Xelerated Is First to 40-Gig
From the beginning, network processor startup Xelerated AB vowed it would be first to reach 40 Gbit/s, and, despite significant delays to its schedule, the company appears to have reached its target.
Xelerated has been showing working silicon in private demonstrations since last week, Light Reading has learned.
"We got first silicon back to the lab in the first week of January, and it worked so well that we decided to take it on the road in February," confirms Xelerated's VP of marketing Gary Lidington. "It's robust enough to go traveling -- that's pretty amazing for a device of this complexity."
"On paper, it always looked like a good design, but the big question was whether it would work. Now it looks as though Xelerated has achieved its mission," says Simon Stanley, principal consultant of Earlswood Marketing Ltd. and, at one stage, the director of networking products at a competitor of Xelerated (see All Change at Clearspeed ).
So, kudos to Xelerated for getting a product out of the door. However, some say that being first is easy in a niche of one. "Other vendors pursuing the 40-Gbit/s market have dropped out or re-targeted their designs for 10 Gbit/s," points out Linley Gwennap, founder of technical consultancy The Linley Group.
"Teradiant Networks Inc. is the only other vendor with firm plans to deliver a 40-Gbit/s packet processor," says Gwennap, who notes that Xelerated's network processor is programmable but Teradiant's is not (see Teradiant Turns Up).
Xelerated takes umbrage at the suggestion that it's virtually the only vendor daft enough to continue developing a 40-Gbit/s network processor. The main reason others aren't targeting higher speeds is because their architectures aren't up to the job, Lidington contends.
It's true that the telecom landscape has changed out of all recognition since Xelerated first set out, with renewed emphasis on data rates such as 2.5 and 10 Gbit/s. But that really misses the point, Lidington says: Folk are confusing the bandwidth (or throughput) of the chip with clear channel speed.
While there is little demand for applications running at port speeds of OC768 (a single 40-Gbit/s pipe), in Lidington's view there are plenty of opportunities for chips that can aggregate lower-speed channels, such as four times OC192 (10 Gbit/s), or even 20 times 1-Gigabit Ethernet.
Indeed, Xelerated's network processor really has four 10-Gbit/s interfaces. This is reflected in the name of the chip -- X10q -- where q stands for quad.
The bottom line is that higher throughput equates to a more efficient design. Customers can cram more ports onto a line card, while using up less space and consuming less power.
Lidington reckons the LAN market -- aggregation of Gigabit Ethernet in particular -- is a potential bright spot for Xelerated. "Everyone else is going after the MAN and WAN, but those aren't high volume markets," he says. "If we can break into the LAN, we hope that will allow us to achieve higher volumes."
So far, he contends, it's been really hard for traditional nework processors to penetrate the LAN market, because of their low efficiency. Basically, they consume too much power -- a price they pay for their increased flexibility.
Lidington points to Intel Corp.'s (Nasdaq: INTC) IXP2800 as an example of a fully featured network processor (see Intel: The Prince of Processors?). The IXP2800 consumes up to 30 W, he says -- a specification that's confirmed on Intel's Website. This rather high figure results from the fact that the clock speed on the IXP2800 is an impressive 1.4 GHz. More clock ticks means more packet operations per second -- in other words, good for flexibility but bad for power consumption.
Xelerated hopes to break this bottleneck. "[We have] potentially the first architecture to allow ASIC-level power consumption," Lidington claims, pointing out that the X10q has a clock speed of only 200 MHz, yet still manages to achieve four times the throughput of the Intel chip. The actual power consumption of the X10q is 11 W maximum.
It's worth pointing out that Xelerated's solution may not support the same range of functions as the IXP2800. However, there will be plenty of applications where the functions it does support -- which include Layer 4 classification, access control lists, VLAN, policing, and statistics for IPv4, IPv6, and MPLS traffic, according to the Linley Group's Gwennap -- will be adequate.
For those that want it, Xelerated also offers cheaper, lower-bandwidth versions of its network processor. The other products are called X10d and X10s, where d stands for dual, and s for single, again referring to the number of 10-Gbit/s interfaces on the chip. Its chips are sampling to two early adopter customers, the company says.
Xelerated plans to release more details of its architecture at the end of its customer tour, sometime in late March or early April.
For more on Xelerated and its high-speed history, see previous Light Reading news analyses: Xelerated Packet Devices AB, Swedes Claim Processor Advance, and Xelerated Touts 40-Gig Toolbox; as well as last August's report on Network Processors.
— Pauline Rigby, Senior Editor, Light Reading