Abstract: Data overflows can be prevented in edge computing systems. For example, an edge computing system (ECS) can include a memory buffer for storing incoming data from client devices. The ECS can also include a local storage device. The ECS can determine that an amount of available storage space in the local storage device is less than a predefined threshold amount. Based on determining that the amount of available storage space is less than the predefined threshold amount, the ECS can prevent the incoming data from being retrieved from the memory buffer. And based on determining that the amount of available storage space is greater than or equal to the predefined threshold amount, the ECS can retrieve the incoming data from the memory buffer and store the incoming data in the local storage device. This may prevent data overflows associated with the local storage device.

Abstract: Methods and systems are provided for assigning nodes to execute functions in a serverless computing environment. In one embodiment, a method is provided that includes receiving a function for execution in a serverless computing environment and identifying a storage pool needed during execution of the function. The serverless computing environment may include nodes for executing functions and a first set of nodes may be identified that implement the storage pool. Colocation measures may be determined between the first set of nodes and a second set of nodes. Available computing resources may be determined for the second set of nodes, such as available processing cores and available memory. The second set of nodes may be ranked according to the colocation measures and the available computing resources and a first node may be selected based on the ranking. The first node may be assigned to execute the function.

Abstract: Software performance can be automatically managed in a distributed computing environment. In one example, a system that can receive metrics information describing resource usage by a first instance of a service in a distributed computing environment. The system can also determine a quality-of-service (QoS) constraint for the service. The system can then modify a definition file based on the metrics information and the QoS constraint, the definition file being configured for deploying instances of the service in the distributed computing environment. The system can deploy a second instance of the service in the distributed computing environment using the modified definition file. As a result, the second instance can more closely satisfy the QoS constraint than the first instance.

Abstract: Resource utilization in edge-computing clusters and other computing clusters can be improved according to some aspects described herein. For example, an event handler can detect an event involving a state change to a data object. In response to detecting the event, the event handler can access an event registry storing relationships between a group of controllers and a group of events that are to be handled by the group of controllers. The event handler can determine, using the event registry, a controller that is configured to handle the event. The event handler can then transmit a command over a network to a computing cluster that includes a reconciler associated with the controller, where the command is for causing the reconciler to perform a reconciliation operation with respect to the data object. Separating the event-handling logic from the reconciliation logic in this way may improve resource utilization.

Abstract: Methods and systems for storing and injecting bytecode are provided. In one embodiment, a method is provided that includes receiving, at a first time, a first function for execution within a serverless computing environment; generating, by an interpreter, a first bytecode based on the first function; storing the first bytecode in association with an identifier of the first function; receiving, at a second time after the first time, a second function for execution within the serverless computing environment; identifying the second function as corresponding to the first function; injecting the first bytecode into a container for execution of the second function; receiving performance metrics regarding execution of the second function; and determining, based on the performance metrics, whether to allow or prevent future injection of the first bytecode.

Abstract: Resource utilization in edge-computing clusters and other computing clusters can be improved according to some aspects described herein. For example, an event handler can detect an event involving a state change to a data object. In response to detecting the event, the event handler can access an event registry storing relationships between a group of controllers and a group of events that are to be handled by the group of controllers. The event handler can determine, using the event registry, a controller that is configured to handle the event. The event handler can then transmit a command over a network to a computing cluster that includes a reconciler associated with the controller, where the command is for causing the reconciler to perform a reconciliation operation with respect to the data object. Separating the event-handling logic from the reconciliation logic in this way may improve resource utilization.

Abstract: Container-image layers can be managed. For example, a computing device can determine a first score for a first layer of a container image and a second score for a second layer of the container image. The computing device can determine that the first score corresponds to a first storage destination among several possible storage destinations. The computing device can also determine that the second score corresponds to a second storage destination among the possible storage destinations. The second storage destination can be different from the first storage destination. The computing device can then store (i) the first layer in the first storage destination based on the first layer being correlated to the first score, and (ii) the second layer in the second storage destination based on the second layer being correlated to the second score.

Abstract: Aspects of the disclosure provide for mechanisms for resource allocation in computer systems. A method of the disclosure may include running, on a plurality of nodes of a computer system, a benchmark workload utilizing a plurality of compute resources, and generating, by a processing device, a plurality of benchmarking results representing performance of the computer system utilizing the plurality of compute resources. The method may further include creating, by the processing device, a resource definition for the compute system in view of the benchmarking results, wherein the resource definition comprises a first resource allocation unit indicative of a first computing capacity of the computer system utilizing a first compute resource and a second resource allocation unit indicative of a second computing capacity of the computer system utilizing a second compute resource, and wherein the plurality of compute resources comprises the first compute resource and the second compute resource.

Abstract: Virus scanning of container images can be managed. For example, container images can be received in a sequential order. The container images can then be analyzed to determine the contents of the container images. The container images can be arranged in a virus-scanning queue in an order that is different from the sequential order in which the container images were received based on the contents of the container images. The container images can then be scanned for viruses in the order in which the container images are arranged in the virus-scanning queue.

Abstract: Systems and methods for provisioning system components to execute jobs are provided. In one embodiment, a method is provided that includes receiving a request to provision system components for executing a job. Aggregate latencies for computing units may be calculated based on startup latencies for jobs executing on the computing units. A particular computing unit may be selected from among the plurality of computing units based on the aggregate latencies, and system components may be provisioned from the computing unit.

Abstract: Sensitive information can be managed using a trusted platform module. For example, a system can encrypt target information using a cryptographic key to generate encrypted data. The system can also receive an encrypted key from a trusted platform module, where the encrypted key is a version of the cryptographic key that is encrypted using a public key stored in the trusted platform module. The system can then transmit the encrypted data and the encrypted key to a remote computing system, for example to store the encrypted data and the encrypted key on the remote computing system. Using these techniques, the target information may be secured and stored in remote locations.

Abstract: Data overflows can be prevented in edge computing systems. For example, an edge computing system (ECS) can include a memory buffer for storing incoming data from client devices. The ECS can also include a local storage device. The ECS can determine that an amount of available storage space in the local storage device is less than a predefined threshold amount. Based on determining that the amount of available storage space is less than the predefined threshold amount, the ECS can prevent the incoming data from being retrieved from the memory buffer. And based on determining that the amount of available storage space is greater than or equal to the predefined threshold amount, the ECS can retrieve the incoming data from the memory buffer and store the incoming data in the local storage device. This may prevent data overflows associated with the local storage device.

Abstract: Dynamic storage provisioning for nested clusters is disclosed. A hosted cluster (HC) storage provisioner, executing in a hosted cluster that is hosted by an infrastructure cluster (IC), receives a request to dynamically provision a hosted cluster (HC) persistent volume object that is coupled to a physical storage. The HC storage provisioner causes an IC control plane executing on the IC to generate IC volume metadata that is backed by a storage volume on the physical storage. The HC storage provisioner determines that the IC volume metadata has been generated. The HC storage provisioner creates HC volume metadata on the hosted cluster that is linked to the IC volume metadata, the HC volume metadata comprising an HC persistent volume object that represents a persistent volume for use by the hosted cluster that is backed, via the IC volume metadata, by the physical storage.

Abstract: Computer workloads can be managed across distributed computing clusters according to some aspects of the present disclosure. In one example, a system can receive a request from a workload manager for identifying a computing cluster to which to assign a workload. The system can determine that the workload is to be assigned to a particular computing cluster among a plurality of computing clusters based on historical information about replica deployment by the particular computing cluster. The system can then transmit a response to the workload manager for causing the workload manager to assign the workload to the particular computing cluster.

Abstract: A system and method for providing high availability for a thin-provisioned container cluster includes a memory, one or more processors in communication with the memory, a scheduler executing on the one or more processors, and a spot instance market monitor. The spot instance market monitor receives market information about spot instances in a cloud system at a first time. The spot instances are available to a client at the first time. The spot instance market monitor determines, based on the market information, a respective reliability value for each of the spot instances at the first time. Then, the scheduler selects one spot instance among the spot instances based on the reliability value of the spot instance. In response to the selection of the spot instance, the scheduler schedules a container on the spot instance and executes the container on the spot instance.

Abstract: A power consumption optimization system includes a virtual machine (VM) provisioned on a host, a memory, a server, and a processor in communication with the memory. The processor causes the server to store a power consumption profile of the VM. The VM runs at a processor frequency state. Additionally, the processor causes the server to receive a request to lower a processor frequency for the VM from an original processor frequency state to a reduced processor frequency state. The request has request criteria indicating a time duration associated with the request. The server validates the request criteria and a requirement of another tenant on the host. Responsive to validating the request criteria and the requirement the other tenant on the host, the server confirms the request to lower the processor frequency. Additionally, the server lowers the processor frequency to the reduced processor frequency state during the time duration.

Abstract: Automated quantum circuit job submission and status determination is disclosed. A classical computing system receives a request to execute a quantum circuit that includes quantum program instructions that perform one or more quantum operations on one or more qubits. The classical computing system requests, from a first quantum computing system that includes a first set of quantum computing devices, metadata describing characteristics of the first set of quantum computing devices. The classical computing system determines, based on the quantum circuit and the metadata, a selected quantum computing device of the first set of quantum computing devices, and sends, to the first quantum computing system, quantum circuit information that identifies the quantum circuit and instructions to execute the quantum circuit on the selected quantum computing device.

Abstract: Aspects and features of the present disclosure can provide a trusted, privacy-preserved deduplication process by executing deduplication functions in a trusted execution environment (TEE). In some examples, encrypted, incoming user data blocks are decrypted in the TEE to produce unencrypted user data blocks. An incoming digital fingerprint or each unencrypted user data block is produced. A processing device can compare the incoming digital fingerprint to existing digital fingerprints stored in the TEE to determine a presence of the incoming digital fingerprint and hence the presence of a copy of the data block in the storage platform, and writes the encrypted. Incoming data blocks are written to storage only when necessary. The technique allows public mass storage systems to meet cybersecurity objectives while achieving the storage space efficiency that deduplication provides.

Abstract: Events to be executed by one or more serverless functions are received from an events queue. A determination that a first set of the events are to be executed by one or more serverless functions at a local host system and a second set of the events are to be executed at one or more serverless functions at a remote host system is made. The first set of the events are directed to the one or more serverless functions at the local host system and the second set of the events are directed to the one or more serverless functions at the remote host system.

Abstract: One or more metrics associated with performance of a serverless function chain are received. The one or more metrics are used by a reinforcement learning (RL) model to tune the serverless function chain.