x
ATCA/Standard Servers

ATCA Ramps 40G Generation

The 40-Gbit/s generation of ATCA chassis is rolling out now, with Long Term Evolution (LTE) a key market driver, according to systems integrators in the space.

That made wireless networks a key focus during Light Reading's xTCA & COTS Virtual Event held online Thursday.

It's not so radical an idea; vendors including ALU, NEC, and ZTE have already announced Evolved Packet Cores built from ATCA-based gear. Still, with LTE expected to create a crushing wave of data demand during the next two years, it made for a nice backdrop to a keynote from Anthony Ambrose, a vice president at Radisys Corp. (Nasdaq: RSYS).

Carriers are going to need frequent upgrades to the network, and vendors can accommodate that through the speed of ATCA-based development -- especially if another suddenly popular product like an iPhone or iPad hits the market, he said. "If you can't develop your product in less than one year, you're going to be behind people in the industry that will have this kind of capability."

The use 40-Gbit/s slots, which adds up to chassis capacity of 560 Gbit/s, could provide a bandwidth boost to help accommodate LTE needs. The technology marks the fourth generation of ATCA, following the third generation and its 10-Gbit/s slots, which came out in 2008. "We've seen most companies shift over to these third-generation platforms," said Simon Stanley of Earlswood Marketing Ltd. , who moderated every talk at the virtual conference. (See Who Makes What: ATCA, AMC & MicroTCA.)

Even though the equipment is available -- RadiSys and Emerson Electric Co. announced 40-Gbit/s platforms in October -- the PCI Industrial Computer Manufacturers Group (PICMG) standard for a 40-Gbit/s platform isn't yet complete. (PICMG is the industry group that handles ATCA standards.) (See RadiSys Launches 40-Gig ATCA and Emerson Sports 40GE ATCA .)

PICMG had hoped to ratify the standard by April but missed. Still, in a chat session in the virtual tradeshow's virtual lounge (BYOB), representatives from Continuous Computing Corp. , Kontron AG , and RadiSys agreed that the technology is in a stable enough state for products to start coming out.

Getting 40 Gbit/s per slot will almost always require a new chassis, though, because the backplane has to accommodate those speeds. In the case of ATCA, with its Ethernet backplane, that means supporting the 10GBASE-KR and 40GBASE-KR4 standards for 10- and 40-Gbit/s backplanes.

Beyond the wireless market, Ambrose thinks 40-Gbit/s optical gear might make use of ATCA as well.

"With ATCA 40 Gbit/s, I think you will see more systems crossing over to open modular platforms," Ambrose "said" in the online chat lounge. But 100 Gbit/s interfaces are "a bridge too far right now," so that equipment "will stay proprietary in this generation," he said.

An archive of the virtual event will be available starting May 11, on Light Reading's Events page.

— Craig Matsumoto, West Coast Editor, Light Reading

Pete Baldwin 12/5/2012 | 4:36:54 PM
re: ATCA Ramps 40G Generation

Interesting poll result from late in the conference: Asked what the main reason for using ATCA would be, 17% of the audience cited 40G support.


The top answers were the surrounding COTS ecosystem (33%) the fact that ATCA speeds up development (22%), and the lower costs (17%).  I'd have expected those threereasons to make up more like 95% of the responses; it was interesting to see that a reasonable number of attendees cared that much about 40G capability.


ATCA still has its handicaps -- power limitations being one.  For those who've passed on using it, I'd be interested in hearing why.

goundan 12/5/2012 | 4:36:53 PM
re: ATCA Ramps 40G Generation

Craig,


Those are BS reasons, unless someone is planning on less than 50 chassis per year and are dealing with relatively low volume control plane intensive data through the system. At 50 chassis and above, depending on the product requirements, customs jobs are cheaper by a factor of 2x to 4x compared to ATCA.


1. 40Gbps - Do they mean real 40Gbps or just the fact that 40Gbps Ethernet means they can run around 16Gbps effective throughtput per slot? Nevermind the aggregation required to bring traffic in and out of the chassis.


2. Where is end-to-end QoS in this story? If you dealing with the likes of SGSNs/MMEs or enterprise firewalls, I would say it makes sense. If you dealing with PDN GWs or any kind of BRAS/Edge Routers, I'd say fuggedaboutit.


3. Speed of development again depends on the kind of equipment the vendor is trying to build. For contol plane processing intensive jobs this is true. For data plane or 50/50 comtrol/data plane jobs, this is not true given the nature of 3rd party boards available and the kludges that one has to implement to get it to work right.


3. Lower costs again is applicable only for control plane intensive jobs. Does not hold water for others. Even for control plane intensive work, blade servers are cheaper by 2x, and are much better put together from a HW/SW integration perspective. For control plane intensive work, look at IBM blade center as a cost point against ATCA before you jump into ATCA. Even with single vendor lock-in, it is still cheaper.


4. ATCA ecosystem - more joke than reality once you step out of control plane intensive requirements. The ATCA crowd needs to get their act toegther on this one. In the last 3 years, I have yet to come across a decently put together board or system or ecosystem for anything other than x86 boards. Let the ATCA vendors create a decent bonding driver, before they talk of their ecosystem (and yes, that goes for all the 4 big boys of the ATCA world).


5. Every ATCA system requires customization to convert it into a real product. The problem is that these customizations are not portable across vendors, creating single vendor lock-in issues. So what exactly is the point of a ecosystem that works only at pointy haired manager power point level?


6. Power and board space limitations - lets not even go there please.

Pete Baldwin 12/5/2012 | 4:36:49 PM
re: ATCA Ramps 40G Generation

> At 50 chassis and above, depending on the product requirements, customs jobs are cheaper by a factor of 2x to 4x compared to ATCA.


Thanks for the comments, goundan.  Great input. 


#3 and #6 are familiar comments ... regarding #4, I keep hearing a lot about Cavium support, but maybe it's not at the true ecosystem level you're talking about?


#5 is particularly interesting, though - the fact that ATCA projects still use custom assistance. That would make sense; I remember one company (in pre-ATCA years) showing me how many FPGAs got slapped onto boards as last-minute fixes and kludges. This would seem to be an extension of the same principle. 

Pete Baldwin 12/5/2012 | 4:36:48 PM
re: ATCA Ramps 40G Generation

'Course I should add that some OEMs, AlcaLu in particular, seem to be getting a lot of mileage out of ATCA. But that could be just for specific types of gear - I haven't looked much into how broadly they use ATCA.

goundan 12/5/2012 | 4:36:39 PM
re: ATCA Ramps 40G Generation

Cavium/Raza XLR work fine as chips by themselves. Where it goes wrong is how the boards are designed by ATCA vendors who seem to have no idea of how to design boards that operate at the system level. Most boards are designed to operate for primarily security configurations and not real life data plane work. So any system vendor using these boards has to implement enormous software kludges to get to it to work about half way right. The throughput of the boards themselves (based on multicore RISC) isn't much to talk about  given the handicaps the chips have been subjected to. (Sorry, can't go too much into the details on these boards given disclosure agreements).


The nature of custom assistance in ATCA projects has been mostly limited to software only workarounds primarily by restricting the choice of boards/AMCs to specific vendors/models and then implementing software workarounds for the specific problems posed by the combinations chosen. Change any one item in the combination and it gets very expensive soon.


 


Craig wrote:


"regarding #4, I keep hearing a lot about Cavium support, but maybe it's not at the true ecosystem level you're talking about?



#5 is particularly interesting, though - the fact that ATCA projects still use custom assistance. That would make sense; I remember one company (in pre-ATCA years) showing me how many FPGAs got slapped onto boards as last-minute fixes and kludges. This would seem to be an extension of the same principle. "


HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE