Abstract: Techniques are provided for implementing asynchronous checkpointing of in-memory data in a distributed computing system. For example, a method includes processing a stream of data records by an operator executing on a computing node, maintaining in a system memory, an operator state which is generated in response to the operator processing the stream of data records, and performing an asynchronous checkpointing process. The asynchronous checkpointing process includes enqueuing a checkpoint of the operator state in a first queue, wherein the first queue is maintained in the system memory, and executing a background worker thread to dequeue the checkpoint of the operator state from the first queue and store the checkpoint of the operator state in a data store. The operator continues with processing the stream of data records during the asynchronous checkpointing process.

Abstract: A method for managing container images in a container hosting environment, comprising a set of machine nodes for hosting container images, comprises the following steps. At a machine node of the set of machine nodes, at least one image layer of a container image comprising multiple image layers is identified for compaction. The identified image layer is compacted by merging data of the identified layer with data of another image layer of the container image. The data of the compacted image layer is transmitted to a backup store remote from the machine node for storage. The set of machine nodes and the backup store are implemented via processing devices operatively coupled via a communication network to form the container hosting environment. The compacted image layer may then be recovered from the backup store.

Abstract: A method for processing a computing task comprises: dividing multiple computing resources into multiple groups on the basis of topology information describing a connection relationship between the multiple computing resources; selecting at least one computing resource from at least one group of the multiple groups; determining processing performance of processing the computing task with the selected at least one computing resource; and allocating the at least one computing resource on the basis of the processing performance to process the computing task. Accordingly, the multiple computing resources can be utilized sufficiently, so that the computing task can be processed with better processing performance.

Abstract: A method for processing a computing task comprises: determining parameter data of multiple layers associated with a neural network model in response to receiving a computing task based on the neural network model; ranking at least one part of the multiple layers on the basis of the parameter data so as to obtain a layer list; ranking multiple computing resources on the basis of status information of the multiple computing resources so as to obtain a resource list; and determining a mapping between a corresponding layer among the multiple layers and a corresponding computing resource among the multiple computing resources on the basis of the layer list and the resource list, where the mapping indicates one computing resource among the multiple computing resources is to process parameters associated with one layer among the multiple layers.

Abstract: Techniques for managing a network system are disclosed. For example, a method comprises: receiving, at a network adapter, an access request from a client, the access request including first header information and first payload information; transmitting the first header information to a load balancing server associated with the network adapter; receiving the updated first header information from the load balancing server, wherein the updated first header information is determined by the load balancing server based on the received first header information and an identifier of the load balancing server, and wherein the updated first header information indicates a destination server to which the access request is to be transmitted; and, generating, based on the updated first header information and the first payload information, an updated access request. Accordingly, the processing capability of load balancing servers and the performance of load balancing servers in a network system may be improved effectively.

Abstract: An apparatus in one embodiment comprises a host device that includes at least one processor and an associated memory. The host device is configured to implement a plurality of containers each configured to access one or more portions of the memory. The containers are implemented as respective kernel control groups utilizing operating system level virtualization of the processor of the host device. The host device is further configured to assign the containers to groups in accordance with one or more designated criteria, and to generate checkpoints for respective groups of the containers. In conjunction with generation of a given one of the checkpoints for a particular one of the groups of containers, the host device identifies one or more pages of the memory that are shared by at least first and second containers of the particular group of containers, and generates the given checkpoint without duplicating the one or more shared pages to persistent storage.

Abstract: A method is used in managing deletion of replicas of files. A request is received to delete a replica of a file of a file system. The replica of the file represents a state of the file at a particular prior point in time. The replica is associated with a group including a set of replicas of the file system identified for deletion. Each replica is deleted in the group concurrently by using a delete marker. The delete marker is used for tracking deletion of each replica in the group.

Abstract: A graphics processing unit (GPU) service platform includes a control server, and a cluster of GPU servers each having one or more GPU devices. The control server receives a service request from a client system for GPU processing services, allocates multiple GPU servers within the cluster to handle GPU processing tasks specified by the service request by logically binding the allocated GPU servers, and designating one of the at least two GPU servers as a master server, and send connection information to the client system to enable the client system to connect to the master server. The master GPU server receives a block of GPU program code transmitted from the client system, which is associated with the GPU processing tasks specified by the service request, processes the block of GPU program code using the GPU devices of the logically bound GPU servers, and returns processing results to the client system.

Abstract: A plurality of data sets to be moved from a source site to a target site in a cloud computing platform is received at a plurality of a containerized data ingest components located at the source site. The received plurality of data sets are provided from the plurality of data ingest components to a staging cluster comprising a plurality of containerized broker components located at the source site, wherein the plurality of containerized broker components queue the plurality of data sets. The queued plurality of data sets are provided from the plurality of containerized broker components to a processing cluster comprising a plurality of containerized data processing components, wherein the plurality of containerized data processing components process the plurality of data sets. The plurality of data sets is transmitted from the plurality of containerized data processing components to the target site.

Abstract: An apparatus in one embodiment comprises a processing platform configured to communicate over a network with a plurality of Internet of Things (IoT) devices. The processing platform receives at least a first intermediate message from a first gateway of the network, receives one or more additional intermediate messages from each of one or more additional gateways of the network, associates the first and additional intermediate messages with one another based at least in part on a common message identifier detected in each such intermediate message, and processes the associated first and additional intermediate messages to recover a device message from a given one of the IoT devices. The first intermediate message is based at least in part on at least one application of a designated cryptographic function to the device message utilizing a corresponding key. At least one of the one or more additional intermediate messages provides at least a portion of the key.

Abstract: A method of scheduling a dedicated processing resource includes: obtaining source code of an application to be compiled; extracting, during compiling of the source code, metadata associated with the application, the metadata indicating an amount of the dedicated processing resource required by the application; and obtaining, based on the metadata, the dedicated processing resource allocated to the application. In this manner, performance of the dedicated processing resource scheduling system and resource utilization is improved.

Abstract: Implementations of the present disclosure relate to a method, apparatus and computer program product for processing a computing task. The method comprises: obtaining status information of multiple computing resources; in response to receiving a neural network model-based computing task, determining configuration information of multiple layers associated with the neural network model; obtaining parameter data associated with at least one part of the multiple layers on the basis of the configuration information; and based on the status information and the parameter data, selecting from the multiple computing resources a group of computing resources for processing the computing task. According to the example implementations of the present disclosure, multiple computing resources may be utilized sufficiently, and it may be guaranteed that a load balance may be stricken between the multiple computing resources.

Abstract: Implementations of the present disclosure relate to a method, apparatus and computer program product for processing a computing task. According to one example implementation of the present disclosure, there is provided a method for processing a computing task, comprising: in response to usage of multiple computing resources indicating that at least one part of computing resources among the multiple computing resources are used, determining a direction of a communication ring between the at least one part of computing resources; in response to receiving a request for processing the computing task, determining the number of computing resources associated with the request; and based on the usage and the direction of the communication ring, selecting from the multiple computing resources a sequence of computing resources which satisfy the number to process the computing task. Other example implementations include an apparatus for processing a computing task and a computer program product thereof.

Abstract: Embodiments of the present disclosure provide methods, apparatuses and computer program products for transmitting data. A method comprises determining, at a source node, a traffic type of a packet to be sent to a destination node, the source node and the destination node having therebetween a plurality of network paths for different traffic types. The method further comprises including a mark indicating the traffic type into the packet. In addition, the method further comprises sending the packet including the mark to the destination node such that the packet is forwarded along one of the plurality of network paths specific to the traffic type. Embodiments of the present disclosure can transmit data using different network paths based on different traffic types of data so as to optimize network performance for different network requirements.

Abstract: Systems and methods are provided to optimize checkpoint operations for deep learning (DL) model training tasks. For example, a distributed DL model training process is executed to train a DL model using multiple accelerator devices residing on one or more server nodes, and a checkpoint operation is performed to generate and store a checkpoint of an intermediate DL model. A checkpoint operation includes compressing a checkpoint of an intermediate DL model stored in memory of a given accelerator device to generate a compressed checkpoint, and scheduling a time to perform a memory copy operation to transfer a copy of the compressed checkpoint from the memory of the given accelerator device to a host system memory. The scheduling is performed based on information regarding bandwidth usage of a communication link to be utilized to transfer the compressed checkpoint to perform the memory copy operation, wherein the memory copy operation is performed at the scheduled time.

Abstract: Techniques are provided for implementing a parameter server within a networking infrastructure of a computing system to reduce the communication bandwidth and latency for performing communication synchronization operations of the parameter server. For example, a method includes executing a distributed deep learning (DL) model training process to train model parameters of a DL model using a plurality of worker nodes executing on one or more server nodes of a computing system, and executing a parameter server within a networking infrastructure of the computing system to aggregate local model parameters computed by the plurality of worker nodes and to distribute aggregated model parameters to the plurality of worker nodes using the networking infrastructure of the computing system.

Abstract: A method of managing a distributed system comprises: in response to receiving data that changes in a first memory region of a client, updating data in a second memory region with the received data based on a mapping between the first memory region and the second memory region; in response to receiving a calling request for a computing node in a server from the client, causing the computing node to perform a processing step based on the updated data in the second memory region; determining data that changes after the processing step in the second memory region; and transmitting, to the client, the data that changes after the processing step, such that the client updates the data in the first memory region based on the mapping.

Abstract: A resource management method comprises: in response to receiving, from an application operating on a client, a resource allocation request indicating an amount of dedicated processing resources required by the application, acquiring a mapping between a group of physical dedicated processing resources provided by a group of servers and a group of logical dedicated processing resources, the group of physical dedicated processing resources being divided into the group of logical dedicated processing resources; determining allocation statuses of the group of logical dedicated processing resources; determining, based at least on the mapping and the allocation statuses, a first amount of logical dedicated processing resources to be allocated to the application from the group of logical dedicated processing resources; and indicating the first amount of logical dedicated processing resources to the application, to allow the application to utilize physical dedicated processing resources provided by at least one of the

Abstract: An apparatus in one embodiment comprises a host device that includes at least one processor and an associated memory. The host device is configured to implement a plurality of processes each configured to access a shared region of the memory. The host device is further configured to establish a multi-process control group for the shared region, to maintain state information for the multi-process control group, and to track usage of the shared region by the processes based at least in part on the state information. At least a subset of the processes may comprise respective containers implemented utilizing operating system level virtualization of the processor of the host device. The multi-process control group established for the shared region illustratively comprises a coarse-grained control group having a granularity greater than a single page of the shared region.

Abstract: Techniques are provided for topology-aware provisioning of computing resources in a distributed heterogeneous environment.