x
Optical/IP

Avici, Riverstone Pick Processors

At the most recent Network Processors Conference, last October, keynote speaker Doug Spreng noted that 2003 could be a brutal year (see Net Processors Brace for Shakeout). And as expected, some shutdowns have already begun. The flip side was that network processors were finally gaining some respect and even design wins inside production-worthy boxes.

Going into next week's NPC, being held in Boston, some positive news is starting to appear. Avici Systems Inc. (Nasdaq: AVCI; Frankfurt: BVC7) recently announced it's using an Intel Corp. (Nasdaq: INTC) network processor; and Riverstone Networks Inc. (Nasdaq: RSTN) is readying a metro core box using chips from startup Sandburst Corp..

The wins are key, as many of the network-processor startups are running out of time (see Fast-Chip Flees the Market and Cogni-Gone?). With plenty of doubt still lingering about startups, that first customer win is an all-important piece of evidence that the chip maker might still be around later.

"It really depends on the wins you catch in this first wave," says Vince Graziani, Sandburst CEO. "Every win we get, we throw everything at it."

The downturn has played into chip makers' hands by gutting the R&D staffs of many systems companies. Left without the engineering teams needed to design an ASIC, OEMs are turning to off-the-shelf chips, including network processors.

That's the theory, anyway, but chip vendors have relatively few announced design wins to show for it. The claim has been that OEMs don't want to disclose which network processor they're using -- or in some cases don't even want to fess up to using one -- because they consider the chips to be a competitive advantage.

Still, some vendors are beginning to talk. Avici recently announced a multiservice blade based on the IXP2400 from Intel (see Avici Intros Multiservice Line Cards). Among the deciding factors for Avici was that network processors finally had the performance to be viable in a core router.

"In the past there were a lot of promises made, and the technology just wasn't mature enough," says Esmeralda Swartz, Avici vice president of marketing. "We wanted to make sure it could support line rate when advanced features [such as policing and shaping] were turned on."

Another deciding factor was the work Intel had put into its tools for programming the network processor's software. "Intel had a very rich and mature development infrastructure. The SDK [software developers' kit] was very critical to us."

Intel is an established name in the relatively young network processor market. By contrast, Riverstone chose to go with newcomer Sandburst, whose four-chip set includes a network processor, a switch fabric, and an arbiter that handles traffic management.

Sandburst got its first chips back from the fab in the first week of October, and Riverstone had its box running by December. "They were in full production by the end of January, and by early February they were in a live system," says Sandburst's Graziani. (If that seems fast, keep in mind that Riverstone probably got a headstart. Most processor vendors offer simulation environments that let customers design their systems before having the actual chips in hand. Riverstone couldn't be reached for comment.)

Riverstone's first box using Sandburst is aimed at the metro core. Having seen that the chips work, the company is considering using them in lower-end systems as well, Graziani says.

Sandburst's big hope is that the Riverstone win can open doors to other customers -- particularly those hoping to compete with the Catalyst 6509 from Cisco Systems Inc. (Nasdaq: CSCO), according to Graziani. He's particularly hoping to catch the eye of prospects such as Extreme Networks Inc. (Nasdaq: EXTR) and Foundry Networks Inc. (Nasdaq: FDRY). "This is the first generation of product at that high end where [those companies are] considering merchants."

— Craig Matsumoto, Senior Editor, Light Reading

Page 1 / 6   >   >>
chipsischips 12/4/2012 | 11:52:36 PM
re: Avici, Riverstone Pick Processors The chance of winning a major design at Cisco, Foundry or Extreme is so extremely low for any of these start-ups. These three companies pride themselves on their own in-house ASICs. The start-ups would be better off chasing 3Com/Huawei or working with D-Link, Netgear & Accton to enable them to build boxes further up the foodchain.

Of course, Intel will win a bunch of designs, but Intel is learning the hard way that this market is just not very big and will probably never justify the huge investment they've made.
null0 12/4/2012 | 11:52:35 PM
re: Avici, Riverstone Pick Processors Interesting point.
================================================
Left without the engineering teams needed to design an ASIC, OEMs are turning to off-the-shelf chips, including network processors.
================================================

or maybe?
Without a team to develop a network processor they either had to stay with ASIC's or opt for an off the shelf NPU that they can customize to suit their customers needs?

Point being, which is best ASIC or NPU and why?

Null0
BobbyMax 12/4/2012 | 11:52:34 PM
re: Avici, Riverstone Pick Processors There are certain cost advantages to using NPUs rather than Asics. About 2-3 years ago a lot of vendors such as Vittesse and AMCC lost a lot of money on usinf NPUs. Vitesse lose a lot of money when it acquired Xaqti and Sitera. NPUs are used in rare instances and that is why the NPU market has disappeared
Jutebox 12/4/2012 | 11:52:34 PM
re: Avici, Riverstone Pick Processors NPUs are easier to develop with than ASICs. While development is required with either approach, NPUs are programmed by embedded SW engineers, which are more common than ASIC designers, require less expensive tools, etc.

But the key difference is that with ASICs, the OEM pays the $1M mask charge(s), adding directly to the program's NRE. With NPUs, the NPU vendor pays and the OEM is not exposed to that liability.

Additionally, NPUs provide programmability that allows OEMs to adapt to changing market needs in a way that ASICs can never do.
skeptic 12/4/2012 | 11:52:31 PM
re: Avici, Riverstone Pick Processors Point being, which is best ASIC or NPU and why?
----------
It depends.

The problem with ASIC these days is that the
costs of spinning your own (and maintaining it)
are huge.

If its non-programmable, you end up with re-spins
and re-designs to add functionality. If its
programmable, you are essentially doing something
close to a custom CPU which (these days) requires
a whole lot of infrastructure. And without
lots of volume, its really difficult to support
that kind of effort.

And its very easy to get a ASIC for ip forwarding
"wrong" in an expensive way.

The problem with NPU is that you have to work
with a vendor you can trust. You will also
pay a higher per-unit cost (but as it avoids
the up-front costs of ASICs, its a good tradeoff
where its important to conserve cash). Its also
the case that the NPU never turns out to be
as good as it seems.

alchemy 12/4/2012 | 11:52:31 PM
re: Avici, Riverstone Pick Processors The day the IETF, IEEE and ITU are disbanded, I'll consider replacing an NPU with an ASIC.

I think the real issue with NPUs is that the tools chain, libraries, and support for most are awful. Most people doing the selection are more concerned with the whiz-bang features of the device rather than trying to optimize for time to market. Intel isn't typically an NPU vendor I think of when I think of good tools and good tech support. I'd also be very concerned with picking an NPU from a startup.
lift 12/4/2012 | 11:52:29 PM
re: Avici, Riverstone Pick Processors ....I think the real issue with NPUs is that the tools chain, libraries, and support for most are awful. Most people doing the selection are more concerned with the whiz-bang features of the device rather than trying to optimize for time to market.

=================================================

In addition to the difficulty of programming tools, the performance of NPUs has been very disappointing. This was found very late in the
development cycle - it was too late to change.
mrcasual 12/4/2012 | 11:52:28 PM
re: Avici, Riverstone Pick Processors In addition to the difficulty of programming tools, the performance of NPUs has been very disappointing. This was found very late in the
development cycle - it was too late to change.


If you choose an NPU without knowing EXACTLY what it's performance characteristics you have not done your homework.

No solution, ASIC or NPU, is "perfect". They all have some form of tradeoff. You need to understand them and make a decsision about which ones matter and which ones don't.

If your vendor is reputable you shouldn't get any surprises.
Lightmare 12/4/2012 | 11:52:27 PM
re: Avici, Riverstone Pick Processors http://www.sandburst.com/leade...

Piyush Patel President, Chairman of the Board of Riverstone Networks - Piyush Patel is currently chairman of the board of Riverstone Networks and former chairman of the board and CEO of Cabletron Systems.
drone387 12/4/2012 | 11:52:27 PM
re: Avici, Riverstone Pick Processors "If your vendor is reputable you shouldn't get any surprises. "

Your vendor is desperate for sales and will/won't tell you whatever to close the deal. It's like's it's always been: Buyer Beware! Your first statement, "you have not done your homework.", was correct.
Page 1 / 6   >   >>
HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE