Abstract: A multi-site data storage monitoring system communicates executable actions to remote element managers in non-executable text files. Each non-executable text file includes a human-readable textual description of the action, a top-level Java class of a known task type, and essential properties. The element managers generate executable jobs from the non-executable text files by instantiating an empty task object based on the top-level Java class, partially populating the empty task object based on the essential properties, adding the partially populated task object to a job structure, and using a token to provide authorization at job run-time. The token is created using user authentication information provided in response to presentation of the human-readable textual description of the action.

Abstract: Systems and methods described herein relate to the assignment of user devices to workload clusters. Resource utilization on a plurality of user devices is monitored. A device agent on each user device may be used to monitor the resource utilization. A workload execution request identifies resource requirements of a cluster workload. The workload execution request is assigned to a cluster based on the resource requirements of the cluster workload and the resource utilization on the plurality of user devices. The cluster comprises the plurality of user devices. The workload execution request is caused to be executed on the cluster. Each user device executes user workloads and a respective portion of the cluster workload.

Abstract: Techniques implemented by batch-processing service to provide users with visibility around execution metrics for their batch jobs, and determine more optimal scheduling policies and/or scaling algorithms for executing batch jobs. The batch-processing service may collect telemetry data for batch jobs, and analyze the telemetry data to generate execution metrics for the batch jobs. The batch-processing service uses these execution metrics to generate various visualizations that may be presented to users to provide the users with more visibility into the performance of their batch jobs. Additionally, the batch-processing service may analyze the execution metrics, and determine more optimized scheduling policies and/or scaling algorithms to use for running the batch jobs. In this way, the batch-processing service can increase the visibility given to users around the performance of their batch jobs, and can also identify modifications to scheduling and scaling in order to improve the performance of the batch jobs.

Abstract: Instantiation of a deployment image is accelerated by associating images with fingerprints, including fingerprints of constituent images (e.g., container and application images) of the deployment image. A database is maintained that associates hosts with the fingerprints of images instantiated thereon. When deploying the deployment image, a host associated with at least a portion of the fingerprint of the deployment image is identified and only the portion of the deployment image that is not already present on the host are transmitted to the host. An application image may be loaded into an already-executing container. The container may be restarted and invoke an entrypoint that references an orchestrator agent that retrieves and loads the application image into the container and invokes the entrypoint of the application image.

Abstract: A computer implemented method comprises a server processing work requests of a work requester. The work requester can communicate to the server a processing dependency of one work request on a second work request. The server can associate the dependency with the work requests and/or a queue of work requests. The dependency include a condition to be met in association with processing the work requests, and the condition can include an action for the server to take in association with processing a work request. A computing system can comprise a work requester, a server, and a set of dependency-aware queues for processing a set of work requests. A queue and/or work requests on the queues can be associated with a processing dependency and the server can process work requests enqueued to the queues in an order based on the dependencies. A work requester/server interface can comprise a dependency framework.

Abstract: A system and a method for allocating limited resources to agents that reveal their stochastic demands on arrival over a finite horizon in a proportionally fair manner that exhausts available resource budgets are provided. The allocation of limited resources includes receiving a request for a resource for which a predetermined maximum amount of the resource is available for allocation during a predetermined time interval, and estimating a number of future requests expected to be received and amounts of the resource to be requested within the predetermined time interval. The initial estimates are then adjusted by calculating a standard deviation of an uncertainty of at least one future request, and an amount of the resource to allocated to a resource requesting entity is determined by applying an algorithm with respect to the adjusted estimates and the predetermined maximum available amount of the resources.

Abstract: In one aspect, an example methodology implementing the disclosed techniques includes, by a computing device, determining an average time to finish for a first task to be executed and determining whether there is a resource instance that is designated to service tasks whose average time to finish matches the average time to finish for the first task. The method also includes, responsive to a determination that there is a resource instance that is designated to service tasks whose average time to finish matches the average time to finish for the first task, determining whether the resource instance has available capacity to service the first task. The method further includes, responsive to a determination that the resource instance has available capacity to service the first task, assigning the first task to the resource instance.

Abstract: Embodiments of this application disclose a data labeling system and method, and a data labeling manager. An example system includes a data labeling manager, a labeling model storage repository, and a basic computing unit storage repository. The data labeling manager receives a data labeling request, obtains a target basic computing unit, allocates a hardware resource to the target basic computing unit, establishes a target computing unit, obtains first storage path information of basic parameter data of a first labeling model, and sends the first storage path information to the target computing unit. The target computing unit obtains the basic parameter data of the to-be-used labeling model by using the first storage path information, combines a target model inference framework and the basic parameter data of the first labeling model to obtain the first labeling model, and labels to-be-labeled data by using the first labeling model.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for distributing a top k computation across multiple computing units of an integrated circuit One of the methods includes computing, by each of the plurality of computing units and for each candidate vector in a respective subset of the candidate vectors assigned to the computing unit, a respective distance between the query vector and the candidate vector; initializing, by the integrated circuit, a cut-off distance value; determining, by the integrated circuit, a final cut-off distance value; and providing, by the integrated circuit and as an output of a top k computation for the query vector and the set of candidate vectors, the candidate vectors that have respective distances that satisfy the final cut-off distance value.

Abstract: A decentralized scheduler node for an unstable network of heterogenous nodes collects status information including a resource capacity of each of a network of nodes with which it is in communication. For each current computation job in its input queue, the scheduler node determines which connected nodes have suitable resource capacity to execute the current job and a reliability of each suitable node based on statistical evaluation of its historical availability for job deployment. From the set of suitably powerful and reliable candidate nodes, the scheduler node selects a candidate node of sufficient resource capacity for deployment of the current job based on the probability of a subsequent job appearing during the execution time of the current job, the subsequent job having an equivalent or greater resource requirement and thus at risk of waiting in the input queue if the selected node is required for executing the subsequent job.

Abstract: Techniques for generating, simulating, and optimizing one or more provider-specific cloud-based architectures from a provider-independent architecture definition are disclosed. An architecture generator maps provider-independent service definitions to provider-specific service components for one or more specific cloud service providers. An architecture simulator simulates execution of a set of operations on the provider-specific cloud-based architectures to determine one or more performance and cost metrics. An architecture optimizer varies one or more design choices or parameters of a provider-specific service component to suggest which variant is optimal with respect to an optimization objective.

Abstract: A data processing method and communication device for configuring at least one split mode corresponding to at least one part of data associated with a data processing model. The method further includes: sending available processing resource to a second communication device; acquiring first configuration information from the second communication device; processing the jth part of data by using a sub-processing model corresponding to the ith split mode. The average quantity of processing resources occupied by processing the jth part of data using the first sub-processing model is matched with the available processing resource to improve a balance of resource occupation and a resource utilization rate of the communication device.

Abstract: The present invention provides systems and methods for tokenization of distributed compute resources. The distributed compute resources are provided in a multi-tenant hybrid fog computing cloud. The plurality of nodes are organized in a peer-to-peer edge network. The tokenization of distributed compute resources is automated and occurs in real time.

Abstract: In some embodiments, a method receives usage data that is based on a delivery of content by a plurality of content delivery network entities. A first value of a selection parameter is used to determine whether to select a content delivery network entity from the plurality of content delivery network entities to process a first request for content. The method allocates the usage data in a first distribution to the plurality of content delivery network entities. The allocating does not use a condition to determine the first distribution. The usage data is allocated in a second distribution to the plurality of content delivery network entities. The allocating uses the condition to determine the second distribution. The method adjusts the first value of the selection parameter to a second value based on the first distribution and the second distribution.

Abstract: This application relates to the field of NFV technologies, and discloses a data flow mirroring method and an apparatus. The method includes: An NFVO receives an instantiation request from an OSS/BSS, and obtains NSD information based on the instantiation request, where the NSD information includes first mirroring object information, the first mirroring object information describes information for creating a first mirroring object, and the first mirroring object is for mirroring a first data flow; and requests, based on the first mirroring object information, a VIM to create the first mirroring object, where the first mirroring object is associated with a first mirroring service. According to the method, the NSD information is extended, to include mirroring object information.

Abstract: A computer implemented method is provided. A number of processor units identify a number of candidate applications for processing requests based on types of application programming interface for each application. The number of processor units validate metadata related to status of applications for each application in the number of candidate applications. The number of processor units select a subgroup of applications from the number of candidate applications to process requests based on the validated metadata.

Abstract: Data flow analysis is provided. A program level data flow analysis is performed for each respective data flow path in a set of control flow chains corresponding to each respective program called by a particular job step in each respective job of a plurality of jobs in a sequence of job execution corresponding to an application. A particular field of a plurality of fields is identified in a record of each dataset of a plurality of datasets corresponding to the application that is included in a particular data flow path to form a field level data flow analysis for each particular data flow path. Results of the field level data flow analysis for each particular data flow path in the set of control flow chains corresponding to each respective program called by the particular job step in each respective job of the plurality of jobs is aggregated.

Abstract: In some examples, a system including physical graphics processing units (GPUs) receives a request to schedule a new job to be executed in the system that is accessible by a plurality of tenants to use the physical GPUs. The system allocates the new job to a collection of vGPUs of the physical GPUs based on an operational cost reduction objective to reduce a cost associated with a usage of the physical GPUs and based on a tenant isolation constraint to provide tenant isolation wherein a single tenant of the plurality of tenants is to use a physical GPU at a time.

Abstract: Apparatuses, systems, and techniques to determine metrics for paths connecting hardware components, select a plurality of groups of the hardware components based at least in part on the metrics, and perform at least a portion of a workload using a selected group of the plurality of groups.

Abstract: A cluster computing system is provided. The cluster computing system includes: a host including a first processor and a first buffer memory; computing nodes, each of which includes a second processor and a second buffer memory configured to store data received from the host; a network configured to connect the host and the computing nodes; and storage devices respectively corresponding to the computing nodes. The first processor is configured to control a task allocator to monitor a task performance state of each of the computing nodes, select at least one of the computing nodes as a task node, and distribute a background task to the task node, and the second processor of the task node is configured to perform the background task on sorted files stored in the second buffer memory, the sorted files being received by the second buffer memory from the first buffer memory.