30 results about "Cluster expansion" patented technology

The invention discloses a multi-tenant-oriented network architecture. The cloud network architecture comprises computing nodes, a virtual router cluster and a cloud gateway. Virtual machines which are included in the computing nodes perform message exchange with a public server in a private network through the virtual router cluster. Furthermore message exchange between the virtual machines and a public network is realized through the virtual router cluster and the cloud gateway, wherein the virtual router cluster comprises at least two virtual routers, and each virtual router transmits a same IP address to a private network switch. The cloud gateway comprises at least two gateway nodes. Each gateway node transmits an equivalent default router to the private network switch. Furthermore each gateway node transmits a same floating IP address to a public network router or a public network switch, thereby realizing cluster expansion of the cloud network architecture, preventing serviceability reduction of the whole network caused by fault of a single node and improving defensive capability of the network to attacks.

The invention relates to a method for extraction and deviation warning of a lane line. The method is characterized in that the method comprises the following steps that: an image preprocessing step is executed; more specifically, a camera is calibrated to obtain an inverse projection transmission picture, a single-line DLD feature image and a double-line DLDLD feature image are extracted, operators are extracted according to the two features and binarization processing is carried out on the operators, and clustering expansion processing is carried out on white pixels in the binarization images from bottom to up; continuous linear piecewise function fitting is carried out on the feature points in a predefined grid and coordinates of starting points and end points as well as corresponding gradients of all the piecewise functions are recorded; dashed and solid line determination is carried out on the feature points; and a multi-segment tracking step, a multi-segment extension step, an inspection step and a lane line determination step are carried out to determine a lane, thereby making a deviation warning decision. Advantages of the method are as follows: the lane line detection algorithm is suitable for various complicated urban roads; and because less computation is required and thus operation can be carried out under the circumstance of low-end hardware, the method is simple, secure and reliable.

The invention relates to a distributed acceleration method and system for a deep learning training task. The method comprises the following steps: (1) building a distributed GPU training cluster; (2)adopting a swap-in swap-out strategy, and adjusting a minibatch size on a single GPU working node in the distributed GPU training cluster is adjusted; (3) adjusting the learning rate according to theminibatch size determined in the step (2); And (4) carrying out deep learning training by adopting the hyper-parameters minibatch size determined in the steps (2) and (3) and the learning rate. On thepremise that the training accuracy is not affected, the communication time is greatly compressed simply and efficiently by reducing the number of times of parameter updating communication between clusters, and compared with a single GPU mode, the cluster expansion efficiency can be fully improved in a multi-GPU mode, and acceleration of the training process of the ultra-deep neural network modelis achieved.

The invention provides a data storage method and apparatus. The method comprises the steps of obtaining a to-be-stored first data file and a first LUN identifier corresponding to the first data file;according to first cluster view information, determining multiple first OSDs corresponding to the first LUN identifier, wherein the first cluster view information comprises a corresponding relationship between the first LUN identifier and the first OSDs; splitting the first data file into multiple data blocks, determining the first OSDs corresponding to the data blocks, and sending the data blocksto the corresponding first OSDs, thereby enabling the first OSDs to store the data blocks; and storing metadata corresponding to the first data file in a monitoring server. Through the technical scheme, the metadata quantity is greatly reduced; the metadata access frequency and quantity are reduced; the cluster access performance is improved; and storage resources occupied by the metadata are saved. Meanwhile, an expansion scheme based on zero data migration of a cluster view version is provided, so that the influence of the cluster expansion process on the system performance is reduced.

The invention provides a method and a device used for realizing cluster dynamic expansion in a distributed framework. The method comprises the steps of determining that a cluster dynamic expansion demand exists, wherein the cluster comprises multiple slave nodes, and the multiple slave nodes are divided into a calculation resource group and a storage resource group based on the provided resource services; and adjusting the number of the slave nodes in the calculation resource group and/or the storage resource group based on the cluster dynamic expansion demand. The technical scheme provided by the invention can better match a cluster size and processing capability with a demand for a storage resource and a calculation resource in an actual application scene, accordingly efficiently avoids the phenomenon of cluster resource shortage and cluster resource waste, and further improves the performance of the cluster while improving the flexibility of the cluster dynamic expansion.

The invention provides an expansion method for supporting non-halt of a multi-hash map database cluster system. The expansion method comprises the following steps: S1, binding all original tables in a cluster onto an existing hash map which is named old hash map; S2, during cluster expansion, executing a command for generating a new hash map, generating the new hash map through a cluster, and setting the hash map as a current hash map; S3, automatically binding tables which are newly created after cluster expansion onto the current hash map; S4, participating into operation with new tables after performing dynamic data redistribution according to the current hash map, wherein in any operation on the original tables and the new tables, the current hash map is taken as a hash map of current operation, and the original hash characteristics of old tables based on an old hash map are lost; S5, migrating a part of data onto a new node, and binding the tables onto the current hash map after the completion of migration. Through the design, performance attenuation is avoided, and natural data transition is realized.

A cloud communication based collection front-end system relates to a front end system. Currently, since the quantity scale of connection terminals of a power consuming collection system is large, fullrecord loading cannot be accepted for management of a single-node program memory. The technical scheme includes a communication gateway cluster, a communication front end machine cluster connected with the communication gateway cluster, a message queue cluster connected with the communication front end machine cluster and performing queue processing on data and a zookeeper distributed service cluster connected with the communication front end machine cluster. The distributed service cluster forms a zookeeper distributed service system. According to the technical scheme, a zookeeper distributed application program is adopted to achieve share of node manage and data share between front end machines. By adopting the technical scheme, collection front end cluster expansion becomes simple andefficient. The distributed expansion capability of the collection front end system is enhanced through cloud communication technology and load balance capability of the front end machine system is enhanced.

The invention discloses a BDCH-DBSCAN-based taxi passenger-carrying hotspot identification method. The method comprises the steps of firstly providing a large amount of taxi GPS data tracks, and deleting interference data from the data tracks; extracting get-on data points and get-off data points in the GPS data tracks; performing multi-thread block clustering on the extracted data points, separating out core points by adopting a DBSCAN algorithm, obtaining a neighbor node set of the core points, and performing cluster expansion operation on the core points, wherein cluster expansion refers tooperations of performing class cluster division on neighbor points of the core points and adding the neighbor core points with a number greater than a minimum cluster point number and domains of theneighbor core points into clusters; and according to a curved surface distance between two points, finding out neighbor nodes in a scanning radius, finding out a maximum density point in each clusterto serve as a cluster center, and cyclically clustering cluster center sets again until the set quantity and the precision meet the requirements. According to the method, large-scale data can be adapted; class cluster centers are convenient to identify; the multi-thread clustering speed is high; and the accuracy is high.

The embodiments of the invention provide a method, apparatus and system based on an SSL session. The method comprises the following steps: receiving a client greeting message sent by a client; according to the client greeting message, determining session identification content exists and obtaining a determination result; and according to the determination result, scheduling SSL servers in an SSL cluster of the SSL session of the client, and sending the client greeting message to the corresponding SSL server. By using the technical scheme provided by the invention, the method, apparatus and system have the following advantages: the hit rate of the SSL Session is improved, and cluster expansion is also realized.

The invention discloses a multi-purpose aircraft suitable for a multi-carrier environment and a combination body of the multi-purpose aircraft. The multi-purpose aircraft suitable for the multi-carrier environment achieves the multi-purpose multi-carrier function of the multi-purpose aircraft, when an aircraft body is in other states (land cruise or water cruise) other than flight, the power consumption amount is greatly reduced compared with the situation that the aircraft is in the flight state, the average power consumption amount of the multi-purpose aircraft suitable for the multi-carrierenvironment is reduced to a certain degree, and all-terrain all-day application base is provided for achieving cluster expansion of the multi-purpose aircraft of the multi-carrier environment. In addition, the multi-purpose aircraft has the multi-user and multi-role interactive scene, especially networked interactive application, and a lower cost and less power-consuming software and hardware platform is provided.

A high-expanded data center all-optically interconnected network system and a communication method are provided. A network system optical switch and a server of the invention adopt hierarchical network topological structure based on cluster expansion. The routing method comprises the following steps of: 1 generating a data packet; 2 calculating the communication wavelength of the data packet; 3 modulating the operating wavelength of a laser; 4 sending the data packets; 5 receiving the data packet by the optical switch; 6 determining whether the current optical switch satisfies a routing cut-off condition; 7 outputting the data packet from a routing port determined according to the routing method based on topology; 8 transmitting the data packet to a destination server; and 9 receiving the data packet by the destination server. The high-expanded data center all-optically interconnected network system has the advantages of simple structure, low communication delay and large throughput.

The embodiment of the invention discloses a cluster expansion method and device. The method comprises the steps of obtaining the target node group number of nodes in a cluster after transverse expansion is carried out; determining the maximum group number of the nodes contained in the same physical machine after the transverse expansion is carried out, according to the target node group number anda preset proportion; determining the physical machine of which group number of the currently contained nodes is smaller than the maximum group number as a target physical machine; determining the first node group number of currently to-be-allocated nodes based on the first group number of the nodes contained in a first physical machine and the maximum group number; allocating the nodes containedin the node groups of the first node group number within the target physical machine until the transverse expansion of the cluster is finished, otherwise, returning to perform the step of determiningthe physical machine of which group number of the currently contained nodes is smaller than the maximum group number as the target physical machine. Through application of the scheme, it is ensured that after the transverse expansion of the cluster is carried out, when all nodes contained in any physical machine crash due to a fault occurring in the physical machine of the cluster, the whole cluster still can work normally.