Abstract: Managing the resource demand load for edge systems is significantly more complex than for other systems, such as cloud environments. A time period in which an application or task is operating based on initial demand resource load values that are provided by a customer may be inaccurate, which may expose sub-standard execution. Embodiments herein seek to significantly mitigate the potential of sub-standard execution. Embodiments collect a repository of resource demand load usage data over a time period that can be used to accurately determine the statistical moments of uncertain resource demand load. In one or more embodiments, a repository of hypervector and/or hyperspace representations may be generated and used to help with resource demand load estimation.

Abstract: Presented herein are embodiments to implement a temporal queueing system with class-based fair queuing and dynamic resource allocation based on a novel look-ahead capability to manage various models and workloads for utilization/efficiency improvements. Embodiments may be implemented to allocate accelerator resources based on platform-defined timeslots, and therefore significantly increase the ability of workloads to access hardware accelerator resources. Training and inference may be supported with flexible preemption and the ability to support run-to-completion for training tasks while still supporting non-run-to-completion for inference tasks. Embodiments may be implemented by an edge software operation platform through virtual accelerators to allow emulation of different types of hardware accelerators and to map to the hardware accelerators with hardware-specific procedures managed by an edge orchestrator and an edge endpoint.

Abstract: Managing the resource demand load for edge systems is significantly more complex than for other systems, such as cloud environments. A time period in which an application or task is operating based on initial demand resource load values that are provided by a customer may be inaccurate, which may expose sub-standard execution. Embodiments herein seek to significantly mitigate the potential of sub-standard execution. Embodiments collect a repository of resource demand load usage data over a time period that can be used to accurately determine the statistical moments of uncertain resource demand load. In one or more embodiments, a repository of hypervector and/or hyperspace representations may be generated and used to help with resource demand load estimation.

Abstract: Managing the resource demand load for edge systems is significantly more complex than for other systems, such as cloud environments. Unlike cloud systems and other frameworks that are able to use closed-form solutions based on Poisson processes or other tractable Gaussian-based probability distributions, edge systems present complex waveforms, pareto/alpha-stable distributions, and long-range dependence. Based on elaborately designed embodiments that recognize the complexities of edge data, one can estimate scaling and multi-fractal dimensionality to determine predictive models.

Abstract: Managing the resource demand load for edge systems is significantly more complex than for other systems, such as cloud environments. Embodiments herein provide edge resource demand load estimation systems and methods that inform scheduling and associated edge orchestration to ensure that edge system resource capacity is appropriately utilized. Efficient utilization allows an increased number of applications to be deployed at a reduced level of reserved resources. Also presented are embodiments of assurance mechanisms for monitoring edge resource demand load characterizations. In one or more embodiments, when an estimate or estimates are deemed to not be valid (e.g., having experienced stationary drift), updated estimates may be obtained.

Abstract: Managing the resource demand load for edge systems is significantly more complex than for other systems, such as cloud environments. Unlike cloud systems and other frameworks that are able to use closed-form solutions based on Poisson processes or other tractable Gaussian-based probability distributions, edge systems present complex waveforms, pareto/alpha-stable distributions, and long-range dependence. Based on elaborately designed embodiments that recognize the complexities of edge data, one can estimate scaling and multi-fractal dimensionality to determine predictive models.

Abstract: Managing the resource demand load for edge systems is significantly more complex than for other systems, such as cloud environments. Edge resource demand load scheduling systems and methods are disclosed that can ensure that edge systems operate smoothly and efficiently while balancing multiple scheduling objectives. Scheduling techniques disclosed herein may utilize heuristic rules for candidate edge system selection (e.g., utilizing ARMA/ARIMA averages and/or service level objectives) and modified best fit decreasing (mBFD) assignment/allocation techniques.

Abstract: Managing the resource demand load for edge systems is significantly more complex than for other systems, such as cloud environments. Embodiments herein provide edge resource demand load estimation systems and methods that inform scheduling and associated edge orchestration to ensure that edge system resource capacity is appropriately utilized. Efficient utilization allows an increased number of applications to be deployed at a reduced level of reserved resources. Also presented are embodiments of assurance mechanisms for monitoring edge resource demand load characterizations. In one or more embodiments, when an estimate or estimates are deemed to not be valid (e.g., having experienced stationary drift), updated estimates may be obtained.

Abstract: Self-provisioning a computer system is illustrated. Code in a secure base activation image implements an auto-connection service at the computer system. The auto-connection service attempts to perform a local activation by attempting to determine that a private port is open and using the private port, initiating an identification and attestation process with an activation service. The auto-connection service performs a remote activation by determining that the private port is closed, defaulting activation to a public port, and using the public port, initiating an identification and attestation process with the activation service.

Abstract: Data characterization techniques in an information processing system environment are disclosed. In one example, at least one processing device is configured to detect a source application associated with data obtained from execution of at least one of a plurality of applications in an information processing system, wherein the plurality of applications comprise services associated with multiple different policies. The processing device is further configured to classify the data to determine an intent associated with the data, wherein classifying comprises utilizing a machine learning classification process.

Abstract: Data characterization techniques in an information processing system environment are disclosed. In one example, at least one processing device is configured to obtain data associated with execution of at least one of a plurality of applications in an information processing system. The processing device extracts features from the obtained data, and then selects a reduced subset of features from the extracted features for use in a machine learning classification process configured to determine an intent of the obtained data.

Abstract: Data characterization techniques are disclosed. In one example, at least one processing device is configured to obtain information from at least one of a plurality of data feature extraction and selection processes that respectively operate in conjunction with a plurality of machine learning classification processes that determine intent of data generated by execution of at least one of a plurality of applications in an information processing system. The processing device executes a reinforcement learning process on at least a portion of the obtained information to determine one or more parameters, and propagates the one or more parameters to one or more of the plurality of data feature extraction and selection processes for use in training of the one or more of the plurality of data feature extraction and selection processes to respectively operate in conjunction with one or more corresponding ones of the plurality of machine learning classification processes.

Abstract: Application monitoring techniques in an information processing system environment are disclosed. In one example, at least one processing device is configured to obtain an indication of at least one anomalous behavior associated with execution of an application in an information processing system, wherein the application comprises a plurality of services. The processing device is further configured to analyze, across a plurality of time periods, at least one metric associated with the execution of the application to determine at least one critical path associated with the execution of the application, wherein the critical path comprises at least a portion of the plurality of services. The processing device is then configured to analyze the critical path using a set of variance correlation algorithms and identify a set of one or more services in the critical path that are highest in a ranked order determined by the set of variance correlation algorithms.

Abstract: Application monitoring techniques comprising predictive anomaly detection and fault isolation are disclosed for use in an information processing system environment. For example, an apparatus comprises at least one processing device comprising a processor coupled to a memory. The processing device is configured to obtain application-level data generated for an information processing system in accordance with execution of an application and obtain hardware-level data generated for the information processing system in accordance with the execution of the application. The processing device is further configured to utilize an unsupervised machine learning model to predictively detect anomalous behavior in accordance with the execution of the application in the information processing system, based on at least a portion the application-level data and the hardware-level data. The processing device may also initiate fault isolation in addition to predicting anomalous behavior.

Abstract: Techniques described herein relate to a method for deploying workflows. The method may include receiving, by a platform controller of a domain, a workflow portion from a service controller of a federated controller, provisioning a set of devices in the domain to the workflow portion based on a first fit, generating, by the platform controller, a workflow fingerprint based on the provisioning of the set of devices and based on the workflow portion, executing the workflow portion in the domain using the set of devices, making a determination that the workflow portion requires additional resources, based on the determination, provisioning additional resources of the domain to the workflow portion to obtain an updated execution resource set, and updating the workflow fingerprint based on the updated execution resource set to obtain an updated workflow fingerprint, and executing the workflow portion using the updated execution resource set.

Abstract: Techniques described herein relate to a method for managing a distributed multi-tiered computing (DMC) environment. The method includes obtaining, by a local controller associated with a DMC domain, a set of correlated services associated with an anomaly; obtaining a service dependency graph associated with the set of correlated services; generating a causal variance for each service using the correlated services and the service dependency graph; generating a weighted rank order of causal services based on the causal variance associated with each service, and the weighted rank order of causal services includes a portion of the services associated with an application associated with the anomaly; and performing remediation based on the weighted rank order of the causal services.

Abstract: Techniques described herein relate to a method for distributed data management. The method may include making a first determination that data is written to a data structure of storage of a data host; obtaining, based on the first determination, data information associated with the data; making a second determination of intent corresponding to the data; generating a global name and metadata corresponding to the data, wherein the metadata comprises the intent; and publishing the global name and the metadata to a global name repository.

Abstract: Techniques described herein relate to a method for deploying workflows.

Abstract: Techniques described herein relate to a method for managing a distributed multi-tiered computing (DMC) environment. The method includes obtaining, by a local controller associated with a DMC domain, a weighted rank order of causal services associated with a service impairment; obtaining contextual information associated with the weighted rank order of causal services; identifying top causal services based on the weighted rank order of causal services; generating, using a causal symbolic engine, a ranked order of top root causes based on the top causal services and the contextual information; selecting remediation profiles associated with the top root causes; executing remediation profiles in ranked order of the top root causes; making a determination that the remediation profiles were successful; and in response to the determination: logging the service impairment and notifying a user associated with the service impairment.

Abstract: Techniques described herein relate to a method for distributed data management. The method may include obtaining data descriptors for an application executing on a data host, performing a dataset policy analysis using the data descriptors to determine a data placement for a dataset associated with the application using a global name repository, performing, based on the data policy analysis, the data placement, and based on the data placement, updating the global name repository.