcomponents (connectors, faceplates, power supplies, and yes - backplane interconnect, etc.). Most vendors will still want to bend their own sheet metal and build their own boards....

edgecore 12/5/2012 | 1:07:02 AM
re: ATCA's at a Fork in the Road -------
The real battle is between RIO and Ethernet, and in most cases, Ethernet is just too dumb

It's really hard to bet against Ethernet. Plus it also really hard to think of a component market developing around RIO...Ethernet seems to be here to stay for a while....its is pervasive and RIO is still confusing people and not well supported, nit even by Mot...aside from Tundra, who has a lot of their eggs in the RIO basket?

sigint 12/5/2012 | 1:06:54 AM
re: ATCA's at a Fork in the Road It's well known that Intel hasn't yet managed to become the preferred silicon vendor for most networking and telecom OEMs. Pushing forward ATCA allows Intel to flex it's muscle in the communications space in partnership with vendors of low cost gear, i.e., the Huaweis of this world.

This would likely work to the goliath's advantage in two ways:

1) Chip sales to low end OEMs
2) High end OEMs (Cisco), irritated by Intel's romance with others yield and start incorporating Intel silicon at the expense of internal ASIC programmes.

All this while, Intel has been more than willing to accommodate feature demands from large OEMs.

It'll be interesting to see how all of this works out.
Sign In