Our Technology
Extends Data-Intensive Computing into the Network
With businesses transitioning to distributed computing services in cloud data centers and an exponential growth in data whose processing must be coordinated between server processors, Dr. Dalal could foresee that these big data flows would bottleneck server processors. To address this problem, Dr. Dalal conceived of extending computing intelligence into the network for offloading, accelerating, and isolating the processing of these data-intensive workloads independent of server processors, including CPUs and GPUs. Xockets’ approach was against the conventional wisdom for how cloud data centers should be built and operate.
Implements Programmable Hardware Acceleration
Xockets’ DPUs include a virtual switch computing architecture that dynamically connects together hardware accelerators into programmable logic pipelines of accelerators for executing computing operations on packet data at line rate—the speed of the network—and with the versatility and programmability that clouds demand, including for offloading, accelerating, and isolating a cloud’s data plane and/or control plane.
Operates Independent of Server Processors
Xockets’ virtual switch computing architecture goes against the conventional wisdom in putting computing in the network because networks were designed for routing packets and are lossy—they can lose or drop data packets. As result, conventional approaches to addressing big data problems in cloud data centers were used for years until Xockets’ DPU inventions were revealed to the industry, including throwing more hardware at the problem under control of server processors, such as by adding additional server processors as well as using SmartNICs and FPGAs that are purpose-built for a specific task. These conventional approaches did not accommodate the versatility, programmability and independence required for cloud offload, and thus they did not provide the TCO savings or accelerated computing performance of Xockets’ DPU technology.
Implements Data-Centric Computing
Xockets’ virtual switch computing architecture uses the data stream itself moving between server processors for defining the programmable logic pipelines of hardware accelerators to be formed in flight in the cloud network, enabling data-centric computing for processing these data-intensive workloads instead of the application-centric computing of conventional computing architectures in server processors.
Enables an AI Future To Benefit Everyone
Dr. Dalal anticipated that big data analytics and the training of large models for ML/AI across GPUs in cloud servers would compound the big data problems he saw coming, significantly slowing down the computing performance of clouds, greatly increasing energy costs, and diverting a server’s computing cycles away from running revenue-producing customer applications. Xockets’ DPUs enable widespread and affordable adoption of ML/AI in every industry for a new industrial revolution.
