An accelerator system is implemented on an expansion card comprising a printed circuit board having (a) one or more graphics processing units (GPUs), (b) two or more associated memory banks (logically or physically partitioned), (c) a specialized controller, and (d) a local bus providing signal coupling compatible with the PCI industry standards. The controller handles most of the primitive operations to set up and control GPU computation. Thus, the computer's central processing unit (CPU) can be dedicated to other tasks. In this case a few controls (simulation start and stop signals from the CPU and the simulation completion signal back to CPU), GPU programs and input/output data are exchanged between CPU and the expansion card. Moreover, since on every time step of the simulation the results from the previous time step are used but not changed, the results are preferably transferred back to CPU in parallel with the computation.

A method of operating a terminal (130), comprising: transmitting, to a master control node (290) of an edge computing architecture (299), a configuration request message (7001), the configuration request message (7001) being indicative of at least one of a connectivity status of the terminal (130) indicative of an access network and a location of the terminal (130), receiving, from the master control node (290), a configuration response message (7002) indicative of an edge computing system (260-267), and executing an edge computing application (4001, 4002) using the edge computing system (260-267).

The present application discloses a task allocation method and system based on a resource management platform. The method comprises: receiving an artificial intelligence model training and/or testing task and a name of data set required for processing the task; acquiring data set distribution information of a plurality of nodes; judging if the node has the required data sets according to names of the data sets in the node; and selecting a node with the size of the required data set meeting preset requirements for task allocation according to the size of the required data set in the node if the node has the required data set. It may be seen that, in the present application, the situation of data sets in a node is taken into account during task allocation, and the node with the size of the required data set meeting preset conditions is selected for task allocation, such that the node does not need to download the required data set or reduce the frequency of downloading data sets by a node, thereby improving the efficiency in processing tasks.

A non-transitory computer-readable recording medium stores a program that causes a computer to execute a process that includes receiving load arrangement of jobs in a case where a compute server mounted in a compute rack in a server room executes the jobs, the server room being a room where the compute rack in which the compute server is mounted and a storage rack in which a storage is mounted are arranged, and estimating a time at which a predetermined job of the compute server is to be offloaded to the storage that generates less heat than the compute server and estimating setting temperature and an air volume of an air conditioner, based on the load arrangement and time-series data of temperature and power of the server room, such that the power of the server room is reduced within limitation conditions of the compute server, the storage, and the air conditioner.

Systems and methods are presented herein for implementing reinforcement learning in a task-facilitation service. The task-facilitation service may receive a request to delegate an execution of a task. The request can may include a user identifier that corresponds to the task. The task-facilitation service may generate a proposal using a machine-learning process. The proposal may include an implementation of the task and facilitate execution of the task by one or more third-party service providers. The task-facilitation service may facilitate the execution of the task by the one or more third-party service providers according to the proposal. In response to receiving an execution status of the task, the task-facilitation service may train the machine-learning process using the proposal and the execution status to improve subsequent proposals generated by the machine-learning process.

Examples include registering a device driver with an operating system, including registering available hardware offloads. The operating system receives a call to a hardware offload, inserts a binary filter representing the hardware offload into a hardware component and causes the execution of the binary filter by the hardware component when the hardware offload is available, and executes the binary filter in software when the hardware offload is not available.

Implementations detailed herein include description of a computer-implemented method. In an implementation, the method at least includes receiving an application instance configuration, an application of the application instance to utilize a portion of an attached accelerator during execution of a machine learning model and the application instance configuration including: an indication of the central processing unit (CPU) capability to be used, an arithmetic precision of the machine learning model to be used, an indication of the accelerator capability to be used, a storage location of the application, and an indication of an amount of random access memory to use.

Examples are disclosed that relate to processing files between a local network and a cloud computing service. One example provides a computing device configured to be located between a local network and a cloud computing service, comprising a logic machine and a storage machine comprising instructions executable to receive, from a device within the local network, a file at a local share of the computing device, and in response to receiving the file, generate a file event indicating receipt of the file at the local share and provide the file event to a virtual machine executing on the computing device. The instructions are further executable to, based upon a property of the file, provide the file to a program operating within a container in the virtual machine to process the file, and send a result of executing the program on the file to the cloud computing service.

In examples provided herein, upon receiving notification of a computational task requested by a package to provide an experience to a user, a remote node management engine identifies computing nodes for performing the computational task and determining available processing resources for each computing node, where a computing node resides at networked wearable devices associated with the user. The remote node management engine further selects one of the computing nodes as a primary controller to distribute portions of the computational task to one or more of the other computing nodes and receive results from performance of the portions of the computational task by the other computing nodes, and provides to the selected computing node information about available processing resources at each computing node.

An apparatus to facilitate transparent network access controls for spatial accelerator device multi-tenancy is disclosed. The apparatus includes a secure device manager (SDM) to: establish a network-on-chip (NoC) communication path in the apparatus, the NoC communication path comprising a plurality of NoC nodes for ingress and egress of communications on the NoC communication path; for each NoC node of the NoC communication path, configure a programmable register of the NoC node to indicate a node group that the NoC node is assigned, the node group corresponding to a persona configured on the apparatus; determine whether a prefix of received data at the NoC node matches the node group indicated by the programmable register of the NoC; and responsive to determining that the prefix does not match the node group, discard the data from the NoC node.