Chinese Academy of Sciences

Beijing Hili Technology Co. Ltd.

Shanghai 3ntv Network Technology Co. Ltd.

Microsoft Technology Licensing, LLC

Beijing Acell Technology Co., Ltd.

Beijing Intellix Technology Company Limited

Jiali You

Jinlin Wang

Lingfang Wang

Gang Cheng

Haojiang Deng

Yi Chen

Xue Ping Liu

Min Chu

Frank Kao-Ping Soong

Xiaozhou Ye

Feng Deng

Xiangdong Qi

Weining Qi

Yiqiang Sheng

Zhaolin Liu

Tingyi Li

Pufang Ma

A programmable network technology-based multi-homed network routing and forwarding method includes a data packet performing address-to-port matching a forwarding flow table address matching method; classifying addresses to matched ports corresponding to a destination address group; selecting a link information table and a port selection function, an appropriate port and a corresponding destination address group; cropping destination address groups of the remaining ports using an address filtering function and previous hop information carried in data to further crop the remaining destination address groups and addresses in the destination address group, a single destination address can be determined using a single address selection method, taking the single destination address using a universal routing and forwarding method to forward a packet to the destination address; otherwise, packaging the destination address group, and using a single address selection method to determine the single destination address hop by hop until the destination address is reached.

Routing and forwarding method for multi-homed network based on programmable network technology

A big data processing method based on a deep learning model satisfying K-degree sparse constraints comprises: step 1), constructing a deep learning model satisfying K-degree sparse constraints using an un-marked training sample via a gradient pruning method, wherein the K-degree sparse constraints comprise a node K-degree sparse constraint and a level K-degree sparse constraint; step 2), inputting an updated training sample into the deep learning model satisfying the K-degree sparse constraints, and optimizing a weight parameter of each layer of the model, so as to obtain an optimized deep learning model satisfying the K-degree sparse constraint; and step 3), inputting big data to be processed into the optimized deep learning model satisfying the K-degree sparse constraints for processing, and finally outputting a processing result. The method in the present invention can reduce the difficulty of big data processing and increase the speed of big data processing.

Big data processing method based on deep learning model satisfying k-degree sparse constraint

A distributed node intra-group task scheduling method includes: step) collecting resource state information and history service information about each node in a group; step) receiving a service request command and parsing the request command to obtain a request task; step) according to the resource state information and history service information about each node, calculating the resource utilization rate increment after each node in the group loads a task and the occupation condition of each kind of resource in the group; and according to a principle of making the resource utilization rate increment of each node in the group as low as possible and the occupation of each kind of resource in the group as balanced as possible, selecting service execution nodes in the group, and providing the request task by the selected service execution nodes.

Distributed node intra-group task scheduling method and system

A system and method for providing an on-site service includes a plurality of nodes, each containing: a neighborhood node set generation module, for generating a neighborhood node set on the basis of the bidirectional link bandwidth between a local node and a neighboring node; a neighborhood information index table generation module, for generating a neighborhood information index table of the local node; a candidate service point selection module, for selecting according to a selection function a candidate service node from the set of neighboring nodes; the definition of the selection function being: for a current service request, computing the difference between the QoS of the neighboring node i executing the service request and the QoS of the local node executing the service request; if the computed difference is smaller than a set threshold, the neighboring node i serving as the candidate service node; a service scheduling module, for receiving status information and feedback information provided in real time by the candidate service node, and selecting, on the basis of this information, a candidate node or the local node to serve as the service-executing node.

System and method for providing an on-site service

The present invention relates to a multi-protocol routing system, comprising a route determining module and a forwarding module which are located in network layer, wherein the route determining module is used for determining, based on the related information of applications and networks, the types of routable protocols and the next-hop path which are to be selected in the current routing phase; the forwarding module, including different routable protocols and corresponding routing information, is used for periodically calculating the newest remaining network resources and the state of link during a routing process and updating the forwarding table in current state. By integrating various characteristics and states of applications as well as different routable protocols, the present invention realizes real-time protocol selection and dynamic routing. Moreover, the present invention is easy to be expanded to other protocols, thus being adaptive to the continuous development and evolution of network services.

Multi-protocol routing system and method driven by application and network in convergence

The present invention relates to a Layer 2 adapter selecting system, including a Layer 3 packet requirements extracting module, a destination Layer 2 address acquiring module, a Layer 2 adapter metric acquiring module, an appropriate Layer 2 adapter identifier selecting module, a packet caching module and a packet scheduling module; wherein the packet caching module is configured to cache a received Layer 3 packet; the Layer 3 packet requirements extracting module is configured to read the Layer 3 packet and extract the type and a parameter value of the requirements; the Layer 2 adapter metric acquiring module is configured to acquire instant metric parameter values of each of the adapters; the appropriate Layer 2 adapter identifier selecting module calculates an adapter identifier and a Layer 3 packet metric; the destination Layer 2 address acquiring module is configured to acquire a Layer 2 address of each of the adapters associated with a destination Layer 3 address; and the packet scheduling module calls an external Layer 2 adapter driver, and the Layer 2 adapter driver completes packaging and sending of the packet. The present invention may solve the problem of resource waste of multiple links, so that the multiple links may be used dynamically in parallel.

System and method for selecting layer-2 adapter

A storage network structure comprises a primary distributed storage network (a cluster distributed storage network or a DHT (Distributed Hash Table)-based peer-to-peer storage network) and a Peterson graph-based storage network. The storage network structure provides different reliability classes for applications and combines the advantages of a peer-to-peer based distributed network storage with Peterson graph-based network storage, solve the problem of a large quantity of data migrating by utilizing the better media data migrating non-movable property from the Peterson graph-based network storage, and screen the single failure of the cluster storage structure and avoid performance degradation of peer-to-peer wide area storage by combining with the availability and robustness of Distributed Hash Table.

Storage network structure based on the Peterson graph and data read-write method thereof

The language of origin of a word or named entity is predicted using estimates of frequency of occurrence of the word or named entity in different languages. In one embodiment, the normalized frequency of occurrence of the word or named entity in a variety of different languages is estimated and the values are used as features in a feature vector which is scored and used to identify language of origin.

Growing community of inventors

Beijing, China

Jiali You