Tsinghua University

Yongwei Wu

Kang Chen

Mingxing Zhang

Teng Ge Ma

Jinlei Jiang

Weimin Zheng

Ke Yang

Yeh-Ching Chung

Chengying Huan

Tuoyu Gong

Mengxing Liu

Lidong Zhang

Maomeng Su

Shaonan Ma

Yingdi Shan

Shuke Wang

Systems and methods are provided for performing random walk graph computing. One method may comprise generating a subset of walkers on a graph, maintaining the generated subset of walkers in a walker pool in a memory, loading a coarse-grained block of the graph from a non-volatile storage into a block buffer of the memory, generating pre-sampled edges for vertices in the coarse-grained block, storing the pre-sampled edges into a pre-sampled edge buffer allocated for the coarse-grained block and moving one or more walkers of the generated subset of walkers using the pre-sampled edges stored in the pre-sampled edge buffer. The generated subset of walkers may have an initial number determined based on a memory space allocated to the walker pool.

Systems and methods for general-purpose out-of-core random walk graph computing

Systems and methods are provided for persisting a data structure. One method may comprise, at a front-end node in a computing system: generating a data structure operation record for a data structure operation directed to a data structure persisted in a non-volatile memory (NVM) in a back-end node of the computing system, appending the data structure operation record in an operation log, generating a transaction record for a transaction that includes a plurality of memory operations that collectively accomplishing the data structure operation, appending the transaction record in a transaction log, flushing the transaction log to the back-end node after flushing the operation log; and at the back-end node of the computing system: persisting received operation log and received transaction log in the NVM, and accomplishing the data structure operation by performing the plurality of the memory operation records with the data structure operation record as a commit signal.

Systems and methods for implementing persistent data structures on an asymmetric non-volatile memory architecture

Systems and methods are provided for performing temporal graph computing. One method may comprise receiving an input temporal graph that have a plurality of edges with each edge connecting from one vertex instance to another vertex instance, generating in-vertices and out-vertices for each vertex instance, merging the in-vertices and out-vertices into hub vertices for each vertex instance and generating a directed acyclic graph (DAG), receiving a minimum path problem, and scanning the DAG once to provide a solution to the minimum path problem. The merging of vertices and generation of the DAG may be performed by sorting all out-vertices using a 2-dimensional radix sort, generating a respective set of hub vertices for each vertex instance, relabeling the in-vertices and the out-vertices to their respective hub vertices for each vertex instance by a parallel binary search and updating edges affected by the relabeling, and assembling relabeled edges and vertices.

Systems and methods for general-purpose temporal graph computing

Systems and methods are provided for an erasure coding object storage system. One method may comprise receiving an object for storing in an erasure coding object storage system, partitioning the object into a plurality of chunks including a first chunk and a second chunk. The first chunk may have a first chunk size and fall into a first bucket, the second chunk may have a second chunk size equal to the first chunk size multiplied by a ratio q that is larger than one and fall into a second bucket. The method may further comprise encoding the first bucket using a regenerating code with the first chunk size, encoding the second bucket using the regenerating code with the second chunk size and storing the encoded first and second buckets in a plurality of nodes of the erasure coding object storage system.

Systems and methods for geometric partitioning in object storage

Provided are a hybrid distributed file system architecture structure, an applied file storage processing method, a dynamic migration method, and application thereof. The file storage processing method comprises: acquiring storage attributes of a file to be stored, wherein the storage attributes at least include a size of the file; determining, according to a pre-configured storage rule and the attributes of the file to be stored, in which distributed file system the file to be stored is stored; and storing the file to be stored in the determined distributed file system. The method further comprises migrating, according to a predetermined policy, a file that has been stored in a predetermined storage location. The device intelligently selects a file underlying storage policy according to file feature attributes to decide whether to migrate the file and to which file system the file is migrated so as to satisfy usage equalization of different file systems and also minimize performance degradation. By means of experimental comparison, it is concluded that the present disclosure can greatly improve comprehensive file performances such as I/O performance and the usage equalization of the file system.

Hybrid file system architecture, file storage, dynamic migration, and application thereof

Systems and methods are provided for performing a remote procedure call. One method may comprise, at a client device, generating a request including setting a status field in a request header to indicate to a server processor that the request is ready, writing the request to a server memory via a RDMA write operation and fetching a response generated by the server processor from the server memory via a RDMA read operation. The method may further comprise, at a server device, checking a mode flag to determine that an operation mode is set to repeated remote fetching, retrieving the request from a server memory, processing the request to generate a response and writing the response to the server memory for the response to be fetched by a client device. The response includes a response header that comprises a status field for the status of the response and a response time.

Growing community of inventors

Beijing, China

Yongwei Wu