Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for distributing resources to client devices are described. A computer-implemented system receives a request for a resource distribution constraint for a given resource of a first type, determines that a second type of resource that has a sufficient amount of historical distribution data to generate a distribution constraint, and determines the resource distribution constraint using historical distribution data for the second type of resource.

Abstract: A content delivery service that facilitates the selection of virtual machine instances utilized to implement the transcoding function. A management service first implements a test channel configuration routine that selects worst case scenario configurations for implementing the transcoding function and measures performance metrics for different virtual machine instances implementing the transcoding function. The management system can then match incoming requests for video channel transcoding with signature data from the test channel configurations to identify which virtual machine instances can implement the requested transcoding functionality.

Abstract: A customer function on a computer system comprising processing cores. The customer function comprises applications with a multitude of different tasks. The computer system provides containers and is configured to activate and de-activate said containers so that a container is active or inactive, wherein all tasks of the applications are assigned to containers, wherein all tasks of each specific application are assigned to exactly one specific container, wherein in a timeframe, during which a container is active, one or more cores of the computer system are exclusively reserved for the execution of the tasks of the application of said container, and wherein the computer system is configured such that when a container is inactive, the tasks of said container cannot be executed on the computer system. For each container a task sequencer is provided.

Abstract: Techniques for memory access management in a distributed computing system are described herein. In some aspects, the techniques described herein relate to a method for memory access management in a distributed computing system, where the method includes: receiving a first request to execute a first operation using a distributed architecture and in a uniform memory access (UMA) mode, wherein the distributed architecture comprises a first processor, a first memory that is local to the first processor, and a second memory that is remote to the first processor; subsequent to receiving the first request and a first delay period, transmitting first data associated with the first operation to the first processor, wherein the first data is stored in the first memory; and subsequent to receiving the first request, transmitting second data associated with the first operation to the first processor, wherein the second data is stored in the second memory.

Abstract: Disclosed embodiments may include a system for tracing cloud service costs. The system may receive data corresponding to a first node cluster, the first node cluster indicative of a plurality of cloud services utilized by one or more users. The system may extract metrics associated with the plurality of cloud services. The system may determine a relationship between each cloud service of the plurality of cloud services based on the metrics. The system may determine one or more costs associated with the first node cluster, the one or more costs based on the relationship. The system may determine whether the first node cluster is utilized within a predetermined threshold. Responsive to determining the first node cluster is not utilized within the predetermined threshold, the system may automatically conduct one or more actions associated with the one or more costs.

Abstract: A computing device includes (a) circuitry for managing at least (1) a first thread and (2) a second thread, (b) circuitry for causing the first thread to initiate or receive a priority task, (c) circuitry for determining a status of the second thread, (c) circuitry for, responsive to the second thread having an idle status, copying at least some state associated with the first thread to registers of the second thread, activating the second thread, and changing its availability status from idle to active; (d) circuitry for at least one of concurrently or simultaneously: (1) copying the at least some state from the registers of the second thread to a stack; and (2) performing the priority task at least in part with the first thread; and (e) circuitry for, upon completion of the copying, deactivating the second thread and changing its availability status to idle.

Abstract: A conditional locking system receives, from a request source, a request to execute an operation on a target resource on behalf of a client thread. The system generates a unique key based on an attribute of the request and searches a concurrency dictionary for an entry identifying the unique key. In response to determining that the concurrency dictionary does include an entry identifying the unique key, the system identifies a lock object associated in the entry with the unique key and transmits a request to acquire the lock object. Once the lock object associated with the unique key has been acquired on behalf of the client thread, the system transmits a lock acquisition confirmation to the request source.

Abstract: Described herein are systems and methods for sharing vitals among service replicas to enable processing of long running workflows within a container orchestration system. A method can provide a container orchestration system that provides within one or more container orchestration environments, a runtime for containerized workloads and services. The method can provide a healthbus within the container orchestration system, the healthbus comprising a memory. The method can deploy a plurality of pods within the container orchestration system, each pod comprising a memory. The method can periodically publish, by each pod, a health message to the healthbus, the health message comprising at least an indication of an identification of the pod and an indication of a time interval in which the pod has been active. The method can periodically query, by each pod, the healthbus to determine a world view of the container orchestration system.

Abstract: A method includes obtaining an agentic workflow from an artificial intelligence (AI) agent execution framework. An execution graph of the agentic workflow is generated by an LLM, including a multitude of nodes corresponding to a multitude of tasks. The tasks are partitioned by complexity into first complexity level tasks and second complexity level tasks. A first set of candidate models performs the first complexity level tasks. A first candidate model is assigned to a first task, based on an optimization function value of performance metrics of the first candidate model. A second set of candidate models performs the second complexity level tasks. A second candidate model is further assigned to a second task. The candidate models are assigned to respective nodes of the execution graph, to modify the execution graph. An updated agentic workflow is generated from the modified execution graph and deployed to the AI agent execution framework.

Abstract: A method includes receiving (1) a request to perform of one or more analytics operations, and (2) data associated with an industrial automation system, wherein the industrial automation system comprises a plurality of devices configured to perform a plurality of automation operations, and wherein each of the plurality of devices comprises a compute surface configured to perform one or more software tasks, determining a plurality of data processing tasks included in the one or more analytics operations, identifying a portion of the plurality of devices to perform the plurality of data processing tasks based on the compute surface available for each of the plurality of devices, deploying one or more containers to each of the portion of the plurality of devices, and providing first instructions to a first container of the one or more containers to perform a first data processing task of the plurality of data processing tasks.

Abstract: Some embodiments provide a method for operating a physical server in a network. The method stores multiple copies of a virtual machine (VM) image at a network-accessible storage. The method uses a first copy of the VM image as a virtual disk to execute a VM on a hypervisor of a first physical computing device. The method uses a second copy of the VM image as a virtual disk accessible via a smart network interface controller (NIC) of a second physical computing device to execute an operating system of the second physical computing device.

Abstract: Some embodiments provide a method of providing distributed storage services to a host computer from a network interface card (NIC) of the host computer. At the NIC, the method accesses a set of one or more external storages operating outside of the host computer through a shared port of the NIC that is not only used to access the set of external storages but also for forwarding packets not related to an external storage. In some embodiments, the method accesses the external storage set by using a network fabric storage driver that employs a network fabric storage protocol to access the external storage set. The method presents the external storage as a local storage of the host computer to a set of programs executing on the host computer. In some embodiments, the method presents the local storage by using a storage emulation layer on the NIC to create a local storage construct that presents the set of external storages as a local storage of the host computer.

Abstract: A system and method for orchestrating distributed operations to be executed in a distributed computing system with multiple virtual infrastructures uses a distributed operation descriptor to find any Cloud-Native Network Function (CNF) entry in the distributed operation descriptor. For each found CNF entry, a CNF descriptor is retrieved from a CNF catalog and parsed to find an overridable property for a CNF described in the CNF descriptor for which a property override is defined. Then, a target virtual infrastructure is selected from the multiple virtual infrastructures to perform a lifecycle management operation of the CNF. Instructions are then transmitted to a local operator in the target virtual infrastructure with the property override so that information regarding the CNF is transmitted to a local orchestrator of the target virtual infrastructure to perform the lifecycle management operation of the CNF at the target virtual infrastructure using the property override.

Abstract: Disclosed are a parallel method and device for convolution computation and data loading of a neural network accelerator. The method needs two input feature maps and two convolution kernel cache blocks, and sequentially stores the input feature maps and 64 convolution kernels into cache sub-blocks according to a loading length, so as to execute convolution computation and simultaneously load data of a next group of 64 convolution kernels.

Abstract: An example methodology includes declaring a node in a cluster of multiple nodes as a lead node to perform a cluster upgrade. While the lead node is performing the upgrade operations, the remaining nodes in the cluster continue to provide the cluster services. The upgrade operations on the lead node include migrating applications on the lead node to one of the other nodes in the cluster and performing upgrade of the lead node. In some cases, the upgrade operations include rebooting the lead node with a new OS image and installing new versions of the applications or re-installing the same applications on the lead node. The upgrade operations on the lead node may also include, once the lead node is upgraded, stopping applications and cluster services on the other nodes in the cluster, creating a new cluster that includes the lead node, and starting the applications on the lead node.

Abstract: A request for completion by an artificial intelligence (AI) model is received. Tasks required to fulfill the request are identified. An AI model is selected based on the tasks and capabilities of the AI model. The request is transmitted to the selected AI model. A response is received from the AI model. The response is then transmitter to a requester.

Abstract: A method of collecting and reporting inventory of resources deployed in a data center that includes hardware resources, a virtualization management software executed to provision virtual resources from the hardware resources, and a cloud management server executed to provision the virtual resources for tenants of the data center, includes the steps of: executing a first API call to the virtualization management software to collect first inventory of virtual resources deployed in the data center and a second API call to the cloud computing management software to collect second inventory of virtual resources deployed in the data center; storing the first and second inventory; and in response to an inventory request from a central orchestrator, initially sending a subset of the stored first and second inventory to the central orchestrator in accordance with parameters included in the inventory request, and thereafter sending updates to the subset to the central orchestrator periodically.

Abstract: A first hypervisor running on a first processor cluster is provided. During operation, the first hypervisor can determine a first set of processing nodes and a first memory unit of the first processor cluster in response to the booting up of a first Basic Input/Output System (BIOS) of the first processor cluster. The first hypervisor can discover a second hypervisor running on a second processor cluster comprising a second set of processing nodes and a second memory unit. The first hypervisor can operate, with the second hypervisor, a distributed system comprising the first and second sets of processing nodes and the first and second memory units. The first hypervisor can then operate, with the second hypervisor, a global virtual machine on the distributed system. The virtual memory space of the global virtual machine can be mapped to respective memory spaces of the first and second processor clusters.

Abstract: A system can be provided for deploying bare metal clusters that satisfy custom resource requests. For example, the system can receive from a client device, a custom resource request. The custom resource request can include a set of requirements for a bare metal cluster. The set of requirements can include a number of nodes for the bare metal cluster. The system can determine a set of resources that satisfies the set of requirements. The set of resources can include virtual Internet Protocol (IP) addresses and a set of baseband management controller (BMC) IP addresses. A number of BMC IP addresses in the set of BMC IP addresses can be equal to the number of nodes for the bare metal cluster. Additionally, the system can generate, based on the set of resources, a configuration file for the bare metal cluster and deploy, based on the configuration file, the bare metal cluster.

Abstract: In one embodiment, an apparatus includes: a plurality of execution circuits to execute and instruct micro-operations (?ops), where a subset of the plurality of execution circuits are capable of execution of a fused ?op; a fusion circuit coupled to at least the subset of the plurality of execution circuits, wherein the fusion circuit is to fuse at least some pairs of producer-consumer ?ops into fused ?ops; and a fusion throttle circuit coupled to the fusion circuit, wherein the fusion throttle circuit is to prevent a first ?op from being fused with another ?op based at least in part on historical information associated with the first ?op. Other embodiments are described and claimed.