Abstract: Methods, systems, and computer-readable media for execution of replicated tasks using redundant resources are disclosed. Replicas of a task are generated. Computing resources are selected from at least one pool of computing resources of a provider network. The provider network includes a plurality of pools of computing resources that vary in a characteristic, and the computing resources are selected based (at least in part) on the characteristic. Concurrent execution of the replicas of the task is initiated using the selected computing resources. Input data for the concurrent execution does not vary from one of the replicas to another of the replicas, and at least a portion of the replicas produce individual results for the input data. Based (at least in part) on a policy, an individual result of one or more of the replicas is selected as a final result of the task.

Abstract: Techniques for API-based endpoint discovery and centralized load balancing involving provider substrate extension resources are described. A discovery coordinator service located within a provider network can identify one or more endpoints from a set of potentially distributed endpoints for a client to utilize, where endpoints may be located within provider substrate extensions of the provider network. The discovery coordinator service can analyze the loads of these endpoints, via client lease information, to identify nearby, low-load resources that may be most optimal for the client to use via providing minimal latency of access.

Abstract: Techniques for utilizing dual-stack network addressing for compute instances hosted in an edge location of a cloud provider network along with communications service provider (CSP) network addresses are described. A first network address is assigned to the compute instance from a pool of network addresses of the cloud provider network, and a second network address is associated with the compute instance that is provided by the CSP network. A gateway of the edge location is updated to direct packets addressed to the second network address to the compute instance via use of the first network address.

Abstract: Methods, systems, and computer-readable media for job scheduling based on job execution history are disclosed. A request is received to schedule a workload comprising a plurality of jobs. A resource allocation score for the workload is determined. The resource allocation score represents (at least in part) an estimated likelihood of successful execution of the workload. A first portion of the workload is scheduled for execution, and a remaining portion (if any) of the workload is delayed. A quantity of jobs in the first portion of the workload is determined based (at least in part) on the resource allocation score. Execution of the first portion of the workload is initiated.

Abstract: Methods, systems, and computer-readable media for job execution with managed compute environments are disclosed. A specification of a managed compute environment comprises one or more constraints associated with computing resources in the managed compute environment. A queue or other data structure that is associated with the managed compute environment is monitored. The data structure is configured to store jobs. Data indicative of a job is detected in the data structure. One or more computing resources are reserved for the job from a pool of available computing resources. The one or more computing resources are selected for the job based at least in part on the one or more constraints associated with computing resources in the managed compute environment. Execution of the job using the one or more computing resources is initiated.

Abstract: Techniques are described for enabling users of a cloud provider network to discover “availability groups” provided by a cloud provider network and to request the launch of computing resources into selected availability groups. Some cloud provider networks are expanding the definition of traditional “availability zones” to include new types of availability zones representing various types of provider substrate extension edge locations—including, for example, cloud-provider managed substrate extensions associated with separate control planes, 5G-enabled provider substrate extensions connected to communications service provider networks, and the like. Availability groups can be used to represent various defined collections these new types of provider substrate extensions, where each availability group may be defined such that includes a set of provider substrate extensions with a similar set of characteristics and capabilities.

Abstract: Techniques for distributed network connectivity monitoring of provider network edge location resources from cellular networks are described. A central service transmits test suites of commands to agents executed by test devices, which can execute the commands to test network characteristics between the test devices and target locations via one or multiple cellular communications networks. Results of the testing are sent back to the central service for processing, and the resultant metrics can be used for intelligent latency-based routing of clients, latency-based placement of resources, and/or performance monitoring of deployed resources.

Abstract: A scheduler of a batch job management service determines that a set of resources a client is insufficient to execute one or more jobs. The scheduler prepares a multi-dimensional statistical representation of resource requirements of the jobs, and transmits it to a resource controller. The resource controller uses the multi-dimensional representation and resource usage state information to make resource allocation change decisions.

Abstract: Methods, systems, and computer-readable media for job execution with managed compute environments are disclosed. A specification of a managed compute environment comprises one or more constraints associated with computing resources in the managed compute environment. A queue or other data structure that is associated with the managed compute environment is monitored. The data structure is configured to store jobs. Data indicative of a job is detected in the data structure. One or more computing resources are reserved for the job from a pool of available computing resources. The one or more computing resources are selected for the job based at least in part on the one or more constraints associated with computing resources in the managed compute environment. Execution of the job using the one or more computing resources is initiated.

Abstract: Techniques for distributed network connectivity monitoring of provider network edge location resources from cellular networks are described. A central service transmits test suites of commands to agents executed by test devices, which can execute the commands to test network characteristics between the test devices and target locations via one or multiple cellular communications networks. Results of the testing are sent back to the central service for processing, and the resultant metrics can be used for intelligent latency-based routing of clients, latency-based placement of resources, and/or performance monitoring of deployed resources.

Abstract: Techniques for distributed network connectivity monitoring of provider network edge location resources from cellular networks are described. A central service transmits test suites of commands to agents executed by test devices, which can execute the commands to test network characteristics between the test devices and target locations via one or multiple cellular communications networks. Results of the testing are sent back to the central service for processing, and the resultant metrics can be used for intelligent latency-based routing of clients, latency-based placement of resources, and/or performance monitoring of deployed resources.

Abstract: Techniques are described for establishing connectivity between compute instances hosted by an edge location of a cloud provider network (also referred to as a provider network substrate extension (PSE)) deployed within a communications service provider (CSP) network and the communications service provider network. A compute instance is hosted on a PSE of a cloud provider network, the PSE connected to a CSP network. A first network address is assigned to the compute instance from a pool of network addresses of the CSP network. A gateway of the PSE is updated to direct packets addressed to the first network address to the compute instance.

Abstract: A method for processing computationally intensive workloads includes monitoring a storage system to determine what portion of results of computations associated with an application have been received by the storage system. Responsive to the monitoring determining that the portion of results is sufficient to trigger allocation of one of a plurality of event handlers to the application to process the portion of results: an event handler is allocated to the application, the portion of the results is retrieved from the storage system, a specification of a first computation to be performed and the portion of the results are provided to the event handler, the event handler applies the first computation to the portion of results to produce a new result, the new result is stored in the storage system, and the event handler is deallocated from the application.

Abstract: Techniques for API-based endpoint discovery involving provider substrate extension resources are described. A discovery coordinator service located within the provider network can identify one or more endpoints from a set of potentially distributed endpoints for a client to utilize, where endpoints may be located within provider substrate extensions of the provider network. The discovery coordinator service can utilize location values of the client provided via an API request, such as its network address or geographic coordinates, to identify a nearby resource that may be most optimal for the client to use via providing minimal latency of access.

Abstract: Techniques are described for enabling users of a cloud provider network to discover “availability groups” provided by a cloud provider network and to request the launch of computing resources into selected availability groups. Some cloud provider networks are expanding the definition of traditional “availability zones” to include new types of availability zones representing various types of provider substrate extension edge locations—including, for example, cloud-provider managed substrate extensions associated with separate control planes, 5G-enabled provider substrate extensions connected to communications service provider networks, and the like. Availability groups can be used to represent various defined collections these new types of provider substrate extensions, where each availability group may be defined such that includes a set of provider substrate extensions with a similar set of characteristics and capabilities.

Abstract: Techniques for reducing latency between resources of a cloud provider network and end user devices are described. In one embodiment, a system includes a cloud provider network including a first control plane service to manage customer compute instances hosted on resources of the cloud provider network. The system further includes a first provider substrate extension of the cloud provider network connected to a first communications service provider network. The first provider substrate extension includes a first computer system having capacity for executing customer compute instances, is controlled at least in part by the first control plane service via a connection through at least a portion of the first communications service provider network, executes at least a first customer compute instance using the first computer system, wherein the first customer compute instance communicates with a first mobile computer system connected via the first communications service provider network.

Abstract: Methods, systems, and computer-readable media for job execution with scheduled reserved compute instances are disclosed. One or more queues are mapped to a compute environment. The queue(s) are configured to store data indicative of jobs. The compute environment is associated with one or more scheduled reserved compute instances, and the one or more scheduled reserved compute instances are reserved for use in the compute environment for a window of time. The queue(s) are mapped to the compute environment prior to the window of time opening. During the window, at least one of the scheduled reserved compute instances is added to the compute environment. The scheduled reserved compute instance(s) are provisioned from a pool of available compute instances. During the window, execution is initiated of one or more jobs from the queue(s) on the scheduled reserved compute instance(s) in the compute environment.

Abstract: Techniques are described for enabling users of a cloud provider network to discover “availability groups” provided by a cloud provider network and to request the launch of computing resources into selected availability groups. Some cloud provider networks are expanding the definition of traditional “availability zones” to include new types of availability zones representing various types of provider substrate extension edge locations—including, for example, cloud-provider managed substrate extensions associated with separate control planes, 5G-enabled provider substrate extensions connected to communications service provider networks, and the like. Availability groups can be used to represent various defined collections these new types of provider substrate extensions, where each availability group may be defined such that includes a set of provider substrate extensions with a similar set of characteristics and capabilities.

Abstract: Methods, systems, and computer-readable media for event-driven scheduling using directed acyclic graphs are disclosed. A directed acyclic graph is generated that comprises a plurality of nodes and a plurality of edges. The nodes represent jobs, and the edges represent dependency relationships between individual jobs. Based (at least in part) on one or more events, a job scheduler determines that one of the nodes represents a runnable job. One or more of the dependency relationships for the runnable job are satisfied by the one or more events. An execution schedule is determined for the runnable job. Based (at least in part) on the execution schedule, execution of the runnable job is initiated using one or more computing resources.

Abstract: A scheduler of a batch job management service determines that a set of resources a client is insufficient to execute one or more jobs. The scheduler prepares a multi-dimensional statistical representation of resource requirements of the jobs, and transmits it to a resource controller. The resource controller uses the multi-dimensional representation and resource usage state information to make resource allocation change decisions.