Computing resources are managed in a computing environment comprising a computing service provider and an edge computing network. The edge computing network comprises computing and storage devices configured to extend computing resources of the computing service provider to remote users of the computing service provider. The edge computing network collects capacity and usage data for computing and network resources at the edge computing network. The capacity and usage data is sent to the computing service provider. Based on the capacity and usage data, the computing service provider, using a cost function, determines a distribution of workloads pertaining to a processing pipeline that has been partitioned into the workloads. The workloads can be executed at the computing service provider or the edge computing network.

A method and system of deployment of a virtualized service on a cloud infrastructure are described. A first service function specification of a first service function is selected. A determination of a set of the computing systems and a set of the links is performed based on availability and characteristics of the computing systems and the network resources in the cloud infrastructure. A selection of a first computing system to be assigned to host the first service function and links is performed based on the first service function specification. In response to determining that there is one or more of the service functions that are not yet assigned, the selection of a service function and the determination of a computing system and links are repeated for each of the remaining functions until all of the service functions are assigned to computing and networking resources in the cloud infrastructure.

Embodiments of the present disclosure provide a method for splitting a neural network model to be processed by a multi-core processor and related products. When a splittable operator is present in the neural network model, the operator is split, and an optimal splitting combination is selected to obtain an optimal splitting result of an entire neural network model, and then sub-operators corresponding to the optimal splitting result are executed through multiple cores in parallel. Thereby, a purpose of reducing resource consumption of a computer device is achieved.

A method for performing a distributed computation on a computing system using computational resources dynamically allocated using a computational resource manager includes storing information specifying quantities of computational resources associated with respective ones of a number of program portions of the program, where the program portions perform successive transformations of data and each program portion uses computational resources granted by the computational resource manager enabling computation associated with that program portion to be performed in the computing system, requesting a first quantity of computational resources associated with a first program portion of the number of program portions from the computational resource manager, receiving a second quantity of computational resources from the computational resource manager, less than the requested first quantity of computational resources, performing computation associated with the first portion of the program using the second quantity of computational resources, while performing the computation associated with the first portion of the program using the second quantity of computational resources, receiving an additional quantity of computational resources from the computational resource manager, and performing an additional computation associated with the first portion of the program using the additional quantity of computational resources while performing the computation associated with the first portion using the second quantity of computational resources.

A method for scheduling tasks includes receiving input that was acquired using one or more data collection devices, and scheduling one or more input tasks on one or more computing resources of a network, predicting one or more first tasks based in part on the input, assigning one or more placeholder tasks for the one or more predicted first tasks to the one or more computing resources based in part on a topology of the network, receiving one or more updates including an attribute of the one or more first tasks to be executed as input tasks are executed, modifying the one or more placeholder tasks based on the attribute of the one or more first tasks to be executed, and scheduling the one or more first tasks on the one or more computing resources by matching the one or more first tasks to the one or more placeholder tasks.

A method for distributed neural network processing, the method may include detecting, by a local neural network that belongs to a local device, and based on sensed information, an occurrence of a triggering event for executing or completing a classification or detection process; sending to a remote device, a request for executing or completing the classification or detection process by a remote device that comprises a remote neural network; wherein the remote neural network has more computational resources than the local neural network; determining by the remote device whether to accept the request; and executing or completing, by the remote device, the classification or detection process when determining to accept the request; wherein the executing or completing involves utilizing the remote neural network.

A method and system of deployment of a virtualized service on a cloud infrastructure while taking into consideration variability/elasticity of the cloud resources are described. A first service function specification is selected. Candidate computing systems and candidate links are determined. The determination of the candidates is performed based on the availability and characteristics of the computing systems and the network resources in the cloud infrastructure which includes varying characteristics. A first computing system and one or more links are determined from the candidates for placing the first service function in the cloud infrastructure.

A method includes identifying, by a first instance of a service, a first number of data partitions of a data source to be processed by the service and a second number of instances of the service available to process the first number of data partitions. The method further includes separating the first number of data partitions into a first set of data partitions and a second set of data partitions in view of the second number of instances of the service, determining a target number of data partitions from the first set of data partitions to be claimed by each of the second number of instances of the service, and claiming, by the first instance of the service, the target number of data partitions from the first set of data partitions and up to one data partition from the second set of data partitions.

A method for multi-task scheduling, a device and a storage medium are provided. The method may include: initializing a list of candidate scheduling schemes, the candidate scheduling scheme being used to allocate a terminal device for training to each machine learning task in a plurality of machine learning tasks; perturbing, for each candidate scheduling scheme in the list of candidate scheduling schemes, the candidate scheduling scheme to generate a new scheduling scheme; determining whether to replace the candidate scheduling scheme with the new scheduling scheme based on a fitness value of the candidate scheduling scheme and a fitness value of the new scheduling scheme, to generate a new scheduling scheme list; and determining a target scheduling scheme, based on the fitness value of each new scheduling scheme in the new scheduling scheme list.

Systems and methods are provided for offloading a task from a central processor in a radio access network (RAN) server to one or more heterogeneous accelerators. For example, a task associated with one or more operational partitions (or a service application) associated with processing data traffic in the RAN is dynamically allocated for offloading from the central processor based on workload status information. One or more accelerators are dynamically allocated for executing the task, where the accelerators may be heterogeneous and my not comprise pre-programming for executing the task. The disclosed technology further enables generating specific application programs for execution on the respective heterogeneous accelerators based on a single set of program instructions. The methods automatically generate the specific application programs by identifying common functional blocks for processing data traffic and mapping the functional blocks to the single set of program instructions to generate code native to the respective accelerators.