Remember Netronome, the startup with roots in Fore Systems and other legendary networking companies, that licensed the Intel IXP family a couple years ago, at a time when traditional network processors were being disowned and spoofed to no end?  Well, who’s laughing now?  Netronome is launching its highly-parallel datapath Network Flow Processor this week, and betting that three discrete markets still need a heavy-lifting co-processor for handling packets.  Why should we pay attention to the NFP-32xx?

Try 40 separate RISC cores at 1.4 GHz per core.  Try hardwired engines for cryptography, PKI (Public Key Infrastructure), and content-addressable memory.  Try Interlaken and PCI Express Gen 2 interfaces optimized for server virtualization.

Pay attention to that last factor.  Netronome’s senior vice president of marketing and sales, Jarrod Siket, admits that if his company had only paid attention to core switches and routers, and put all of its energies into the March 23 Optical Fibers in Communications (OFC) show, Netronome would be reduced to knocking at the door of less than five OEMs who are still important in this space.  But the NFP designers realized during development that the processor could be ideal for network-edge Unified Threat Management platforms, which perform seven-layer deep packet inspection; and for data servers moving to virtualization, even as their line cards support 40G and 100G Ethernet.

Netronome got its foot in the door with the licensed Intel processors, allowing OEMs to become familiar with the company’s ability to design silicon and comprehend market needs.  Now it needs to show that NFP’s security features can meet those of Cavium, RMI/NetLogic, and Freescale, while it offers higher integration and lower power dissipation than ASICs.  When I asked Siket about the FPGA vendors showing up at OFC and RSA shows with these kind of architectures, he said he appreciates the new parallel capabilities of the latest FPGA architectures, but has yet to see a single device compete in high-speed network processing on the data plane.

If Netronome had not adopted a three-way end market philosophy, I might have questioned the company’s ability to find a customer base.  But some interesting datapath processor architectures, like that of C-Port’s, disappeared during the telecom collapse, largely because the software and support necessary for the architectures were unsustainable when the communication industry was in freefall.  Today, Netronome might have found a window in which it can insert a new datapath architecture for 40G and 100G networks.