Abstract: A request to execute a recurring job is received by a cloud computing environment. Thereafter, available computing resources available to execute the job are determined based on historical resource consumption trends. A resource prediction for the job is then generated using an ensemble model ((S)ARIMA model) that combines an autoregressive moving average (ARMA) model and an autoregressive moving average (ARIMA) prediction models. The resource prediction characterizes resources to be consumed for successfully executing the job. Execution of the job can then be scheduled by the cloud computing environment based on the resource prediction and the available computing resources. Related apparatus, systems, techniques and articles are also described.

Abstract: Implementations detailed herein include description of a computer-implemented method.

Abstract: A load balancing method, apparatus and device for a parallel model training task, and a computer-readable storage medium, includes: acquiring data traffic and a theoretical computational amount of each network layer in a target model; determining a theoretical computing capability of each computing device, and obtaining an initial computational amount corresponding to each computing device according to the theoretical computing capability and the theoretical computational amount; performing a load balancing operation according to the initial computational amount by using multiple device critical layer position division rule, so as to obtain a plurality of initial balancing schemes; compiling statistics on time performance parameters corresponding to the initial balancing schemes, and determining an intermediate balancing scheme from the initial balancing schemes according to the time performance parameters; and adjusting the intermediate balancing scheme according to the data traffic, so as to obtain a final

Abstract: A digital environment includes multiple computing nodes and a scheduling system that assigns workloads to computing nodes. The scheduling system includes an equivalence-class-based resource usage prediction system that receives a workload request and predicts an equivalence class for that workload request based on resource usage over time by the workload request or metadata associated with the workload request. The scheduling system also includes a workload assignment system that assigns the workload request to one or more of the computing nodes based on the predicted equivalence class. The number of equivalence classes is small relative to the total number of workloads that are scheduled (as an example, 10 to 15 equivalence classes for a total number of workloads in the tens or hundreds of thousands).

Abstract: Example techniques for execution of functions by clusters of computing nodes are described. In an example, if a cluster does not have resources available for executing a function for handling a service request, the cluster may request another cluster for executing the function. A result of execution of the function may be received by the cluster and used for handling the service request.

Abstract: Disclosed are aspects of workload selection and placement in systems that include graphics processing units (GPUs) that are virtual GPU (vGPU) enabled. In some aspects, workloads are assigned to virtual graphics processing unit (vGPU)-enabled graphics processing units (GPUs) based on a variety of vGPU placement models. A number of vGPU placement neural networks are trained to maximize a composite efficiency metric based on workload data and GPU data for the plurality of vGPU placement models. A combined neural network selector is generated using the vGPU placement neural networks, and utilized to assign a workload to a vGPU-enabled GPU.

Abstract: A method includes receiving a request to migrate a virtual machine from a source host to a destination host, mapping, by a hypervisor running on the source host, a first portion of a memory of the virtual machine to a persistent memory device, where the persistent memory device is accessible by the source host machine and the destination host machine, responsive to determining that a time period to execute a synchronization operation with respect to the first portion of the memory by the persistent memory device is below a threshold, stopping the virtual machine on the source host, and starting the virtual machine on the destination host.

Abstract: A technique is described for managing processor (CPU) resources in a host having virtual machines (VMs) executed thereon. A target size of a VM is determined based on its demand and CPU entitlement. If the VM's current size exceeds the target size, the technique dynamically changes the size of a VM in the host by increasing or decreasing the number of virtual CPUs available to the VM. To “deactivate” virtual CPUs, a high-priority balloon thread is launched and pinned to one of the virtual CPUs targeted for deactivation, and the underlying hypervisor deschedules execution of the virtual CPU accordingly. To “activate” virtual CPUs, the number of virtual CPUs, the launched balloon thread may be killed.

Abstract: The present disclosure relates to a method and system for optimizing a parameter configuration of a distributed computing job.

Abstract: A method and system for collaborative workload placement and optimization. Specifically, the disclosed method and system entail implementing metric exchange between hosts in a clustered or non-clustered environment to perform infrastructure workload placement and optimization decisions. The exchange of metrics may be performed through a lightweight service, which would not impact any existing environment resource utilization. Further, resource utilization maps may be used to offer near real-time tracking of resource utilization across the aforementioned hosts, thereby facilitating resource utilization forecasting for the efficient placement and optimization of workloads throughout the environment.

Abstract: A memory device may receive, from a thread, a first request to read from a location that stores a first value, and may provide, to the thread, the first value based on receiving the first request. The memory device may receive, from the thread, a second request to cause the first value to be incremented to a second value, and may determine that the first value failed to be incremented. The memory device may provide, to the thread, information indicating that the first value failed to be incremented, and may increment a counter value to generate an incremented counter value. The memory device may receive, from the thread, a third request to cause the first value to be incremented to a third value, and may attempt, when the incremented counter value fails to satisfy a threshold, to cause the first value to be incremented to the third value.

Abstract: A hypervisor communicates with a guest operating system running in a virtual machine supported by the hypervisor using a hyper-callback whose functions are based on the particular guest operating system running the virtual machine and are triggered by one or more events in the guest operating system. The functions are modified to make sure they are safe to execute and to allow only limited access to the guest operating system. Additionally, the functions are converted to byte code corresponding to a simplified CPU and memory model and are safety checked by the hypervisor when registered with the hypervisor. The functions are executed by the hypervisor without any context switch between the hypervisor and guest operating system, and when executed, provide information about the particular guest operating system, allowing the hypervisor to improve operations such as page reclamation, virtual CPU scheduling, I/O operations, and tracing of the guest operating system.

Abstract: Disclosed are aspects of task assignment for systems that include graphics processing units (GPUs) that are virtual GPU (vGPU) enabled. In some examples, an algorithm is determined based on predetermined virtual machine assignment algorithms. The algorithm optimizes for a predetermined cost function. A virtual machine is queued in an arrival queue for assignment. A graphics configuration of a system is determined. The graphics configuration specifies a number of graphics processing units (GPUs) in the system. The system includes a vGPU enabled GPU. The algorithm is selected based on a correlation between the algorithm and the graphics configuration of the system. The virtual machine is assigned to a run queue based on the selected algorithm.

Abstract: The present disclosure relates to a method for a computer system comprising a plurality of processor cores including a first processor core and a second processor core, wherein a data item is exclusively assigned to the first processor core, of the plurality of processor cores, for executing an atomic primitive by the first processor core. The method includes receiving by the first processor core, from the second processor core, a request for accessing the data item, and in response to determining by the first processor core that the executing of the atomic primitive is not completed by the first processor core, returning a rejection message to the second processor core.

Abstract: The present disclosure relates to a dataflow optimization method for low-power operation of a multicore system, the dataflow optimization method including: a step (a) of creating an FSM including a plurality of system states in consideration of dynamic factors that trigger a transition in system states for original dataflow; and a step (b) of optimizing the original dataflow through optimization of the created FSM.

Abstract: A system, method and computer-readable media for managing a compute environment are disclosed. The method includes importing identity information from an identity manager into a module performs workload management and scheduling for a compute environment and, unless a conflict exists, modifying the behavior of the workload management and scheduling module to incorporate the imported identity information such that access to and use of the compute environment occurs according to the imported identity information. The compute environment may be a cluster or a grid wherein multiple compute environments communicate with multiple identity managers.

Abstract: Embodiments of an invention related to compacted context state management are disclosed. In one embodiment, a processor includes instruction hardware and state management logic. The instruction hardware is to receive a first save instruction and a second save instruction. The state management logic is to, in response to the first save instruction, save context state in an un-compacted format in a first save area. The state management logic is also to, in response to the second save instruction, save a compaction mask and context state in a compacted format in a second save area and set a compacted-save indicator in the second save area. The state management logic is also to, in response to a single restore instruction, determine, based on the compacted-save indicator, whether to restore context from the un-compacted format in the first save area or from the compacted format in the second save area.

Abstract: A managed object of a virtualized computing environment, which contains the runtime state of a parent virtual machine (VM) and can be placed in any host of the virtualized computing environment, is used for instantly cloning child VMs off that managed object. The managed object is not an executable object (i.e., the state of the managed object is static) and thus it does not require most of the overhead memory associated with a VM. As a result, this managed object can support instant cloning of VMs with a reduction in memory, storage, and CPU overhead relative to when a parent template VM is used.

Abstract: Systems, methods, apparatuses, and computer-readable media for guaranteed quality of service (QoS) in cloud computing environments. A workload related to an immutable log describing a transaction may be received. A determination is made based on the immutable log that a first compute node stores at least one data element to process the transaction. Utilization levels of computing resources of the first compute node may be determined. Utilization levels of links connecting the first compute node to the fabric may be determined. A determination may be made, based on the utilization levels, that processing the workload on the first compute node satisfies one or more QoS parameters specified in a service level agreement (SLA). The workload may be scheduled for processing on the first compute node based on the determination that processing the workload on the first compute node satisfies the one or more QoS parameters specified in the SLA.

Abstract: An approach for a hypervisor to throttle CPU utilization based on a CPU utilization throttling request received for a data flow is presented. A method comprises receiving a request for a CPU utilization throttling. The request is parsed to extract a CPU utilization level and a data flow identifier of the data flow. Upon receiving a data packet that belongs to the data flow identified by the data flow identifier, a packet size of the data packet is determined, and a rate limit table is accessed to determine, based on the CPU utilization level and the packet size, a rate limit for the data packet. If it is determined, based at least on the rate limit, that the CPU utilization level for the data flow would be exceeded if the data packet is transmitted toward its destination, then a recommendation is generated to drop the data packet.