163 results about "Parallel scheduling" patented technology

The invention belongs to the field of cloud computing application, and relates to method for task rely relation description, decoupling, parallel scheduling and the like in cloud service. Rely task relations are provided, and a decoupling parallel scheduling method of rely tasks are constructed. The method comprises first decoupling the task rely relations with incoming degree being zero to construct a set of ready tasks and dynamically describing tasks capable of being scheduled parallelly at a moment; then scheduling the set of the ready tasks in distribution type and multi-target mode according to real time resource access so as to effectively improve schedule parallelism; and during the distribution of the tasks, further considering task execution and expenditure of communication (E/C) between the tasks to determine whether task copy is used to replace rely data transmission so as to reduce the expenditure of communication. The whole scheduling method can schedule a plurality of tasks in the set of the ready tasks in dynamic parallel mode, well considers performance indexes including real time performance, parallelism, expenditure of communication, loading balance performance and the like, and effectively improves integral performance of the system through the dynamic scheduling strategy.

The invention discloses a load-balanced sparse convolutional neural network accelerator and an acceleration method thereof. The accelerator comprises a main controller, a data distribution module, a convolution operation calculation array, an output result caching module, a linear activation function unit, a pooling unit, an online coding unit and an off-chip dynamic memory. According to the scheme provided by the invention, the high-efficiency operation of the convolution operation calculation array can be realized under the condition of few storage resources, the high multiplexing rate of input excitation and weight data is ensured, and the load balance and high utilization rate of the calculation array are ensured; meanwhile, the calculation array supports convolution operations of different sizes and different scales and parallel scheduling of two layers between rows and columns and between different feature maps through a static configuration mode, and the method has very good applicability and expansibility.

The invention relates to the technical field of cluster resource scheduling, and provides a method for scheduling containers in parallel in a cluster. The method comprises the steps of periodically obtaining cluster state information by a scheduler, wherein the cluster state information comprises resource information of all nodes in the cluster and description information of all containers in the cluster, the description information of the containers comprises container IDs of the containers, resource demands of the containers and deployment node IDs of the containers, and the resource information of the nodes comprises IDs of the nodes and resource quantities of the nodes; and according to the cluster state information, the container information of the containers scheduled by the scheduler and the container information of the containers scheduled by other schedulers, scheduling to-be-deployed containers to the nodes in the cluster, wherein the scheduled container information comprises container IDs and deployment node IDs of the scheduled containers. Through the scheme, the problem of resource conflicts in parallel scheduling is effectively reduced.

Featured is a method for reducing the contention of the highly contended global lock(s) of an operating system, hereinafter dispatcher lock(s) that protects all dispatching structures. Such a method reduces the need for acquiring the global lock for many event notification tasks by introducing local locks for event notifications that occur frequently among well defined, or consistent dispatcher objects. For these frequently occurring event notifications a subset of the dispatching structure is locked thereby providing mutual exclusivity for the subset and allowing concurrent dispatching for one or more of other data structure subsets. The method also includes acquiring one or more local locks where the level of protection of the data structure requires locking of a plurality or more of data structures to provide mutual exclusivity. The method further includes acquiring all local locks and/or acquiring a global lock of the system wide dispatcher data structures wherever a system wide lock is required to provide mutual exclusivity.

The invention proposes a task-level parallel scheduling method and system for a dynamically reconfigurable processor. The system comprises a main controller, a plurality of reconfigurable processing units, a main memory, a direct memory access device and a system bus, wherein each reconfigurable processing unit consists of a co-controller, a plurality of reconfigurable processing element arrays in charge of reconfigurable calculation and a plurality of shared memories used for data storage; the reconfigurable processing element arrays and the shared memories are adjacently arranged; and the shared memories can be read and written by the two connected reconfigurable processing element arrays around. According to the task-level parallel scheduling method and system for the dynamically reconfigurable processor, different scheduling modes can be executed for different tasks by adjusting the scheduling method, and basically all parallel tasks can be well accelerated in parallel in the reconfigurable processor.

Techniques for workload management in cloud computing infrastructures are provided. In one aspect, a method for allocating computing resources in a datacenter cluster is provided. The method includes the steps of: creating multiple, parallel schedulers; and automatically selecting a resource allocation method for each of the schedulers based on one or more of a workload profile, user requirements, and a state of the datacenter cluster, wherein an optimistic resource allocation method is selected for at least a first one or more of the schedulers and a pessimistic resource allocation method is selected for at least a second one or more of the schedulers. Due to optimistic resource allocation conflicts may arise. Methods to resolve such conflicts are also provided.

The invention provides a DAG (Directed Acyclic Graph) node optimal path-based multi-model parallel scheduling method and device. The method comprises the following steps of creating a DAG based on a relationship among models according to the relationship among the models; decomposing the DAG into a scheduling sequence set according to a path detection-based optimal path analysis algorithm, wherein the scheduling sequence set comprises a plurality of parallel scheduling sequences; saving all parallel scheduling sequences in the scheduling sequence set as a first adjacency matrix; mapping the first adjacency matrix into an adjacency list; running each sequence in the scheduling sequence set in parallel according to the adjacency list. According to the DAG node optimal path-based multi-model parallel scheduling method and device provided by the invention, the problems that the complexity is high and the running time is long in the existing decomposing method can be solved; the system resources can be fully utilized, and the multi-model scheduling running time is effectively shortened.

The invention discloses a distributed task scheduling method and system, belonging to the computer technical field. The method comprises the following steps: S1, after the client server starts up, alltasks are automatically scanned, a link is established with the intermediate server, and self-information and task information are reported to the intermediate server; S2, that intermediate server obtains the client server information and the task information, and splits a task into a plurality of independent sub-tasks according to the task configuration information, and distribute the task to each client server; S3, the client server executes the sub-tasks assigned by the intermediate server. Through integration of the task scheduling engine in the intermediate server, the dependency relationship of the tasks are conveniently managed. Through the intermediate server, a task is divided into several independent sub-tasks, and the client server is called to execute the assigned task items in parallel to achieve parallel scheduling.

The invention discloses a two-stage self-adaptive scheduling method suitable for a large-scale parallel data processing tasks. Two-stage scheduling is performed on tasks from a task stage and a subtask stage. By the method, the problem of parallel scheduling difficulty caused by complex dependence relationship among the tasks is solved effectively, parallelism degree is increased, orderly and efficient parallel processing of large-scale data processing tasks is realized, task waiting or executing time is reduced, and overall executing time is shortened. In addition, by the method, executor operation statistical information can be fed back to a scheduler, self-adaptive adjusting of executor pool size and task type is realized, and scheduling is constantly optimized, so that system resourceusing efficiency is improved.

Method for complex coloring based parallel scheduling for the switching network that is directed at traffic scheduling in the large scale high speed switching network. The parallel scheduling algorithm is on a frame basis. By introducing the concept of complex coloring which is optimal and can be implemented in a distributed and parallel manner, the algorithm can obtain an optimal scheduling scheme without knowledge of the global information of the switching system, so as to maximize bandwidth utilization of the switching system to achieve a nearly 100% throughput. The algorithm complexity is O(log² N).

The invention discloses a dependency mesh based instruction-level parallel scheduling method. The method includes the steps of 1), acquiring data dependency relations among instructions in a target basic block and information of function units corresponding to the instructions, and setting and computing data dependency priority values of the instructions according to the data dependency relations; 2), partitioning the instructions according to the data dependency priority values and the function units, storing partitioned results according to a mesh form, and establishing dependency meshes obtaining dependency relations between the instructions and data dependency priorities as well as between the instructions and the function units; 3), performing parallelism analysis among the instructions according to the relations, in the dependency meshes obtained in the step 2, between the instructions and the data dependency priorities as well as between the instructions and the function units. The dependency mesh based instruction-level parallel scheduling method has the advantages that the method is capable of describing parallel relations among the instructions and relativity between the instructions and hardware structures by combining the data dependency relations and function unit allocation relations, and is simple in implementation, wide in application range and high in instruction-level parallelism.

The invention discloses a data flow program scheduling method oriented to a multi-core system. The method comprises the following steps that task partitioning of mapping from calculation tasks to processor cores is determined, and software pipeline scheduling is constructed; according to structural characteristics of a multi-core processor and execution situations of a data flow program on the multi-core processor, cache optimization among the cores is conducted. According to the method, data flow parallel scheduling and optimization related to the cache structure of a multi-core framework are combined, and high parallelism of the multi-core processor is brought into full play; according to the hierarchical cache structure and the cache principle of the multi-core system, access of the calculation tasks to a communication cache region is optimized, and the throughput rate of a target program is further increased.

The invention discloses a heterogeneous computing platform. The platform comprises a host and a plurality of programmable devices, wherein the host is connected with each of the programmable devices; the host is used for initializing the programmable devices, carrying out parallel scheduling on each programmable device, sending computing data to each programmable device and obtaining a computing result; and each programmable device is used for processing the distributed computing data in parallel. The plurality of programmable devices of the heterogeneous computing platform can carry out computation at the same time, and the operation speed of the whole heterogeneous computing platform is equal to the sum of operation speeds of the programmable devices; compared with the heterogeneous computing platform with only one programmable device in the prior art, the whole operation speed and degree of parallelism of the heterogeneous computing platform are improved, so that the computing efficiency is improved, and the requirements, for the operation speed of the heterogeneous computing platform, of more and more complicated algorithms and data with larger and larger scales can be better satisfied. The invention furthermore provides an acceleration method on the basis of the heterogeneous computing platform.

The invention discloses a fault processing method and system as well as fault scheduling equipment. The processing method comprises the following steps: judging information related to faults in a fault complaint list; generating a parallel scheduling list containing the identification of the fault complaint list when the fault complaint list reaches upgrade standards; notifying a plurality of fault processing units in parallel; and simultaneously eliminating faults by the plurality of fault processing units, thus shortening the whole processing time for fault complaint lists with special requirements, and improving the processing ageing for important fault complaints with special requirements.

The invention relates to a monitoring video pedestrian real-time detection, attribute identification and tracking method and system based on deep learning. The invention mainly provides an efficient pedestrian detection, attribute identification and tracking method, designs an efficient scheduling method, and performs series-parallel scheduling on modules, so that the modules can perform multi-channel video pedestrian real-time detection, attribute identification and tracking as much as possible on limited computing resources. The problems in the prior art of single monitoring function of a road scene security and protection system and that monitoring efficiency is not high are solved. The efficient multifunctional security and protection system for road scene security and protection monitoring is provided. During application, through monitoring, attribute recognition and tracking on pedestrians, the monitoring strength and intensity are enhanced, especially for important places such as airports, stations and subways, crimes can be effectively restrained, and the personal and property safety of national people is guaranteed.

The invention provides a cloud resource scheduling method and system based on Kubernetes. The method comprises the following steps: creating a Pod according to a resource object creation request, so as to record a type of required schedulers in the resource object creation request into a configuration file of created Pod; after the created Pod is monitored, selecting schedulers from a pre-deployed scheduler group of a required scheduler type, and updating identifiers of the selected schedulers to a configuration file of the Pods, wherein the scheduler group comprises a plurality of schedulers of the same type; after the selected schedulers listen to their own own identifiers, determining a cluster node allocated to the Pod and sending a binding request; after the binding request is received through an access controller, verifying that resources still can meet requirements of information of required resources, and recording the identifiers of the cluster nodes to the Pod; after the corresponding cluster nodes monitor that the Pod records the identifiers of the cluster nodes, starting the Pod, and creating a corresponding container. According to the scheme, more schedulers can schedule cloud resources in parallel, and the resource scheduling efficiency is improved.

The invention relates to a distributed data integration job scheduling method and device. The method aims at a special scene possibly faced by data integration. A job scheduling device is responsiblefor issuing the data integration job to the job operation device, and the job operation device receives the scheduling task, starts job execution, feeds back job operation state information to the jobmanagement module, feeds back working node computing resources to the resource scheduling module, and feeds back lost or fault information to the job preloading module. The method has the following comprehensive characteristics: (1) high availability, fault tolerance and weak consistency; (2) a low delay characteristic for quasi-real-time job scheduling; (3) multi-tenant concurrency control oriented to the cloud service application; (4) computing resource isolation and multi-job parallel scheduling; and (5) a priority scheduling mechanism.

The invention relates to the technical field of a geospatial information system, and provides a three-dimensional space data parallel scheduling method based on data content. The method comprises the steps of receiving three-dimensional space data scheduling requests of a plurality of clients; resolving a three-dimensional space data scheduling task through a scheduling task parallel resolving method based on the data content, generating a three-dimensional space data scheduling sub-task which can be executed in parallel; distributing a usable task executing thread and a database connection for the three-dimensional space data scheduling sub-task, carrying out parallel inquiry of the three-dimensional space database; recombining the obtained data inquired by the three-dimensional space data scheduling sub-task and then returning to the clients. The efficiency of scheduling different types of three-dimensional space data from the three-dimensional space database is improved, and the real-time visualization ability of a three-dimensional geographic information system (GIS) is also improved by resolving and parallel executing of the three-dimensional space data scheduling task.

The present invention discloses a parallel scheduling system and method for distributed cluster load balancing, relating to the technical field of load balancing scheduling, and characterized in that: by virtualizing an IP address to receive a user request message, master servers of all clusters exchange information with one another by sending a Hello packet, to select a cluster with the maximum idle value to bind the virtualized IP address, and the cluster sends a Statement packet to notify the whole network after binding; and a master server of the cluster looks up a task table and calculates data in an idle table and an interactive table according to a certain calculation method, to obtain an optimal response cluster, and pushes the request message to the optimal response cluster by sending a Push packet, so as to implement the load balancing in response to request messages among peer-to-peer clusters. The master server of the cluster is provided with four tables including a task table, an idle table, an interactive table and a calculation result table which are used for scheduling request messages, and four packets including a Hello packet, a Push packet, an Exchange packet and a Statement packet are used for maintaining relationships among the clusters.

The embodiment of the invention provides a scheduling method, a scheduling device and a scheduling system for business data. For each node of a tree structure on which a scheduler is based, the actual number of bytes to be scheduled on request corresponding to each priority is reported to the parent node of the node; the actual number of bytes to be scheduled on request corresponding to each priority, reported by all sub nodes of a root node, is obtained, the number of actually-scheduled bytes corresponding to each priority of the root is calculated according to the preset maximum number of concurrently-scheduled bytes; with a preset scheduling rule, the number of actually-scheduled bytes corresponding to each priority of each node is obtained according to the number of candidate bytes to be scheduled corresponding to each priority of the corresponding node, and the number of scheduled bytes and the number of actually-scheduled bytes corresponding to each priority of the parent node of the corresponding node. The embodiment of the invention provides a parallel scheduling method of business data, the parallel scheduling on a plurality of nodes can be realized in each scheduling process, and the scheduling efficiency of the business data is accordingly improved.

The invention provides a real-time processing method of multiple video data on multi-GPU (multiple graphics processing unit) platform; the method mainly comprises the steps of 1, establishing a layered parallel structure based on task processors; 2, initializing the task processors in the layered parallel structure, and receiving monitoring video data and processing the monitoring video data in real time by each task processor; 3, carrying out environment monitoring, calculating task queue health level, task cache region health level and health level of each GPU according to the results of environment monitoring; 4, scheduling the task processors and distributing tasks according to the task queue health level, the cache region health level and the health level of each GPU; 5, repeating the steps 3 and 4 regularly so that balanced load is maintained among the GPUs. A complete parallel scheduling and data management scheme is designed for the real-time processing of multiple video multiple tasks on the multi-GPU platform, the great calculating capacity of a multi-GPU processor can be utilized efficiently, and calculating speed is increased greatly.