Rymden - stock.adobe.com
The rise of software-based networking, encryption and management functions can place a significant strain on the processing capacity of server CPUs. Smart network interface cards, or smartNICs, are an increasingly popular method of offloading intensive packet processing tasks from servers.
The NIC has long been used to connect servers to the Ethernet network. SmartNICs offload certain functions from the server processor, thus freeing cycles to drive application performance. SmartNICs employ specialized processors -- often field programmable gate arrays -- to power their offload capacity. These FPGAs are programmable using standard development tools.
Benefits of smartNICs
NICs have greatly increased the speed at which servers can connect to the network -- from 10 Mbps to 100 Gbps. The advent of software-based networking -- with virtual switches and Data Plane Development Kit, for example -- has placed more packet processing requirements on server CPU usage.
Typical servers might run software-defined networking, load balancing, encryption, deep packet inspection and other intensive input-output software. This packet processing requires significant CPU capacity, thus necessitating CPU cores to run networking functions instead of mission-critical application processing.
SmartNICs can be programmed to run the network software processes and free up the server processing for its primary application tasks. Additional smartNIC applications include packet capture, intrusion detection, network management, telemetry, data decompression and deduplication.
The economic premise of smartNICs is that their cost is offset by the application processing benefits of letting the server perform its primary tasks instead of spending significant time running network applications.
Market segments for smartNICs
Hyperscale cloud providers such as AWS, Google and Microsoft have used smartNIC technology for years to optimize performance of their data center servers. These cloud providers have the technical staff resources to optimize their smartNIC implementations via customized programming.
Service providers and telecom operators are investigating the potential for smartNICs to accelerate their network functions virtualization applications. A few leading-edge enterprises have also deployed smartNICs.
Customers have a number of supplier options for smartNICs, including independent suppliers and semiconductor vendors. Options vary widely in terms of processors, programmability, performance and cost.
Buyers can choose from vendors such as Achronix, AMD, Broadcom, Ethernity Networks, Intel, Marvell Technology, Napatech, Netronome and Nvidia.
The sophistication of networking, security, analytics and management software running on servers continues to grow. With increased network speeds, these tasks can require a significant percentage of server CPU capacity. SmartNICs represent a viable way to offload packet processing workloads from the primary server cores.
Organizations with large data centers should investigate the benefits of smartNIC technology. SmartNICs vary widely in terms of their architecture, capacity and cost. Considering these factors, organizations should select smartNICs to optimize the specific workload that needs to be offloaded from the server.
Editor's note: This article was originally written by Lee Doyle and updated by editors to reflect changes in the industry.