Disclosed techniques relate to circuitry configured to aggregate and report usage information in a distributed processor (e.g., a GPU). In some embodiments, graphics processor circuitry that includes at least first and second portions that are respectively configured to execute sets of graphics work. First utilization circuitry may track execution time for sets of graphics work on the first portion of the graphics processor circuitry and second utilization circuitry may track execution time for sets of graphics work on the second portion of the graphics processor circuitry. Command queue circuitry may store multiple different command queues. Control circuitry may access the first and second utilization circuitry and aggregate utilization data on a per-command-queue basis, where for a given command queue, the aggregated utilization data indicates respective utilization of the first and second portions of the graphics processor circuitry. The control circuitry may provide the aggregated per-command-queue utilization data in software-accessible registers.

This disclosure describes a method to minimize disruption for throughput oriented jobs in power oversubscription services with a dynamic control. The mechanism controls power in a hardware-agnostic way, and the policy employs a multi-threshold approach that balances power safety with workload impact. Moreover, an alternative control mechanism ensures proper system operation while power measurements are unavailable.

Systems and methods are provided for improving compute job distribution using federated computing nodes. This includes identifying a plurality of independently controlled computing nodes which then receive a token such that they can each be identified as being authorized to participate in a federated computing node cluster. Metrics associated with the particular nodes are then received and based on the received metrics compute jobs are assigned to the particular node by assembling a compute job data packet comprising the one or more compute jobs and transmitting the assembled compute job data packet to the particular node. Other features are also described in which assigned compute jobs and/or unrelated compute tasks can be dynamically modified in order to optimize compute job completion based on the received metrics.

Systems and techniques for adaptive limited-duration edge resource management are described herein. Available capacity may be calculated for a resource for a node of the edge computing network based on workloads executing on the node. Available set-aside resources may be determined based on the available capacity. A service request may be received from an application executing on the edge computing node. A priority category may be determined for the service request. Set-aside resources from the available set-aside resources may be assigned to a workload associated with the service request based on the priority category.

To provide a more efficient resource allocation method and system using a genetic algorithm (GA).

The present technology includes a method for allocating resources to a production process including a plurality of processes, the method including allocating priorities to the plurality of processes, selecting processes executable at a first time among the plurality of processes and capable of allocating necessary resources, allocating the necessary resources to the selected processes in descending order of priorities, selecting processes executable at a second time that is later than the first time among the plurality of processes and capable of allocating necessary resources, and allocating the necessary resources to the selected processes in descending order of priorities. The present technology also includes, as a method of expressing genes of GA, not having direct allocation information for genes but having information (priority) for determining an order for allocation.

Techniques are disclosed for executing a data processing pipeline. The techniques may include receiving a job at a data pipeline queue, setting up one or more distributed processing environments, and allocating the job to one of the distributed processing environments. The techniques may further include receiving the allocated job at a job queue within the distributed processing environment, increasing a priority level of the job, and executing the job within the distributed processing environment. The techniques can further include providing a retry pipeline at the data processing pipeline, and re-executing the job at a stage following a failure of at least one of its components. The techniques may decrement the retry budget as the job is re-executed.

A method, performed by a server device, may include receiving a request to activate an application session, the request being received from a user equipment on behalf of a particular application installed on the user equipment. The method may further include determining one or more application requirements associated with the particular application; determining conditions associated with one or more application servers; selecting a particular one of the one or more application servers based on the determined one or more application requirements and based on the determined conditions; and setting up the application session between the user equipment and between the selected particular one of the one or more application servers.

A method, performed by a server device, may include receiving a request to activate an application session, the request being received from a user equipment on behalf of a particular application installed on the user equipment. The method may further include determining one or more application requirements associated with the particular application; determining conditions associated with one or more application servers; selecting a particular one of the one or more application servers based on the determined one or more application requirements and based on the determined conditions; and setting up the application session between the user equipment and between the selected particular one of the one or more application servers.

An approach for estimating future cloud resource requests. The approach receives information defining a present budget interval. The approach analyzes a cloud resource request database based on entry times in respective past budget intervals occurring within a remaining time in the present budget interval. The approach creates estimates of future expected cloud resource requests based on the analysis.

Described herein are systems, methods, and software to manage the update to hosts in a computing environment. In one implementation, a method of operating an update service includes identifying a request to update a plurality of hosts and identifying host groups for the plurality of hosts. The method further includes prioritizing the host groups for the update and selecting a host group to be updated based on the prioritization. Once the host group is selected, the method also provides for identifying hosts to be updated for the host group based on resource scheduling information for the workloads in the host group. Once the group is updated, the method further includes repeating the update process for other host groups until all the host groups are updated.