Arthur Cole spoke with Jarrod Siket, senior vice president of sales and marketing, Netronome.

Cole: The concept of network flow processing is fairly new in the data center. What exactly does it do?
Siket: As today’s data center and network converge to form a unified computing environment, requirements such as deep packet inspection, security processing, support for multi-core processors and virtualization, increased performance and green computing will have to be addressed and delivered. Eventually, every packet from every flow of communications services will need to be intelligently processed. A Network Flow Processor (NFP) is the common denominator that is capable of providing a high-speed, programmable data-plane with integrated, deep packet inspection, security and I/O virtualization for the servers, appliances and network elements that make up these designs, in both enterprise and carrier networks.

Cole: How is your device implemented? On the motherboard, or does it need to be integrated directly into a multicore chipset?
Siket: The NFP-32xx family of Network Flow Processors will be made available to customers as a processor and as a board solution. Customers interested in entering the market quickly will likely make use of an NFP-32xx based PCIe add-in card that can be acquired from Netronome and connected directly to an x86-based motherboard. Other customers will use the processor in custom hardware designs. In either case, both solutions are highly optimized to act as a network-coprocessor with multicore CPUs.

Cole: Are there any deployment issues so far? Does it need to sync up to virtual I/O or cloud-enabling technology to keep tabs on shifting network patterns?
Siket: The NFP-32xx is in fact the technology that provides the high-performance virtual I/O. In these designs, the flow processor provides a high-speed network interface on one side, then a virtualized IOV interface to the system, with programmable DPI, packet capture, load balancing and zero-copy drivers to the multicore CPUs. These capabilities ensure that network I/O is no longer the bottleneck of virtualized servers and appliances, allowing the multicore CPUs to realize maximum performance and application processing by offloading burdensome networking tasks.