31results about How to "Ensure load balance" patented technology

The invention discloses a method for selecting an access carrier of an N-frequency point cell, which comprises: dividing the N carriers of the N-frequency point cell into two classes, namely full coverage carriers covering the whole cell and central coverage carriers covering only the central area of the cell in advance, wherein the full coverage carriers comprise the main carrier of the N-frequency point cell; determining the ratio flag2 of the number of the neighboring cells having a frequency different from that of the current cell to the number of the total number of the neighboring cells of the current cell; when UE selects an access carrier, calculating the priority of each carrier according to the residual physical resource, uplink interference level and the flag2 of each of the N carriers; determining a candidate access carrier group according to the priority of each carrier and selecting the carrier according to a priority sequence from high to low of the carriers in the candidate access carrier group; and allowing the UE to access the selected carrier in a time slot when the interference is the smallest. The method of the invention can realize the load balance of the carriers of the N-frequency point cell and improve system capacity.

The invention provides a multi-path routing method based on reinforcement learning and transfer learning, which is used for solving the technical problem of poor equivalent path load balance in a network environment with less flow data in the prior art, and comprises the following implementation steps of: constructing a real network Z and an experimental network G consistent with a Z topological structure; establishing a two-dimensional array H; constructing a multi-path routing model based on reinforcement learning; initializing a flow matrix DM and an equivalent path flow proportion matrix PM; performing iterative training on the multi-path routing model based on reinforcement learning in the experimental network G; migrating global neural network weight parameters in a routing decisionmodel obtained by training in an experimental network G to a real network Z based on a transfer learning method; and carrying out adaptive training on the global neural network initialized in the realnetwork Z to obtain a multi-path routing result conforming to the real network environment characteristics. The method can be used for data center networks and other scenes.

The invention provides an agent training method, a cross-domain heterogeneous environment task scheduling method and a related device. The agent training method comprises the following steps: acquiring pre-training data for simulating a task load in a real cross-domain heterogeneous environment; and training a deep reinforcement learning agent according to the pre-training data, so that the deep reinforcement learning agent is used for performing task scheduling on the cross-domain heterogeneous environment, and the deep reinforcement learning agent is constructed based on a D3QN model and a Markov decision process corresponding to a task scheduling problem of the cross-domain heterogeneous environment in advance. According to the task scheduling method and the task scheduling system, the intelligent agent which gives consideration to the real-time performance and the high efficiency of task scheduling can be constructed, the timeliness of task scheduling in a cross-domain heterogeneous environment by applying the intelligent agent can be effectively improved, the throughput of the whole task scheduling system can be increased, and computing resources provided in the cross-domain environment can be utilized to the maximum extent; therefore, a task scheduling result is ensured to consider the high efficiency of task processing and the load balance between the computing nodes.

The invention relates to a three-phase boost rectifier circuit and a control method thereof, and an uninterrupted power supply, which contains at least first battery packs BAT+, second battery packs BAT−, and a boost rectifier module; the boost rectifier module comprises a first bidirectional thyristor SCR1, a second bidirectional thyristor SCR2, a third bidirectional thyristor SCR3, a fourth bidirectional thyristor SCR6, a first unidirectional thyristor SCR4, a second unidirectional thyristor SCR5, a first inductor L1, a second inductor L2, a third inductor L3, a three-phase fully controlled rectifier bridge, a first capacitor C1 and a second capacitor C2. The invention can ensure the balance of positive and negative bus's voltage without the balancing device under battery operated boost mode, while improving the efficiency and reliability of the battery operated boost mode.

The invention relates to a loading module and a system and method for detecting the performance of an elevator safety protection device. The loading module is used for providing loads in the process of detecting the performance of the elevator safety protection device and comprises a clamp used for being fixedly connected with an elevator guide rail, wherein the clamp wraps the elevator guide rail in the circumferential direction of the elevator guide rail; a loader connected between the elevator safety protection device and the clamp and used for loading the elevator safety protection device; and a loading data collector used for collecting loading data of the loader. The loading module is fixedly connected with the guide rail, unbalance loading force cannot be formed, and loading balance can be ensured. The loading module is good in universality and can be applied to elevators with guide rails.

The invention discloses a parallel task partitioning method for an HEVC decoder. According to the method, processing of a network abstraction layer unit, decoding of a parameter set and header information and management work of a decoding image buffer area are executed by one control unit, entropy coding, inverse quantization, inverse transformation, intra-frame prediction, inter-frame prediction and deblocking filter required by a decoding and coding tree unit and pixel self-adaptive compensation are executed by a plurality of work units, and a steamline type work mode is adopted in the control unit and the work units. The parallel task partitioning method for the HEVC decoder can ensure load balance better and eliminate dependence on a lock structure in the synchronous process, so that the computing resource utilization rate is improved, and synchronous overhead is saved.

The invention discloses an S-CSCF (Serving-Call Session Control Function) allocation method based on service capability P2P (Peer-to-Peer) distribution. The method comprises the following steps: whether nodes joining in a Reload (resource location and discovery) overlay network are S-CSCF nodes or I-CSCF (Interrogating-Call Session Control Function) nodes is judged; if the joining nodes are S-CSCF nodes, the S-CSCF nodes join in the Reload overlay network, construct a topology structure, and release on the Reload overlay network according to a service capability set capable of being provided by the S-CSCF nodes; the S-CSCF nodes search in the Reload overlay network by taking all service capability values owned by the S-CSCF nodes as key values, so as to obtain an SIP URI (Session Initiation Protocol Uniform Resource Locator) set of all the S-CSCF nodes corresponding to the service capability values; a ping heartbeat packet test is carried out on each S-CSCF node corresponding to the SIP URI set at set intervals; whether the S-CSCF nodes subjected to the ping heartbeat packet test receive response or not five times is judged; and if the S-CSCF nodes subjected to the ping heartbeat packet test do not receive the response five times, the S-CSCF nodes without the response are deleted from the SIP URI set. The S-CSCF allocation method based on the service capability P2P distribution has the advantages that the S-CSCF load balancing of a system is realized in a service capability layer level, the service capability of the system is improved, and service delay of the system is reduced.

The invention discloses a large-scale cluster system and a building method thereof, which are used for improving high availability redundancy of the cluster system. The system mainly comprises a shared memory module, a sub-cluster and a dispatching node, wherein the shared memory module is used for providing a memory function and a data high availability service function; the sub-cluster is used for carrying out data mapping with the shared memory module, providing the data high availability service function and carrying out parallel input and output with outside; and the dispatching node is used for monitoring the working condition of the sub-cluster, maintaining the mapping relation between the sub-cluster and the shared memory module, and dispatching input and output load. The invention greatly improves high availability redundancy and stability of the large-scale cluster system.

The invention discloses a method for selecting an access carrier of an N-frequency point cell, which comprises: dividing the N carriers of the N-frequency point cell into two classes, namely full coverage carriers covering the whole cell and central coverage carriers covering only the central area of the cell in advance, wherein the full coverage carriers comprise the main carrier of the N-frequency point cell; determining the ratio flag2 of the number of the neighboring cells having a frequency different from that of the current cell to the number of the total number of the neighboring cellsof the current cell; when UE selects an access carrier, calculating the priority of each carrier according to the residual physical resource, uplink interference level and the flag2 of each of the N carriers; determining a candidate access carrier group according to the priority of each carrier and selecting the carrier according to a priority sequence from high to low of the carriers in the candidate access carrier group; and allowing the UE to access the selected carrier in a time slot when the interference is the smallest. The method of the invention can realize the load balance of the carriers of the N-frequency point cell and improve system capacity.

The invention relates to a three-phase boost rectifier circuit and a control, method thereof, and an uninterrupted power supply which contains at least first battery packs BAT+, second battery packs BAT−, and a boost rectifier module; the boost rectifier module comprises a first bidirectional thyristor SCR1, a second bidirectional thyristor SCR2, a third bidirectional thyristor SCR3, a fourth bidirectional thyristor SCR6, a first unidirectional thyristor SCR4, a second unidirectional thyristor SCR5, a first inductor L1, a second inductor L2, a third inductor L3, a three-phase fully controlled rectifier bridge, a first capacitor C1 and a second capacitor C2. The invention can ensure the balance of positive and negative bus's voltage without the balancing device under battery operated boost mode, while improving the efficiency and reliability of the battery operated boost mode.

The present invention proposes an intelligent routing decision-making method based on reinforcement learning, and specifically relates to an intelligent routing decision-making method based on DDPG reinforcement learning algorithm, which aims to design intelligent routing decisions by using reinforcement learning to balance the equivalent path traffic load, To improve the ability of the network to handle burst traffic, the present invention adopts an experience decision-making mechanism based on sampling probability, and the experience with poorer performance has a greater probability of being selected, thereby improving the training efficiency of the algorithm. In addition, adding noise to the neural network parameters is beneficial to the exploration of the system and improves the performance of the algorithm. The implementation steps are: 1) constructing the network topology; 2) constructing the network topology G 0 3) Construct a routing decision model based on the DDPG reinforcement learning algorithm; 4) Initialize the traffic demand matrix DM and the equivalent path traffic ratio matrix PM; 5) Iterate the routing decision model based on reinforcement learning train. The present invention can be used in scenarios such as a data center network.

The invention discloses a method for improving equipment reliability, wherein the equipment comprises at least two pairs of heat-switching primary and standby functional boards and a global control module connected with the primary and standby functional boards, and the method comprises the following steps of: reporting the processing load index of each pair of primary and standby functional boards onto the global control module by each pair of primary and standby functional boards; allocating the business of a user by the global control module according to the processing load index reported by each pair of primary and standby functional boards. In the invention, the global control module is added on the basis of 1+1 primary and standby functional board distributed processing, thereby ensuring the load equilibrium at a resource allocation stage through the business allocation of the global control module.

The invention discloses an S-CSCF (Serving-Call Session Control Function) allocation method based on service capability P2P (Peer-to-Peer) distribution. The method comprises the following steps: whether nodes joining in a Reload (resource location and discovery) overlay network are S-CSCF nodes or I-CSCF (Interrogating-Call Session Control Function) nodes is judged; if the joining nodes are S-CSCF nodes, the S-CSCF nodes join in the Reload overlay network, construct a topology structure, and release on the Reload overlay network according to a service capability set capable of being provided by the S-CSCF nodes; the S-CSCF nodes search in the Reload overlay network by taking all service capability values owned by the S-CSCF nodes as key values, so as to obtain an SIP URI (Session Initiation Protocol Uniform Resource Locator) set of all the S-CSCF nodes corresponding to the service capability values; a ping heartbeat packet test is carried out on each S-CSCF node corresponding to the SIP URI set at set intervals; whether the S-CSCF nodes subjected to the ping heartbeat packet test receive response or not five times is judged; and if the S-CSCF nodes subjected to the ping heartbeat packet test do not receive the response five times, the S-CSCF nodes without the response are deleted from the SIP URI set. The S-CSCF allocation method based on the service capability P2P distribution has the advantages that the S-CSCF load balancing of a system is realized in a service capability layer level, the service capability of the system is improved, and service delay of the system is reduced.

The invention provides a method and a device for improving video image access speed, equipment and a medium. The method comprises the following steps: acquiring the number of operation units and the number of CPU cores in a video image access system; receiving an access request containing a to-be-processed service, selecting a corresponding number of operation units and a corresponding number of CPU cores to execute different load strategies according to different types of the to-be-processed service, when the to-be-processed service is detected to be a video image access service, using the operation units not greater than the number of the CPU cores to disperse the load, and distributing the specified load to the corresponding CPU core for video image access processing; and when the to-be-processed service is detected to be the auxiliary service, dispersing the load by using the operation unit with a preset numerical value according to a concurrent distribution principle, and ensuring the concurrent auxiliary service to be dispersed to a part of the CPU cores to be processed. According to the invention, the concurrency and migration overhead of a large number of continuous time slice operation units is avoided, and the request response, the processing speed and the concurrency amount in the video image access system are improved.

The invention discloses an increment time slice adjustment framework and method in a CREDIT scheduler. An interrupt decision making information maintaining device is responsible for maintaining history interrupt decision making information, and an increment time slice adaptive adjustment device collects data information maintained in the interrupt decision making information maintaining device in a timing mode, adjusts history information in cooperation with an increment time slice and dynamically calculates a new increment time slice. If the newly-calculated increment time slice is different from the original increment time slice, the CREDIT scheduler is informed to update the increment time slice. As an extension of an original time slice mechanism, the increment time slice dynamically maintains a non-negative value to increase the length of the original time slice. The increment time slice mechanism is built on a virtual interrupt independent decision making mechanism, and therefore the length of the time slice can be increased on the premise that the I / O performance of a virtual machine is not influenced.

The present application provides a method, device, device and medium for improving video image access speed, the method includes: acquiring the number of operating units and the number of CPU cores in a video image access system; receiving an access request containing a service to be processed , select the corresponding number of operating units and the number of CPU cores to implement different load strategies according to the type of the business to be processed. When it is detected that the business to be processed is a video image access business, use the number of operating units not greater than the number of CPU cores to disperse Load, the specified load is allocated to the corresponding CPU core for video image access processing; when it is detected that the service to be processed is an auxiliary service, the running unit with a preset value is used to distribute the load according to the concurrency allocation principle, so as to ensure that the concurrent auxiliary service is distributed to some parts. processed on the CPU core. The present application avoids the concurrency and migration overhead of a large number of continuous time slice running units, and improves the request response, processing speed and concurrency in the video image access system.

The present invention proposes a multi-path routing method based on reinforcement learning and transfer learning, which is used to solve the technical problem of poor load balance of equivalent paths in a network environment with less traffic data existing in the prior art, and the implementation steps are as follows : Construct the real network Z and the experimental network G consistent with the topological structure of Z; establish a two-dimensional array H; construct a multi-path routing model based on reinforcement learning; initialize the traffic matrix DM and the equivalent path traffic ratio matrix PM; The multi-path routing model based on reinforcement learning is iteratively trained; the weight parameters of the global neural network in the routing decision model trained in the experimental network G are transferred to the real network Z based on the transfer learning method; The initialized global neural network performs adaptive training to obtain multi-path routing results that conform to the characteristics of the real network environment. The present invention can be used in scenarios such as a data center network.

The invention discloses a container image parallel distribution method, scheduler and storage medium. The method includes: using the node where the data block in the node scheduling queue is located as the distribution node, and transmitting the data block to the requesting node; receiving the requesting node pushed by the super node Confirmation information that the data block write disk operation is successful, and insert the request node into the node scheduling queue; in the scheduling window, according to the data block information, the distribution node and the request node are scheduled to match: if the scheduling match is successful, the request node is controlled Download blocks of data from the distribution node. The present invention can ensure load balance and improve large-scale file distribution efficiency.

The invention relates to a performance detection system and method of a loading module and an elevator safety protection device. The loading module is used to provide load during the performance detection process of the elevator safety protection device, including: a clamper, which is used for fixed connection with the elevator guide rail, and the clamper surrounds the elevator guide rail along the circumferential direction of the elevator guide rail; the loader , connected between the elevator safety protection device and the clamper, and load the elevator safety protection device; and a loading data collector, used to collect the loading data of the loader. The loading module is fixedly connected to the guide rail, which does not form an eccentric load force and can ensure load balance. Moreover, the loading module has good versatility, and it can be applied to any elevator with guide rails.