Abstract: The technologies described herein are generally directed to using object metadata to chunk, store, and recover object data. For instance, a system can, based on a first data object and a second data object, generate a recovery data object. The system may further identify first storage information corresponding to first storage configured to store the first data object, second storage information corresponding to second storage configured to store the second data object, and third storage information corresponding to third storage configured to store the recovery data object. The system may further, based on the first storage information, second storage information, and the third storage information, respectively, generate first metadata, second metadata, and third metadata, respectively included in the first data object, second data object, and the recovery data object.

Abstract: A system can receive a connection message from a remote computer at a remote procedure call (RPC) endpoint. The system can send, to the remote computer, a remote direct memory access identifier via the RPC endpoint. The system can receive an object storage operation message from the remote computer at the RPC endpoint, wherein the object storage operation message identifies an object storage operation and an offset from a starting point of a data storage address of memory. The system can perform the object storage operation on a portion of the memory associated with the remote computer starting at the offset from the starting point of the data storage address of the memory associated with the remote computer, via a remote direct memory access operation. The system can send, to the remote computer, an RPC response that indicates that the object storage operation was successful.

Abstract: A system can receive a request from a requestor device to create computing infrastructure. The system can generate a response to the request that indicates that generating the computing infrastructure succeeded, independently of having generated the computing infrastructure. The system can store an association between the request and the response in a database. The system can send the response to the requestor device.

Abstract: A system can identify respective graphics processing units of respective computing nodes of a group of computing nodes with respect to which a containerized application operates. The system can virtualize the graphics processing units in a virtual node, to produce virtualized graphics processing units. The system can cause the virtual node to be available to a control application of a platform of the containerized application. The system can enable access to the virtualized graphics processing units by the containerized application via the virtual node.

Abstract: A system can read, by a first reader component, a first sub-sequence of a sequence of random computer data starting at a first offset position of the sequence of random computer data. The system can read, by a second reader component, a second sub-sequence of the sequence of random computer data starting at a second offset position of the sequence of random computer data, wherein the first offset position differs from the second offset position. The system can store the first sub-sequence as a first binary large object. The system can store the second sub-sequence as a second binary large object.

Abstract: Architectures and techniques are described that can dynamically determine certain optimization data with respect to node or node pool configurations. For example, a maximum pod per node (MPPN) value, a recommended instance type (RIT), and a cost per pod (CPP) value can be dynamically determined. The MPPN value can be determined to prevent pod eviction during autoscaling functions associated with a node pool. The RIT can be determined to reduce operational costs that may be higher if a different instance type is used instead.

Abstract: The technology described herein is directed towards fault injection to cloud provider resources, through a system that facilitates selection of specific resources in availability zone(s)/region(s). Example resources include VMs (virtual machines), VM clusters, tagged resource grouping and/or managed services. Based on (e.g., user) input data, the system injects faults to specified resources through cloud provider interfaces. The input data can specify availability zone(s), fault injection start time and duration. The input data can also specify fixed mode (fail specified resources together) or chaos mode (randomly inject failures for each resource individually). Failure type can be specified, e.g., graceful (e.g., clean shutdown) or non-graceful (e.g., a non-clean, hard fault). Based on the input, the system injects fault(s) using the modes selected to the specified resource(s) in the selected availability zone(s) for the duration entered.

Abstract: Architectures and techniques are described that can automatically determine proxy variable settings for a containerized workload for a given deployment environment of the containerized workload. Typically, a build pipeline of a container orchestration platform can build a containerized workload from received source code and deploys the containerized workload to a runtime environment. The disclosed techniques can, in response to a build being triggered, scan the source code and generate container images for HTTP requests. Once found, proxy variables can be determined and automatically injected into the container specifications.

Abstract: The technology described herein is directed towards an artificial intelligence scheduler for real-time task execution systems, including an implementation based on a recurrent learning model. In one implementation, the model incorporates a deep Q-network and proximal policy optimization to learn efficient scheduling policies for real-time systems. The scheduler can optimize for various real-time system aspects, including avoiding deadlocks, minimizing blocking, avoiding priority inversion, and reducing starvation. The artificial intelligence scheduler learns from actual system experience data to regularly improve scheduling policies and/or adapt to changing system requirements.

Abstract: Systems and methods support secure storage and use of a cryptographic seed phrase. In one example, a user may enter a text string of the seed phrase into an artificial intelligence (AI) art generator. The AI art generator generates art based upon the text string, where the resulting art is sufficiently non-literal that it may serve as a mnemonic device to a wallet owner, but its meaning would be underivable to someone other than the wallet owner.

Abstract: A method for managing a file system includes obtaining, by a compression optimizing manager, a compression algorithm selection request for the file system, determining a set of selection inputs based on a set of file system parameters of the file system, applying a compression selection model to the set of selection inputs to obtain a compression algorithm selection, and initiating a file system compression implementation of the file system using the compression algorithm selection.

Abstract: The technology described herein is directed towards automatic virtual subnet delegation. An automated process scans the subnets of a virtual network and builds a tree of the IP space, including allocated and unused space designations. User-defined parameters and organization policy data are used to determine the number of IP addresses needed for an application or the like. By traversing the tree of existing IP space, the technology described herein finds and places a new subnet, such as to ensure a high utilization of the overall IP space or based on an alternative type-of-fit criterion. When the virtual subnet space is created in the cloud, a public cloud-specific subnet identifier is returned to the user to utilize in deploying the application. Another use of the technology described herein is to track and optimize IP space allocation for existing virtual networks in the public cloud, including to identify underutilized and/or improperly-sized subnets.

Abstract: The technology described herein is directed towards automatic virtual subnet delegation. An automated process scans the subnets of a virtual network and builds a tree of the IP space, including allocated and unused space designations. User-defined parameters and organization policy data are used to determine the number of IP addresses needed for an application or the like. By traversing the tree of existing IP space, the technology described herein finds and places a new subnet, such as to ensure a high utilization of the overall IP space or based on an alternative type-of-fit criterion. When the virtual subnet space is created in the cloud, a public cloud-specific subnet identifier is returned to the user to utilize in deploying the application. Another use of the technology described herein is to track and optimize IP space allocation for existing virtual networks in the public cloud, including to identify underutilized and/or improperly-sized subnets.

Abstract: The technology described herein is directed towards fault injection to cloud provider resources, through a system that facilitates selection of specific resources in availability zone(s)/region(s). Example resources include VMs (virtual machines), VM clusters, tagged resource grouping and/or managed services. Based on (e.g., user) input data, the system injects faults to specified resources through cloud provider interfaces. The input data can specify availability zone(s), fault injection start time and duration. The input data can also specify fixed mode (fail specified resources together) or chaos mode (randomly inject failures for each resource individually). Failure type can be specified, e.g., graceful (e.g., clean shutdown) or non-graceful (e.g., a non-clean, hard fault). Based on the input, the system injects fault(s) using the modes selected to the specified resource(s) in the selected availability zone(s) for the duration entered.

Abstract: A system can read, by a first reader component, a first sub-sequence of a sequence of random computer data starting at a first offset position of the sequence of random computer data. The system can read, by a second reader component, a second sub-sequence of the sequence of random computer data starting at a second offset position of the sequence of random computer data, wherein the first offset position differs from the second offset position. The system can store the first sub-sequence as a first binary large object. The system can store the second sub-sequence as a second binary large object.

Abstract: A method for managing a file system includes obtaining, by a compression optimizing manager, a compression algorithm selection request for the file system, determining a set of selection inputs based on a set of file system parameters of the file system, applying a compression selection model to the set of selection inputs to obtain a compression algorithm selection, and initiating a file system compression implementation of the file system using the compression algorithm selection.

Abstract: Techniques are provided for domain name system multipathing distributed applications. In an example, a client computer queries a multipath-aware domain name system service with a domain name for a plurality of application servers, where there are a plurality of network paths available between the client computer and the application servers. The multipath-aware domain name system service can select a network path to be used, then determine a server that leads to that network path being used (by determining a hash used in network routing for each server of the plurality of servers). The multipath-aware domain name system service can respond to the client computer with an IP address for a selected application server, which results in communications between the client computer and the selected application server occurring over the selected network path.

Abstract: Techniques are provided for artificial intelligence-based storage monitoring. In an example, a system determines structured and unstructured attributes of a folder in a file system and provides them to an trained artificial intelligence model that outputs whether the folder is interesting or not. The folders labelled interesting by the trained artificial intelligence model can be further refined to a subset of folders that are placed in a watch list, and monitored for changes.

Abstract: Techniques are provided for artificial intelligence-based storage monitoring. In an example, a system determines structured and unstructured attributes of a folder in a file system and provides them to an trained artificial intelligence model that outputs whether the folder is interesting or not. The folders labelled interesting by the trained artificial intelligence model can be further refined to a subset of folders that are placed in a watch list, and monitored for changes.

Abstract: Techniques are provided for domain name system multipathing distributed applications. In an example, a client computer queries a multipath-aware domain name system service with a domain name for a plurality of application servers, where there are a plurality of network paths available between the client computer and the application servers. The multipath-aware domain name system service can select a network path to be used, then determine a server that leads to that network path being used (by determining a hash used in network routing for each server of the plurality of servers). The multipath-aware domain name system service can respond to the client computer with an IP address for a selected application server, which results in communications between the client computer and the selected application server occurring over the selected network path.