Some embodiments provide a novel method for sharing data between user-space processes and kernel-space processes without copying the data. The method dedicates, by a driver of a network interface controller (NIC), a memory address space for a user-space process. The method allocates a virtual region of the memory address space for zero-copy operations. The method maps the virtual region to a memory address space of the kernel. The method allows access to the virtual region by both the user-space process and a kernel-space process.

Roughly described, a system for data parallel training of a neural network on multiple reconfigurable units configured by a host with dataflow pipelines to perform different steps in the training CGRA units are configured to evaluate first and second sequential sections of neural network layers based on a respective subset of training data, and to back-propagate the error through the sections to calculate parameter gradients for the respective subset. Gradient synchronization and reduction are performed by one or more units having finer grain reconfigurability, such as an FPGA. The FPGA performs synchronization and reduction of the gradients for the second section while the CGRA units perform back-propagation through the first sequential section. Intermediate results are transmitted using a P2P message passing protocol layer. Execution of dataflow segments in the different units is triggered by receipt of data, rather than by a command from any host system.

The present invention discloses a method and a network device for processing service data and relates to the technical field of virtualization. The method includes: calling a Virtio to establish a communication connection with a Vhost deployed on a virtual switch when a first processing progress starts; applying for a target storage space by the first processing progress and dividing the target storage space into a plurality of sub-storage spaces; and determining a target sub-storage space in the plurality sub-storage spaces and processing the service data based on the target sub-storage space when a second processing progress starts. The present disclosure may save processing resources of a network device and ensure the efficiency of processing the service data.

A data sharing method includes the steps of establishing an association relationship between an information output device and an associated device; receiving shared data and a sharing manner sent by the associated device; determining a display mode of the information output device based on the sharing manner; and sending the shared data to the information output device to display based on the display mode.

A method for multiparty computation wherein a plurality of parties each compute a preset function without revealing inputs thereof to others, comprises: each of the parties performing a validation step to validate that computation of the function is carried out correctly, wherein the validation step includes: a first step that prepares a plurality of verified multiplication triples and feeds a multiplication triple to a second step when required; and the second step that consumes a randomly selected multiplication triple generated by the first step, wherein the first step performs shuffling of the generated multiplication triples, in at least one of shuffle in a sequence and shuffle of sequences.

Systems and methods herein describe receiving identification from a data pipeline, accessing first data offset information for a first data origin and second data offset information for a second data origin, bisecting the first data origin using the first data offset information, processing the data pipeline with the bisected first data offset information and the second data offset information, receiving a notification indicating a data pipeline status, and causing presentation of the notification on a graphical user interface of a computing device.

Methods for providing simultaneous access to a hardware data stream to multiple applications are disclosed. The first application to access a hardware device is responsible for providing and publishing an application programming interface (API) that provides access to the hardware device's data stream, which other applications can then call to gain access to the data stream. In some examples, the first application may be a server process or daemon dedicated to managing the hardware device data stream and publishing the API. In some further examples, the first application may instead may carry out user functionality unrelated to managing the hardware device.

A barrier-free atomic transfer method of multiword data is described. In the barrier-free method, a producer processor deconstructs an original parameter set of data into a deconstructed parameter set; and performs a series of single-copy-atomic writes to a series of single-copy-atomic locations. Each single-copy-atomic location in the series of single-copy-atomic locations comprises a portion of the deconstructed parameter set and a sequence number. A consumer processor can read the series of single-copy-atomic locations; verifies that the sequence number for each single-copy-atomic location in the series of single-copy-atomic locations is consistent (e.g., are all the same sequence number); and reconstructs the portions of deconstructed parameter set into the original parameter set.

The objective of the present invention is to prevent a conflict between variable names and consequently the unintentional overwriting of data when a plurality of programs that define a shared variable exist. A control device (1) is equipped with a program management part (11), a data storage part (13a), and a shared variable symbol table (13b). The program management part (11) acquires an identifier for a first user program, generates a shared variable name that includes the identifier and a variable name of a shared variable defined in the first user program, associates the shared variable name and an address of the shared variable with each other, and records these in the shared variable symbol table (13b).

Methods and systems are provided for processing a stream of incoming messages sent from a specific input message source and validating each incoming message of that stream before sending them to a specific target system.