Editor’s note: This post is the first in a new “member spotlight” series that will highlight the P4-related activities of various P4 Language Consortium members. It is based on an interview with Gordon Brebner, who was recently awarded the inaugural P4.org Service Award for his contributions to the community as inaugural co-chair of the P4 Language Design Working Group.
Please tell us a bit about Xilinx.
Xilinx pioneered Field Programmable Gate Array (FPGA) technology, and is the current market leader. As FPGA has evolved, with the embedding and integration of other programmable functions, Xilinx has recently revealed itself as pioneering a new category: Adaptive Compute Acceleration Platform (ACAP) technology. The important words here are “programmable”, “adaptive”, and “acceleration”, all key attributes of Xilinx products. This resonates with increasing programmability of networks, at increasingly high speeds.
What was Xilinx’s motivation for joining the P4 Language Consortium?
Xilinx was one of the first companies to join P4.org. Xilinx already had its SDNet programming environment (launched as a product in early 2014), for compiling the Xilinx Labs-created PX language to FPGA. PX had a similar level of abstraction to P4 and indeed a considerable intersection with P4. Xilinx felt that the value of SDNet would be increased if P4 was to become a standard packet processing language and Xilinx adopted it, initially alongside PX and ultimately to replace PX. Xilinx also felt that its past experience with PX, including customer feedback, would be of benefit to active participation in defining and working on the new P4.org activities.
What are some of the benefits you see for increased network programmability?
Xilinx’s customer base is primarily in communications (wired and wireless) and data centers. In these areas, there is an increasing customer demand for network programmability, whether on the line card, in the NIC or SmartNIC, or on the network switch. With SDNet, Xilinx is catering for this demand at the packet forwarding and switching level. Further, especially in NFV appliances and in network-attached data center functions, this programmability is coupled with acceleration of programmable L4-L7 and payload processing in the data fast path. Beyond the conventional networking markets, Xilinx also sees interest in programmable networking more broadly: for example, in aerospace, defense, and industrial, markets. Here, customization of protocols is often important.
How is Xilinx involved with P4 community efforts?
Xilinx has been involved in P4 community efforts from the very beginning. This started with a presentation and a 100Gb/s rate demonstration at the first P4 Workshop in June 2015, and has continued at the subsequent events. As a leader and pioneer with P4, Xilinx has contributed to the Developer Days, as well as tutorials at other high-profile conferences for differing types of experts, such as FPGA, Hot Chips, and SIGCOMM. Together with Stanford and Cambridge Universities, Xilinx has founded the P4->NetFPGA community, centered around providing a “big green button” tool flow for compiling P4 to the NetFPGA SUME platform much used by networking researchers working on high line-rate systems. Xilinx has been part of the P4 Language Design Working Group since its inception, indeed co-chairing it, and has been following three other P4 Working Groups since their establishment. In particular, Xilinx recently showcased an Inband Network Telemetry (INT) inter-operability demonstration with Barefoot Networks at the MWC and OFC conferences, based on the draft INT specification from the P4 Applications Working Group, which Xilinx has also made contributions to.
Does Xilinx have any products based on P4?
The May 2017 release of SDNet included support for P416, which had just been released. Various improvements have been made in subsequent releases. For example, Xilinx has recently demonstrated a prototype Xilinx SmartNIC platform—comprising NIC card with FPGA, P4 programming environment, and DPDK runtime software—developed in Xilinx Labs. This is being further developed by Xilinx as a new board-level product, to be launched in 2019.