Abstract: This disclosure describes techniques and mechanisms for providing an intelligent de-scheduler filtering system that minimizes service disruptions within a network. The techniques may provide continuous monitoring of clusters within a dynamic system and provide an intelligent determination of pod(s) within a duster to move by utilizing a disruption score based on disruption policies, balancer policies, service level agreement policies, and other data associated with the cluster and/or pods. The techniques enable a subset of pods that are flagged as violating compute usage to be selected to be moved, such that the subset will result in the least disruption to move to help restore or realign compute resources in the system.

Abstract: This disclosure describes techniques and mechanisms for providing an intelligent de-scheduler filtering system that minimizes service disruptions within a network. The techniques may provide continuous monitoring of clusters within a dynamic system and provide an intelligent determination of pod(s) within a cluster to move by utilizing a disruption score based on disruption policies, balancer policies, service level agreement policies, and other data associated with the cluster and/or pods. The techniques enable a subset of pods that are flagged as violating compute usage to be selected to be moved, such that the subset will result in the least disruption to move to help restore or realign compute resources in the system.

Abstract: A method of tuning telemetry collection parameters may include, with a collector, receiving source data defining at least one application running on a plurality of nodes, the nodes utilizing a finite number of compute resources. With the collector, a number of score models within a scoring agent of the collector may be executed to define telemetry collection parameters used by the collector for source data collection. The method may also include computing, with the scoring agent, a number of scores based on disturbance features and adaptive feedback, and tuning, with the collector, the telemetry collection parameters based on the scores to obtain tuned telemetry collection parameters.

Abstract: This disclosure describes techniques for scheduling workloads relating to data collection from network devices. The techniques may include weighting attributes that impact the data collection workload on computational, storage, and/or other resources of a data collector. The scheduling process may also include consideration of current usage rates at a data collector, to avoid overloading a particular collector while potentially underutilizing others. The weighting of attributes and/or consideration of usage rates may improve the efficiency of workload scheduling among a set of collectors. The overall result may include better utilization of the collectors and/or other resources of the computing network.

Abstract: Presented herein are embodiments for collecting and sharing telemetry data. A plurality of requests are received, from one or more applications, for telemetry data that is stored in a hierarchical tree representation comprising a plurality of nodes, wherein the telemetry data indicates an operational status and performance of a device, and wherein each request includes a cadence indicating a timespan at which the request repeats, and a path in the hierarchical tree where a requested portion of the telemetry data is stored. Processing the requests produces an optimized set of requests by merging any two or more requests whose paths indicate portions of data that share a common parent node in the hierarchical tree, wherein a merged request is assigned a cadence corresponding to a shortest cadence among the two or more requests. The hierarchical tree is polled using the optimized set of requests to satisfy the received requests.

Abstract: Presented herein are embodiments for collecting and sharing telemetry data. A plurality of requests are received, from one or more applications, for telemetry data that is stored in a hierarchical tree representation comprising a plurality of nodes, wherein the telemetry data indicates an operational status and performance of a device, and wherein each request includes a cadence indicating a timespan at which the request repeats, and a path in the hierarchical tree where a requested portion of the telemetry data is stored. Processing the requests produces an optimized set of requests by merging any two or more requests whose paths indicate portions of data that share a common parent node in the hierarchical tree, wherein a merged request is assigned a cadence corresponding to a shortest cadence among the two or more requests. The hierarchical tree is polled using the optimized set of requests to satisfy the received requests.