x
Comms chips

Vitesse Pumps Up its Net Processor

Today chip maker Vitesse Semiconductor Corp. (Nasdaq: VTSS) unveiled its roadmap towards higher-performance network processors -- chips that sit inside switches and routers, reading the headers of incoming data packets to figure out how they should be handled (see Vitesse Details Net Processor Roadmap).

The vendor pre-announced an upgrade of its existing product, the IQ2000, that will allow it to carry out twice as many instructions per second (XIPS) as the old version. It also outlined plans to offer chips capable of 20 times the number of XIPS, which will achieve OC192 (10 Gbit/s) speeds.

But potential customers should be aware that none of the new chips will start sampling until next year, with OC192 chips slated for mid-2002 at the earliest.

So far, the network processor market has been characterized by vendors claiming a lot and delivering little. But in Vitesse's case, its seems reasonable to believe that it can meet its promises. Here's why.

For starters, Vitesse was early to the market, thanks to its $750 million purchase of Sitera Corp. last year. It inherited the startup's IQ2000 processor, which offers dual gigabit Ethernet processing capabilities.

The IQ2000 has notched up a considerable number of design wins, including iPolicy Networks, and Quarry Technologies Inc., and is now shipping in volume, according to Stephen Bassett, senior product manager, Advanced Network Products division at Vitesse

Analysts say Vitesse's strength lies in the fact that its architecture is already proven. "Vitesse may not be the technology leaders, but they've got a solid architecture and they've got it out there," says Russell Johnson, director of switching and routing semiconductors with RHK Inc.'s ChipWatch group.

Vitesse's new generation of network processor chips will use an extended version of the architecture in the IQ2000.

The upgrade, called the IQ2200, will have the same speed as the IQ2000 but will have more headroom for doing "deep packet lookups" to get higher-level information for billing and other purposes. It achieves this via a doubling of the clock rate on the chip, from 200 to 400 MHz.

Vitesse is also working on a chip that is nominally an OC192 (10 Gbit/s) solution, although it actually handles two streams of bidirectional OC48 traffic. To get real OC192 performance would entail doubling up the chip and interconnecting the pair with a switch chip that Vitesse calls a "Focus Connect."

Dubbed the IQ10GS, the 10 Gbit/s chip will contain "more than 10" RISC processor cores, making it a fairly complicated beast, and is expected to be able to handle 20 times more XIPS than its little brother, which contains only 4 processor cores. It's been under development for about a year.

The big advantage of buying from a vendor with an established product line, rather than a startup, is that existing customers "will be able to port all their old software over to the new chips," according to Bassett.

Why announce now, if the products aren't ready? Vitesse says the time is right to put itself on the network processor map. "We've spent the last few years educating the public about what a network processor is," says Bassett. "We're through that chasm now, and the opportunity is huge going forward."

Many startups are hyping their next-generation network processor developments, and Vitesse doesn't want to be overlooked. Some, like Cognigine Corp. and ZettaCom Inc., are intending to start shipping samples of full-duplex OC192 chips early next year (see Startup Spins Novel Network Processor and Zettacom Set to Score $47.5M). Others, like Xelerated Packet Devices AB, are jumping straight in with OC768 (40 Gbit/s) chips, which they claim will be ready about a year from now (see Swedes Claim Processor Advance).

And of course there are the established vendors of gigabit Ethernet and OC48 network processors to take into consideration, notably Agere Systems (NYSE: AGR), Applied Micro Circuits Corp. (AMCC) (Nasdaq: AMCC), and IBM Corp. (NYSE: IBM).

— Pauline Rigby, Senior Editor, Light Reading
http://www.lightreading.com
jrj 12/4/2012 | 7:52:18 PM
re: Vitesse Pumps Up its Net Processor How come network processors claims zillions of instructions per second but hardly talk about the lookup rate?

Seems like they havenGÇÖt grasped what sets the limit in packet processing?

-jrj
skeptic 12/4/2012 | 7:41:29 PM
re: Vitesse Pumps Up its Net Processor How come network processors claims zillions of instructions per second but hardly talk about the lookup rate?
----------------
Because they are programmable. There are multiple
options for lookup algorithms in a fully
programmable part and lots of different
tradeoffs to make.

Lookup rate matters in fixed ASICs and CAMS,
but its much more subjective in the
programmable versions of these devices.

pablo 12/4/2012 | 7:41:29 PM
re: Vitesse Pumps Up its Net Processor
They all do document the lookup rate, i.e. how wide the search is at a certain rate. But most publications do not publish it, since it is perceived as a more ethereal parameter.
HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE