Comms chips

OC192 Processors: Who's First?

All the signs are that systems integrators will soon be able to get their hands on the first OC192 (10 Gbit/s) network processors.

At the Network Processors Conference in San Jose, Calif., which started today, Applied Micro Circuits Corp. (AMCC) (Nasdaq: AMCC) announced the commercial availability of its nP7510 network processor (see AMCC Adds to Chip Family).

Also, EZchip Technologies says it will be shipping its NP-1 network processor "very soon" (see EZchip Redoes It ).

Both companies separately confided to Light Reading that "according to Linley Gwennap [analyst with the Linley Group and program chair at the conference], we're going to be first on the market."

Gwennap concurs. "AMCC and EZchip seem to be neck and neck," he says. Who is actually first may be too close to call.

And about time too. It's taken so long for OC192 network processors to get close to being a commercial reality that some systems integrators have decided that the only way to get packet processing silicon fast enough for their needs was to develop their own ASICs (application-specific integrated circuits).

"When we began development of our system two years ago, there obviously weren't any OC192-capable network processors available or even on the roadmap, so we built our own," Doug Broad, public relations manager at Celox Networks wrote in an email to Light Reading. He claims that Celox already has an internally-developed OC192 network processor, codenamed BitRipper, up and working.

"BitRipper is currently passing packets inside our Celox SCx 192 IP service switch in AT&T's lab," says Broad. "We've got four other trials starting this month and expect general availability [of the switch] in Q4."

However, Celox's answer to OC192 packet processing looks a bit different from the single-chip solutions being developed by AMCC and EZchip. According to information on the company's Website, each card in the Celox platform uses a combination of BitRipper ASICs and off-the-shelf PowerPC microprocessors.

In some ways, its not surprising that it's taken so long to develop OC192 network processors. OC48 network processors are still in their infancy, and many can't maintain wirespeed when required to perform more functions per packet. Given that systems vendors want more functions per packet, and there's a fourfold jump in speed on top of that, the required performance increase is eightfold or more. In the PC world, CPU designers are lucky to double performance when they develop their next-generation devices.

In many cases, the required leap in performance has forced chip makers, both incumbents and startups, to go back to their drawing boards and come up with completely new architectures.

AMCC is particularly pleased that it's managed to scale its architecture up to OC192 data rates, says Robin Melnick, a director of product marketing at the company. "As a result, we'll be the first of the incumbents to get a 10-gig chip out the door," he contends. Some of the incumbents are behind AMCC by as much as 18 months because their designs didn't scale, he claims.

One chip vendor that isn't far behind is Agere Systems (NYSE: AGR), which expects to start sampling its OC192 network processor in Q1 2002, according to Robert Munoz, the company's product marketing manager for network processors. Agere is due to present the design of its NP10/TM10 chipset at the conference tomorrow.

A couple of startups also had announcements. Entridia Corp. introduced the "Rhapsody," actually an ASIC, rather than a software-programmable chip, but which nevertheless can handle a wide range of packet processing functions at full-duplex OC192 speeds (see Entridia Offers 'Rhapsody'). Rhapsody is also due to start sampling in Q1 2001.

In addition, Silicon Access Networks Inc. unveiled a packet processor that, rather unusually, is capable of full-duplex data rates of 20 Gbit/s (see Silicon Access Processes Packets). It's also claiming that it will start shipping the product in Q1 2002.

Other companies that did not make formal announcements but are rumored to be nearly ready with OC192 are rival startups Internet Machines Corp. and ZettaCom Inc. (see Internet Machines Takes Aim at Zettacom).

Even though there won't be much volume in OC192 packet processing for a while yet, vendors still feel compelled to show how they are going to reach OC192. Motorola Inc. (NYSE: MOT) and Vitesse Semiconductor Corp. (Nasdaq: VTSS) both shared their roadmaps at the show (see Motorola Has a Roadmap and Vitesse Gets Technical).

But AMCC's Melnick points out that it's not enough to be first with a network processor. "What matters most to customers is not a point product, but the whole product family," he says. Customers will want to buy not just a network processor, but a traffic manager and a switch fabric as well. That puts established companies like AMCC and Agere at an advantage, since they have a broader portfolio. It also complicates life for startups, which need to develop two or three complicated chips simultaneously, he says.

— Pauline Rigby, Senior Editor, Light Reading
Page 1 / 4   >   >>
hitecheer 12/4/2012 | 7:41:13 PM
re: OC192 Processors: Who's First? After burning millions of cash and spending years of time, these startups must be frustrated to see such low market demand for 192 NPs and intensive competition . To make things worse, big players like Cisco don't even consider using NPs designed by someone else. I'll see most of NP startups belly up before the mid of '02.
mcasaes 12/4/2012 | 7:41:12 PM
re: OC192 Processors: Who's First? Cisco used to have MMC Networks gear on their portfolio (Catalyst 85xx).
uspaltho 12/4/2012 | 7:41:09 PM
re: OC192 Processors: Who's First? As far as I have read:
In July 2001, Cisco announced to use Motorola's MPC7450 PowerPC in its next-generation high-end routers . In addition, Cisco and Motorola are developing a fourth-generation multiple processor chip set for even higher performance and scalability.
Currently Cisco pursues internal Network Processor R&D in cooperation with Tioga and in addition uses various other external Network Processors, from AMCC/MMC, Intel, Broadcom, PMC Cierra.
lostinlight 12/4/2012 | 7:41:02 PM
re: OC192 Processors: Who's First? Whats the deal with network processors? How will the few router companies differentiate their
products if they all base their products on network processors? Will routers become commodity like PCs? I dont see such a huge market for high end routers that they become differentiated by price alone!
mcasaes 12/4/2012 | 7:41:00 PM
re: OC192 Processors: Who's First? From what I understand:
1- Cisco NP is the PXF, I don't think they ever used the term NPs on regard to the Engine'n' ASICs on GSR (maybe I've missed it ?)
2- MPC7450 is not a NP.
hitecheer 12/4/2012 | 7:40:54 PM
re: OC192 Processors: Who's First? One of the stories I heard is that off shelf NPs will enable system companies ("God box" builders) to use the chipset available to them without wasting significent amount of resources on ASIC developement, assuming these off shelf NPs work. So far, however, it is just too risky to trust the design from these 1 or 2 years old NP startups. I heard customers who tried SiByte (Broadcom) NPs, their comments are basically "SiByte stuff is a joke". A bounch Silicon Access designers I know who have very limited design experience, are doing core design of their chipset. No wonder system company are very reluctant to give up their own in house ASIC development effort.
green 12/4/2012 | 7:40:53 PM
re: OC192 Processors: Who's First? system integrators don't have to worry about building the basic building blocks themselves. however before that happens the NP vendors should come up with a common interface so that the system vendors are not tied down to one NP vendor (C-PIX anyone?). NP are just the bricks in a system. how you arrange your bricks to build a system will determine the scalability, cost etc., the flip side is that individual companies don't have the control of the product roadmap.

I don't agree with the article's contention that there is no market for 10gig speeds. the current generation of NP just don't have enough juice to do all the things that the market is asking for. that's the reason Cisco and other are developing their own ASICs..
ackronym 12/4/2012 | 7:40:51 PM
re: OC192 Processors: Who's First? For the record, AMCC had 180 design wins this qtr for OC192 to 75 different customers.
asmo 12/4/2012 | 7:40:50 PM
re: OC192 Processors: Who's First? Sure I can see NP vendors lining up to come up with a common interface so that there customers can dump them and go with someone else.

The problem with many companies that are designing NPs is that it helps to have experience in building previous systems to know what the problems are and how to design an NP to scale well. You need an NP to do a lot more than packet forwarding these days, unless you have intimate knowledge of building previous systems and seeing the real problems that occur when the customer uses the product (i.e. not just in the lab) its hard to create a bigger next-gen NP.

The idea of companies creating NP's and them being a commodity like a PC processor is a nice idea, problem is the NPs may lag a generation or two behind the big boys designing their own silicon with experience and expertise in building whole systems.


rjmcmahon 12/4/2012 | 7:40:48 PM
re: OC192 Processors: Who's First? Cisco used to have MMC Networks gear on their portfolio (Catalyst 85xx).

The 8500 is (was)a stop gap product.

The challenge a NP supplier faces is that when the product goes volume then the box vendor will likely design out the NP (due their recurring cost). If the product stays low volume the NP supplier doesn't make any money.
Page 1 / 4   >   >>
Sign In