This disclosure describes techniques and mechanisms for enabling a user to run heavy deep learning workloads on standard edge networks without off-loading computation to a cloud, leveraging the available edge computing resources, and efficiently partitioning and distributing a Deep Neural Network (DNN) over a network. The techniques enable the user to split a workload into multiple parts and process the workload on a set of smaller, less capable compute nodes in a distributed manner, without compromising on performance, and while meeting a Service Level Objective (SLO).

Approaches presented herein enable dynamic optimization of a degree to which a query is parallelized for execution. More specifically, a priority associated with an obtained user query for execution is identified. A real-time metric indicating availability of one or more runtime resources is checked. An optimal degree of parallelism is calculated based on the priority associated with the obtained user query and the real-time availability metric. A plan is generated for executing the query using the calculated optimal degree of parallelism.

A computer system is constituted by a plurality of physical computers including a first physical computer and a second physical computer. One or more application instances that perform an application service and a storage service instance that provides a storage service including a volume used by the application instance operate on the first physical computer. The computer system predicts a future resource usage status of the first physical computer, creates a plan to move the one or more application instances operating on the first physical computer to the second physical computer based on the predicted future resource usage status, and executes the created plan.

Technologies for providing a multi-tenant local breakout switching and dynamic load balancing include a network device to receive network traffic that includes a packet associated with a tenant. Upon a determination that the packet is encrypted, a secret key associated with the tenant is retrieved. The network device decrypts a payload from the packet using the secret key. The payload is indicative of one or more characteristics associated with network traffic. The network device evaluates the characteristics and determines whether the network traffic is associated with a workload requesting compute from a service hosted by a network platform. If so, the network device forwards the network traffic to the service.

A network-based control method for power consumption of an application, a terminal device and a non-transitory computer-readable storage medium are disclosed. The network-based control method may include: identifying a target application having a frequent wakeup or heartbeat detection behavior in response to a freezing function being enabled; monitoring whether a peer server to which the target application is connected is accessible; and freezing the target application by the freezing function in response to the peer server being inaccessible.

A computer-implemented resource allocation method is provided, which comprises, in a computing environment comprising a resource management unit and a cluster comprising a cluster management node and a cluster node running an application program: receiving, by the resource management unit, a request for allocating one or more system resources to the application program; retrieving, by the resource management unit, from the cluster management node, an identifier of the cluster node running the application program; dynamically updating system physical resources allocated to the cluster node by updating a resource allocation file managed by an operating system of a computing machine on which the cluster is running, based on the identifier of the cluster node and the received request.

A system and a method for routing a message to an application over a connection oriented session in a Kafka messaging platform environment are provided. The method includes: acquiring a plurality of partitions from the Kafka messaging platform; designating a first partition from among the plurality of partitions as a sticky partition; generating a plurality of routing keys that are configured to route to the sticky partition; receiving a subscription from a service that corresponds to a first application; transmitting, to the first application, a first routing key that identifies the subscription from among the plurality of routing keys; and receiving messages from Kafka services that are routed by the first routing key to the first application. For any particular application or set of applications, a plurality of connection oriented sessions may be used to achieve load balancing and high availability.

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.

A method of initiating a transfer of an active first-type slave process, executed in a first processing entity of an edge cloud computing layer, to a second processing entity of the edge cloud computing layer, includes, at a first mobile entity, receiving a first heat map relating to the first-type master-slave process, ranking, based on a cost function, possible process sharing connections, between the first mobile entity and one or more second processing entities, for the current location of the first mobile entity and/or a location of the first mobile entity in the near future, determining, based on the ranking, one or more second processing entities as potential target processing entities to transfer the first-type slave process to, and transmitting a processing entity transfer request to a control process executed in the edge cloud computing layer. The request includes an identification of the active first-type slave process and indicates at least one of the second processing entities determined, based on the ranking, as potential target to transfer the active first-type slave process to.

Methods, apparatus, systems, and articles of manufacture are disclosed to improve workload domain management of virtualized server systems. An example apparatus includes a resource pool handler to generate a pool of virtualized servers including a first virtualized server based on a policy, ones of the virtualized servers to be allocated to a workload domain to execute an application, a resource status analyzer to determine a health status associated with the workload domain and determine whether the health status satisfies a threshold based on the policy, and a resource allocator to allocate the first virtualized server to the workload domain to execute the application when the health status is determined to satisfy the threshold.