Abstract: Various embodiments of the present disclosure provide methods, apparatuses, systems, computing devices, computing entities, and/or the like for determining optimized solutions to input problems in a containerized, cloud-based (e.g., serverless) manner. In one embodiment, an example method is provided. The method comprises: receiving a problem type of an input problem originating from a client computing entity; mapping the problem type to one or more selected solver types; generating one or more container instances of one or more compute containers, each compute container corresponding to a selected solver type; generating a problem output using the one or more container instances; and providing the problem output comprising a solution to the input problem to the client computing entity.

Abstract: A system allocates scratchpad memory (SPM) to heterogeneous devices for neural network computing. The system executes the operations of a global optimization manager. The global optimization manager receives compilation states from compilers, which compile corresponding subgraphs of a neural network model into corresponding subcommands that run on the heterogeneous devices. The global optimization manager unifies records of a same object across different ones of the compilation states, and allocates the SPM to the subgraphs according to the unified records of the compilation states.

Abstract: An information handling system includes a processor, first and second plug-in connector interfaces coupled to the processor, and first and second accelerator modules installed into respective first and second plug-in connector interfaces. The processor instantiates machine learning code. The information handling system instantiates a workload on the processor. The machine learning code determines a processing need of the workload, determines a first processing capability of the first accelerator module and a second processing capability of the second accelerator module, and allocates a processing resource of the first accelerator module to the workload based upon an evaluation of the processing need, the first processing capability, and the second processing capability.

Abstract: Described herein are systems, methods, and software to manage the migration of virtual machines between hosts. In one example, a management service for a cluster of hosts identifies a virtual machine to migrate and identifies current resource requirements of the virtual machine and current resource availability at the hosts of cluster. The management service further determines that two or more of the hosts possess current resource availability to support the current resource requirements of the virtual machine and identifies available headroom resources on each of the two or more hosts. Based on the available headroom resources, the management service further selects a host from the two or more hosts and initiates a migration of the virtual machine to the selected host.

Abstract: One or more aspects of the present disclosure relate to controlling threads across several processors. For example, at least one input/output (IO) workflow message can be stored in a storage array's hardware queue. Additionally, an IO workflow message can be read from the hardware queue. Further, a local thread wake-up or an interrupt-wakeup operation can be performed based on a target of the IO workflow message.

Abstract: The present application discloses a method, system, and computer system for automatically scaling task-processing capacity. The method includes obtaining input data from one or more data sources, determining, by a data layer, a set of the input data to process, obtaining a current measure of queued tasks and/or a task-processing capacity, determining a set of tasks to process based at least in part on the set of the input data to process, and automatically scaling the task-processing capacity based at least in part on (i) the current measure of the queued tasks, a measure of the task-processing capacity, and/or a measure of utilization of the task-processing capacity, and (ii) the set of tasks to process.

Abstract: One example method includes analyzing a load factor regarding a workload for one or more actors in a data storage platform, wherein the workload includes a group of two or more connected jobs, applying one or more criteria to an output of the load factor analyzing, based on the applying a criterion from the one or more criteria, determining how many actors should be included in a swarm of actors necessary to perform all of the connected jobs, spawning the swarm of actors, and load balancing the connected jobs across the swarm of actors.

Abstract: A plurality of virtual processing units associated with a physical processing unit is identified. Each of the plurality of virtual processing units is associated with a virtual machine of a plurality of virtual machines that run on respective virtual processing units in round-robin order using respective assigned execution time periods. A first overhead time value associated with running of the first virtual machine on a first virtual processing unit of the plurality of virtual processing units is obtained for a first virtual machine of the plurality of virtual machines. A second overhead time value associated with running of the second virtual machine on a second virtual processing unit of the plurality of virtual processing units is obtained for a second virtual machine of the plurality of virtual machines. The first overhead time value associated with the running of the first virtual machine and the second overhead time value associated with running of the second virtual machine are compared.

Abstract: Apparatuses, systems, and techniques to allocate memory based on a part of a sequence of items. In at least one embodiment, memory is allocated based on the size of a sliding window used to analyze images with neural networks.

Abstract: Distributing quantum jobs are disclosed. When a quantum processing unit is underutilized or when wait times are long, quantum jobs may be distributed from the job queue of one vendor to another vendor. This improves utilization and reduces wait times.

Abstract: A storage controller, including a processor configured to perform a plurality of tasks; and a scheduling module configured to schedule the plurality of tasks through reinforcement learning, and provide a scheduling result to the processor, wherein the scheduling module includes: a resource analysis module configured to analyze a usage history and a usage status of a resource; an access pattern analysis module configured to analyze an access pattern; a queue analysis module configured to analyze information included in a queue; and a performance analysis module configured to analyze a task execution result, wherein the scheduling module is further configured to perform the reinforcement learning using state information and reward information, and wherein the state information and the reward information are determined based on the performing of the plurality of tasks.

Abstract: As one aspect of the present disclosure, a memory access device is disclosed. The device comprises: a command manager configured to receive and manage one or more commands from a plurality of cores; a programming engine unit comprising one or more programming engines configured to perform calculations in response to the one or more commands; and a direct memory access (DMA) controller configured to perform at least one of read access and write access to a memory external to the memory access device in response to the one or more commands, wherein the command manager comprises a command memory configured to store the one or more commands, and a slot management unit configured to allocate one or more cores that correspond to the one or more commands, respectively, to the one or more programming engines in sequence.

Abstract: A method in a computing environment for application start up includes: generating a directed acyclic graph (DAG) for tasks that are performed during application start-up; performing topological sorting of the tasks based on the DAG to define an order for the tasks; allocating a tenant-specific priority queue; loading the tasks in the tenant-specific priority queue in the order determined by the topological sorting; allocating a plurality of computing threads to executing the tasks; executing the tasks using the plurality of computing threads that were allocated, wherein a plurality of the tasks are executed in parallel, and the tasks are assigned to the computing threads in an order based on the order in which the task are loaded in the tenant-specific priority queue; preparing a start page for use in accessing the application; and causing the start page to be displayed on a user access device for accessing the application.

Abstract: Apparatuses, systems, and techniques to allocate memory based on a part of a sequence of items. In at least one embodiment, memory is allocated based on the size of a sliding window used to analyze images with neural networks.

Abstract: The technology disclosed herein identifying a workflow file in a storage location associated with a trigger event responsive to detecting the trigger event, generating, using the identified workflow file, a new specification definition, storing, in a temporary service registry, the new specification definition, wherein the new specification definition is one of a plurality of temporary specification definitions in the temporary service registry, and identifying from the list of specification definitions the one or more specification definitions based on a parameter of a request responsive to receiving the request for one or more specification definitions of a list of specification definitions. The list of specification definitions including the plurality of temporary specification definitions in the temporary service registry and a plurality of predefined specification definitions in a remote service registry.

Abstract: There is provided a device for allocation of memory configured for: in response to a request for allocation of a region of a pool of a memory by a process being executed by a processor, a memory allocator is configured to perform in a single atomic operation: read a data structure indicating a state of allocation of the pool, check the data structure for a condition, update the data structure according to an outcome of the check, and return an address of the allocated region of the memory.

Abstract: A system for accelerator functions as a service is disclosed. The system may receive a job that includes a computer (CPU) portion and an accelerator portion. When the accelerator portion is performed, an execution time associated with a time for the accelerator to return results is determined. Resources or a portion thereof allocated to the job, or the CPU portion are freed or reallocated to another job at least during the execution time. The job is queued to receive resources when the results are received from the accelerator.

Abstract: Disclosed systems and methods align virtual machines (VMs) with hyperconverged infrastructure (HCI)-based hardware based on a determination of the VM's characteristic type. Each cluster node may be provisioned with a specific combination of hardware. In some embodiments, disclosed methods include determining, for each of the plurality of nodes, a target workload type based at least in part on the combination of information handling resources provisioned on each node. Disclosed methods may manage deployment of the VMs among the cluster nodes based on one or more factors including a workload compatibility factor determined in accordance with a workload type of each VM and a target workload type of each node, where the target workload type may reflect a hardware characteristic of the node.

Abstract: Systems and methods for generating runtime predictions in distributed computer architectures are disclosed. According to one embodiment, a method for generating runtime predictions in a distributed computer architectures may include: (1) receiving, by a runtime prediction computer program executed by an electronic device, training data regarding completion of a plurality of risk compute jobs; (2) extracting, by the runtime prediction computer program, bucketing statistics and instrument level features from the data; (3) training, by the runtime prediction computer program, a machine learning model with the training data, the extracted bucketing statistics, and the instrument level features; (4) receiving, by the runtime prediction computer program, a plurality of jobs for a period of time; and (5) calculating, by the runtime prediction computer program, an instrument cost for each of the jobs using the machine learning model.

Abstract: Processing methods is provided. A method is designated as schedulable. A completion stage is generated for the method. A return of the completion stage corresponding to the method is received. It is determined whether a state of the completion stage indicates continued execution of the method. In response to determining that the state of the completion stage does indicate continued execution of the method, a set of executions of the method is continued based on an execution return of the method being a null return.