Abstract: A method includes receiving a reservation request corresponding to resource requirements of an application and including role-based access and first priority level indicators. A reservation subset of existing reservations is identified based on the role-based access indicator, and server clusters are evaluated based on the resource requirements, the first priority level indicator, the reservation subset, and server cluster resources. When the server clusters are determined to include at least one qualified server cluster, a first server cluster is identified either as a sole qualified server cluster or by ranking multiple qualified server clusters by calculating a score of each qualified server cluster, an existing reservation of the reservation subset assigned to the first server cluster is pre-empted based on the first priority level indicator relative to priority level indicators of the one or more existing reservations, and the reservation request is assigned to the first server cluster.

Abstract: The present disclosure provides a multi-queue multi-cluster task scheduling method and system. The method includes: constructing a training data set; training and optimizing a plurality of parallel deep neural networks (DNN) by using the training data set to obtain a plurality of trained and optimized parallel DNNs; setting a reward function, where the reward function minimizes the sum of a task delay and energy consumption by adjusting a reward value proportion of the task delay and a reward value proportion of the energy consumption; inputting a to-be-scheduled state space into the plurality of trained and optimized parallel DNNs to obtain a plurality of to-be-scheduled action decisions; determining an optimal action decision among the plurality of to-be-scheduled action decisions based on the reward function for output; and scheduling the plurality of task attribute groups to a plurality of clusters based on the optimal action decision.

Abstract: Apparatuses, systems, and techniques to optimize processor resources at a user-defined level. In at least one embodiment, priority of one or more tasks are adjusted to prevent one or more other dependent tasks from entering an idle state due to lack of resources to consume.

Abstract: The disclosed approach works without the individualized credentials of failed machines when setting up recovery VMs in a cloud computing environment. Each recovery VMs is customized to properly correspond to the system state of its failed counterpart. An illustrative data storage management system recovers backup data and system states collected from the counterpart computing devices, custom-configures recovery VMs in the cloud computing environment, and injects the desired drivers into each recovery VM during an enhanced bare-metal restore process. The enhanced bare-metal restore process works without the failed computer's credentials. The system also restores the backed up data to recovery volumes attached to the recovery VMs. The present approach is both scalable and secure.

Abstract: Scaling an ordered event stream (OES) based on an information-unit (IU) metric is disclosed. The IU metric can correspond to an amount of computing resources that can be consumed to access information embodied in event data of an event of the OES. In this regard, the amount of computing resources to access the data of the stream event itself can be distinct from an amount of computing resources employed to access information embodied in the data. As such, where an external application, e.g., a reader, a writer, etc., can connect to an OES data storage system, enabling the OES to be scaled in response to burdening of computing resources accessing event information, rather than merely event data, can aid in preservation of an ordering of events accessed from the OES.

Abstract: A deployment system enables a developer to generate a deployment plan according to a logical, multi-tier application blueprint defined by application architects. The deployment plan includes tasks to be executed for deploying application components on virtual computing resource provided in a cloud infrastructure. The deployment plan includes time dependencies that determine an execution order of the tasks according to dependencies between application components specified in the application blueprint. The deployment plan enables system administrators to view the application blueprint as an ordered workflow view that facilitates collaboration between system administrators and application architects.

Abstract: In one set of embodiments, a computer system can receive a request to provision a virtual machine (VM) in a host cluster, where the VM is associated with a virtual graphics processing unit (GPU) profile indicating a desired or required framebuffer memory size of a virtual GPU of the VM. In response, the computer system can execute an algorithm that identifies, from among a plurality of physical GPUs installed in the host cluster, a physical GPU on which the VM may be placed, where the identified physical GPU has sufficient free framebuffer memory to accommodate the desired or required framebuffer memory size, and where the algorithm allows multiple VMs associated with different virtual GPU profiles to be placed on a single physical GPU in the plurality of physical GPUs. The computer system can then place the VM on the identified physical GPU.

Abstract: Embodiments of the present disclosure present a hyper-square interconnect topology and advanced ring-based AllReduce operations. In some embodiments, a topology is provided that is an improvement over conventional interconnect topologies by eliminating delays associated with long wirings. In some embodiments, computing nodes are divided into sub-sections to better allocate computing tasks, and the system can be optimized to divide up the computing nodes by maximizing the number of square sub-sections in the topology. In some embodiments, the system can be optimized to select square sub-sections first for each computing task. Each sub-section can comprise some computing nodes or all computing nodes in the hyper-square interconnect topology. This flexibility allows the hyper-square interconnect topology to utilize the computing nodes more efficiently by assigning appropriate numbers of computing nodes to each computing task based on the computing need of the computing task.

Abstract: Controlling allocation of resources in network function virtualization. Data defining a pool of available physical resources is maintained. Data defining one or more resource allocation rules is identified. An application request is received. Physical resources from the pool are allocated to virtual resources to implement the application request, on the basis of the maintained data, the identified data and the received application request.

Abstract: Representative apparatus, method, and system embodiments are disclosed for configurable computing. A representative system includes an asynchronous packet network; a plurality of configurable circuits arranged in an array, each configurable circuit coupled to the asynchronous packet network and adapted to perform a plurality of computations; and a dispatch interface circuit adapted to partition the plurality of configurable circuits into one or more separate partitions of configurable circuits and to load one or more computation kernels into each partition of configurable circuits. The dispatch interface circuit may load balance across the partitions of configurable circuits by starting threads for execution in the partition having the highest number of available thread identifiers. The dispatch interface may also assert a partition enable signal to merge the one or more separate partitions and assert a stop signal to all configurable circuits of the one or more separate partitions of configurable circuits.

Abstract: In at least some embodiments, a store-type operation is received and buffered within a store queue entry of a store queue associated with a cache memory of a processor core capable of executing multiple simultaneous hardware threads. A thread identifier indicating a particular hardware thread among the multiple hardware threads that issued the store-type operation is recorded. An indication of whether the store queue entry is a most recently allocated store queue entry for buffering store-type operations of the hardware thread is also maintained. While the indication indicates the store queue entry is a most recently allocated store queue entry for buffering store-type operations of the particular hardware thread, the store queue extends a duration of a store gathering window applicable to the store queue entry. For example, the duration may be extended by decreasing a rate at which the store gathering window applicable to the store queue entry ends.

Abstract: Techniques are disclosed for reallocating host resources in a virtualized computing environment when certain criteria have been met. In some embodiments, a system identifies a host disabling event. In view of the disabling event, the system identifies a resource for reallocation from a first host to a second host. Based on the identification, the computer system disassociates the identified resource's virtual identifier from the first host device and associates the virtual identifier with the second host device. Thus, the techniques disclosed significantly reduce a system's planned and unplanned downtime.

Abstract: At an interface an analytic model for processing data is received. The analytic model is inspected to determine a language, an action, an input type, and an output type. A virtualized execution environment is generated for an analytic engine that includes executable code to implement the analytic model for processing an input data stream.

Abstract: A spinlock circuit connected to one or more first processors through one or more broadside interfaces. The spinlock circuit is configured to receive a plurality of requests for use of a computing resource from one or more first processors, and reply to each of the plurality of requests within a single clock cycle of the one or more first processors. The spinlock circuit can reply to each of the plurality of requests within a single clock cycle of the one or more first processors by alternately assigning the computing resource to a requesting processor from among the one or more first processors or indicating to the requesting processor from among the one or more first processors that the computing resource is not available.

Abstract: A global state associated with a garbage collection (GC) process is a state assumed by a reference after GC operations for a current GC phase are performed for the reference. A frame state of a frame on a call stack of an application thread tracks a progress of GC operations for references therein. A color watermark is added after a global state change. The frame originally marked by a stack watermark is now marked by the new color watermark. The new color watermark marks a deepest frame on the call stack having the old global state. A color watermark is deleted when the stack watermark meets the color watermark. If the stack watermark is shifted in the call stack to mark the same frame as a color watermark, then the color watermark is deleted. The marked frame is a deepest frame on the call stack having the current global state.

Abstract: Systems, methods, and computer programming products leveraging recurrent neural network architectures to proactively predict workload demand of container orchestration platforms. The platform continuously collects metric data from clusters of the platform and train multiple parallel neural networks with different architectures to predict future platform workload demands. At periodic intervals, the registered neural networks in consideration for controlling the scaling operations of the platform are compared against one another to identify the neural network demonstrating the highest performance and/or most accurate workload prediction strategy for scaling the orchestration platform. The selected neural network is enforced as controller for the platform to implement the workload prediction strategy.

Abstract: Systems and methods are described for providing an application-level gateway to an on-demand network code execution system. An on-demand network code execution system may allow users to submit code to be executed in a serverless environment, and may provide an interface for executing the user-submitted code on demand. The interface may require that users authenticate, provide input in a particular format, or meet other criteria when sending a request to execute the code. An application-level gateway may thus provide an interface that implements these functions, thereby allowing computing devices to interact with the code as though it were running on a server (e.g., by using HTTP). The application-level gateway may also use on-demand code execution to provide load balancing for servers that are running the user-submitted code, and seamlessly provide access to code that runs on both server-based and serverless environments.

Abstract: Methods, apparatus, systems, and articles of manufacture to provide tenant aware runtime feature toggling in a cloud or other virtualized computing environment are disclosed. An example method includes determining a feature toggle associated with a resource of a provisioning request; retrieving the feature toggle from a database using a first tenant identifier, the feature toggle to have a first value for the first tenant identifier and a second value for a second tenant identifier; processing the feature toggle to provision the resource according to the first value of the feature toggle; and facilitating provisioning of the resource according to the first value.

Abstract: Systems and methods are described for compiling a specified instruction from a first virtual application to a second virtual application. Each virtual application may be associated with different programming languages. In an example method, a computing device receives a request to execute the specified instruction in the second virtual application. A target data structure may be created, using a library of the second virtual application, where a template directory may be stored. First syntax features, each defining a respective variable may be identified. An abstract syntax tree may be used to derive, for each first syntax feature, a modified definition for the respective variable. Second syntax features may be generated that define the respective variables more precisely than the first syntax features. The specified instruction may be rendered the second virtual application and may be expressed via the second syntax features and their respective variables.

Abstract: A method for distributed processing includes receiving an idle time and at least one task execution characteristic corresponding to each respective processor of a plurality of processors, wherein at least one processor of the plurality of processors is remotely located from other processors of the plurality of processors. The method also includes identifying a target processor of the plurality of processors to execute a task based on the idle time and the at least one task execution characteristic of the target processor. The method also includes communicating, to the target processor, a task request requesting the target processor execute the task and, in response to receiving a communication from the target processor indicating acceptance of the task, communicating, to the target processor, instructions for executing the task.