Comms chips

EZchip Redoes It

EZchip Technologies, which is developing silicon for packet processing at OC192 (10 Gbit/s) rates, has been tinkering with its product roadmap again.

Today, the startup announced that it will be adding a traffic manager to its chip set (see EZchip Unveils OC192 Manager). QX-1 is a hardwired chip that offloads advanced scheduling functions from a network processor. Put simply, its task is to choose the data with the highest priority and make sure it gets through.

According to EZchip's CEO Eli Fruchter, its NP-1 network processor will ship "very soon." The company decided, therefore, to unveil the traffic manager now, some nine months or so ahead of time, so that it can talk about both products together.

The announcement amounts to an admission that EZchip didn't get its product design quite right to start with.

"The original plan was that they would have one chip and not have a traffic manager at all," says Linley Gwennap, a senior analyst with the Linley Group, who has been following the startup closely.

EZchip also promised to deliver its network processor in February 2001 -- something that clearly did not happen. Gwennap blames the delays on constantly moving goalposts.

"Originally EZchip was working closely with just a few customers, particularly Avaya Inc. [NYSE: AV]," he says (see Avaya Chooses EZchip Chip). "As it started talking to other companies to see what their requirements were, it realized how specialized the design was. It started adding more features and interfaces to make the chip more broadly suitable, but every time it went back to the design, it added a few months."

EZchip's early customers had their own traffic manager chips in-house, so there was no need for the startup to develop them initially, he adds.

However, even after all the setbacks, Gwennap still thinks that EZchip may be first to market with an OC192 network processor. "That's the good news."

He's also positive on NP-1's architecture -- a "pipelined" system in which different microprocessor cores are allocated different functions, such as parsing (classifying), searching, forwarding, and modifying. In concept it is similar to that being developed by Xelerated Packet Devices AB (see Swedes Claim Processor Advance).

The bad news is that it's not possible to use EZchip's network processor in any serious networking applications without a traffic manager, Gwennap says. Since the QX-1 will not follow until the middle of next year, customers without their own traffic manager solution will still have to wait.

By the time the QX-1 appears, there could be quite a few other OC192 packet processing solutions available. Applied Micro Circuits Corp. (AMCC) (Nasdaq: AMCC) is already shipping a traffic manager and is only weeks away from shipping its network processor, according to Robin Melnick, AMCC's product marketing manager (see AMCC Introduces Network Processors). And Agere Systems (NYSE: AGR) is due to start shipping both chips in Q1 next year (see Agere Announces Products, Contracts).

Then there are a bunch of startups, such as Internet Machines Corp. and Cognigine Corp., which may not be far behind (see Internet Machines Takes Aim at Zettacom and Startup Spins Novel Network Processor).

— Pauline Rigby, Senior Editor, Light Reading
andrewkim 12/4/2012 | 7:41:35 PM
re: EZchip Redoes It i read somewhere that ibm has actually invested some money in ezchip. ibm also has an oc192 np. could ibm be oem'ing ezchip's oc192 np by any chance? why would ibm be investing in ezchip then if they both have oc192 offerings? i don't think ibm would be investing in ezchip if ezchip is going to go head to head with ibm. this is my two cents.

any word on where ibm is with the rainier chip? i hear that they're late. is it any good compared to motorola's c-port?
pablo 12/4/2012 | 7:41:30 PM
re: EZchip Redoes It Most large NP vendors make their money with OC-48 volumes, and regard the OC-192 market as too far out and somewhat crowded, it seems. I think whenever the Intel-IBM-Motorola-etc really want a piece of the OC-192 pie, they'll do it. Throwing RISC cores at the OC-192 NPU "problem" doesn't seem enough of a core expetise to defend that market from the eventual onslaught, and the most critical software will be developed by the network system vendors themselves. The OC-192 NPU market is merely an acquisition game: be first and prove it works, and you'll get acquired. Inconvenient that - last year all you would have had to do was *claim* you were going to have a working device real soon.
Sign In