Embodiments are generally directed apparatuses, methods, techniques and so forth to select two or more processing units of the plurality of processing units to process a workload, and configure a circuit switch to link the two or more processing units to process the workload, the two or more processing units each linked to each other via paths of communication and the circuit switch.

A system for secure data transfer using air gapping. A first module includes: a first module communication interface configured to communicate with a public network. A second module includes: a first read-only memory storing an operating system; a second read-only memory storing sets of private keys of the second module and at least one public key of another remote entity; a cryptographic unit configured to encrypt and/or decrypt data using the keys stored in the second read-only memory. A bridge module includes: a bridge module controller; memory for storing data; a switch configured to selectively connect the bridge module data interface to either the first module data interface or to the second module data interface such that the first module data interface is never connected with the second module data interface.

An extensible multi-precision data pipeline system, comprising, a local buffer that stores an input local data set in a local storage format, an input tensor shaper coupled to the local buffer that reads the input local data set and converts the input local data set into an input tensor data set having a tensor format of vector width N by tensor length L, a cascaded pipeline coupled to the input tensor shaper that routes the input tensor data set through at least one function stage resulting in an output tensor data set, an output tensor shaper coupled to the cascaded pipeline that converts the output tensor data set into an output local data set having the local storage format and wherein the output tensor shaper writes the output local data set to the local buffer.

Techniques are disclosed for transferring a message between a sender agent and a receiver agent via a shared memory having a main memory and a cache. Feedback data indicative of a number of read messages in the shared memory is generated by the receiver agent. The feedback data is sent from the receiver agent to the sender agent. A number of unread messages in the shared memory is estimated by the sender agent based on the number of read messages. A threshold for implementing a caching policy is set by the sender agent based on the feedback data. The message is designated as cacheable if the number of unread messages is less than the threshold and as non-cacheable if the number of unread messages is greater than the threshold. The message is written to the shared memory based on the designation.

A method for messaging between operating system containers includes receiving, by a first proxy in a first user space container, a first message from a first service in the first user space container, the first message sent to the first proxy using a first messaging mechanism, forwarding, by the first proxy, the first message to a second proxy in a second user space container, the first message sent to the second proxy using a second messaging mechanism that is different than the first messaging mechanism, and delivering, by the second proxy, the first message to a second service in the second user space container.

Embodiments of the present disclosure provide a method, a device, and a computer program product for data sharing. The method includes acquiring first parameter information corresponding to a source process and second parameter information corresponding to a target process, and selecting a desired data sharing method from methods for sharing data between the source process and the target process based on the first parameter information and the second parameter information. The method further includes realizing data sharing between the source process and the target process based on the desired data sharing method. Through this solution, the data sharing efficiency between processes can be improved.

A method for improving performance of an access triggered architecture for a computer implemented application is provided. The method first executes typical operations of the access triggered architecture according to an execution time, wherein the typical operations comprise: obtaining a dataset and an instruction set; and using the instruction set to transmit the dataset to a functional block associated with an operation, wherein the functional block performs the operation using the dataset to generate a revised dataset. The method further creates a pipeline of the typical operations to reduce the execution time of the typical operations, to create a reduced execution time; and executes the typical operations according to the reduced execution time, using the pipeline.

Examples herein describe a peripheral I/O device with a hybrid gateway that permits the device to have both I/O and coherent domains. As a result, the compute resources in the coherent domain of the peripheral I/O device can communicate with the host in a similar manner as CPU-to-CPU communication in the host. The dual domains in the peripheral I/O device can be leveraged for machine learning (ML) applications. While an I/O device can be used as an ML accelerator, these accelerators previously only used an I/O domain. In the embodiments herein, compute resources can be split between the I/O domain and the coherent domain where a ML engine is in the I/O domain and a ML model is in the coherent domain. An advantage of doing so is that the ML model can be coherently updated using a reference ML model stored in the host.

Briefly, an encryption/decryption algorithm providing for consistent encryption entropy and encryption/decryption performance that is independent of the type of input data.

A method and system for computing one or more outputs of a neural network having a plurality of layers is provided. The method and system can include determining a plurality of sub-computations from total computations of the neural network to execute in parallel wherein the computations to execute in parallel involve computations from multiple layers. The method and system also can also include avoiding repeating overlapped computations and/or multiple memory reads and writes during execution.