Netronome just started shipping the NFP-3240 network flow processor, so I sat down with them to hear their pitch.  I came away convinced that the NFP-3240 is the go-to processor for network, security, and content processing to 40 and 100 Gbps.  Here’s why.

The NFP-3240 is the successor to Intel’s IXP28xx line of processor.  Intel handed the line and the technology licenses over to Netronome in 2007, and Netronome has been responsible for development of the technology ever since.  This history is a big part of why I think the NFP-3240 will succeed: As recently as a year ago, The Linley Group said that the IXP retained a 36% market share.  That’s a pretty solid foundation to build on.  The NFP-3240 takes things to the next level with a number of upgrades that bring the chip up to 40/100Gps performance levels.

As you might expect given the Intel history, the NFP-3240 is designed to work with Intel multi-core processors.  This multi-chip strategy is a good sense for high-end networking apps: Intel’s processors are hard to beat for high-performance general-purpose processing, and its Xeon line is picking up a lot of application-plane design wins.  However, Xeon not designed for high-speed data routing.  That’s where the NFP-3240 comes in—bolt it the Xeon processors and you get a “best of both worlds” solution that combines the top general-purpose processor with a networking accelerator.

This coupling is what makes Netronome stand out from its competition.  Other players like Cavium take an SoC approach where the general-purpose processors (typically MIPS) and network accelerators all live on the same chip.  This SoC approach is superior for low-to-mid-range applications, where the SoC’s size and power advantages are a major plus.  However, the MIPS processors in these devices simply can’t scale up to the level of performance offered by a Xeon.  Once you bring Xeon into the picture, the SoCs make less sense.  They just aren’t designed to interface to an off-chip multi-core processor.

In contrast, the NFP-3240 is designed specifically to interface with multi-core IA processors.  The key point to note here is that the NFP3240 makes use of PCI-SIG I/O Virtualization (IOV).  (On Intel chipsets, this is referred to as Intel VT-d.  For an intro to this and other Intel virtualization tech, check out my article on virtualization.)  The short story here is that IOV lets you avoid the sever bottlenecks that would otherwise occur when piping data between a multi-core processor (which may be running many virtual machines) and an NFP.  Only Netronome supports IOV, so it has a unique advantage in this area.

In short, I agree with The Linley Group that Netronome has a compelling offering for high-end apps, but it is not going to win any battles in the mid- to low-end.  With the markets always pushing for higher performance, that seems like a perfectly fine place for Netronome to be.