Google Inc.

Facebook, Inc.

The UAB Research Foundation (the University of Alabama at Birmingham)

Grzegorz Malewicz

Grzegorz Jan Czajkowski

James C Dehnert

Matthew Harold Austern

Aart J C Bik

Joshua L Levenberg

Marian Dvorsky

Derek P Thomson

Christopher B Colohan

Naty Leiser

Ilan Horn

Mohamed Elbassiony Mohamed Abou El Alaa Khalefa

Charles Reiss

Aart Jc Bik

Data are maintained in a distributed computing system that describe a graph. The graph represents relationships among items. The graph has a plurality of vertices that represent the items and a plurality of edges connecting the plurality of vertices. At least one vertex of the plurality of vertices includes a set of label values indicating the at least one vertex's strength of association with a label from a set of labels. The set of labels describe possible characteristics of an item represented by the at least one vertex. At least one edge of the plurality of edges includes a set of label weights for influencing label values that traverse the at least one edge. A label propagation algorithm is executed for a plurality of the vertices in the graph in parallel for a series of synchronized iterations to propagate labels through the graph.

Label propagation in a distributed system

Embodiments are disclosed for using an improved locality sensitive hashing (LSH) operation for the K-means clustering algorithm. In some embodiments, parameters of an LSH function are optimized with respect to a new cost model. In other embodiments, an LSH operation is applied with optimized parameters to a K-means clustering algorithm.

Clustering using locality-sensitive hashing with improved cost model

Data are maintained in a distributed computing system that describe a directed graph representing relationships among a set of items. The directed graph models a condition having an associated problem. The directed graph has graph components having associated data fields. The relationships are analyzed to identify a solution to the problem. As part of the analysis, a new value for the data field associated with a graph component is identified responsive to an operation performed during the analysis. The new value is compared with an existing value of the data field, and the data field is modified. The modification may include inserting the new value into an overflow vector of data, and replacing the existing value in the data field with exception information identifying the location of the new value. An exception flag associated with the data field is set to indicate that the exception information is being used.

Data storage in a graph processing system

A large-scale data processing system and method including a plurality of processes, wherein a master process assigns input data blocks to respective map processes and partitions of intermediate data are assigned to respective reduce processes. In each of the plurality of map processes an application-independent map program retrieves a sequence of input data blocks assigned thereto by the master process and applies an application-specific map function to each input data block in the sequence to produce the intermediate data and stores the intermediate data in high speed memory of the interconnected processors. Each of the plurality of reduce processes receives a respective partition of the intermediate data from the high speed memory of the interconnected processors while the map processes continue to process input data blocks an application-specific reduce function is applied to the respective partition of the intermediate data to produce output values.

System and method for limiting the impact of stragglers in large-scale parallel data processing

Data are maintained in a distributed computing system that describe a directed graph representing relationships among items. The directed graph has a plurality of vertices representing the items and has edges with values representing distances between the items connected by the vertices. A multiple reference point algorithm is executed for a plurality of the vertices in the directed graph in parallel for a series of synchronized iterations to determine shortest distances between the vertices and the source vertex. After executing the algorithm on the vertices, value pairs associated with the vertices are aggregated. The aggregated value pairs indicate shortest distances from the respective vertices to the source vertex. The aggregated value pairs are outputted.

Multiple reference point shortest path algorithm

Executing a confined recovery in a distributed system having a plurality of worker systems including a failed worker system at a current superstep. The confined recovery includes determining states of the partitions of the worker systems during the supersteps preceding the current superstep, and determining a recovery initiation superstep preceding the current superstep in which all messages for recovery initiation superstep are available. The recovery initiation superstep is determined responsive to determining the states of the partitions. Additionally, a recovery set of partitions is determined for which messages in supersteps after the recovery initiation superstep are not available. The worker systems having the partitions in the recovery set are instructed to execute the defined function for the partitions in the recovery set starting at the recovery initiation superstep to recover the lost exchanged messages.

Confined recovery in a distributed computing system

Embodiments are described for determining and/or estimating a nearest neighbor to a data vector in a dataset are presented. Some embodiments redistribute data vectors between clusters based upon the character of the clusters to more evenly balance the computational load. Some embodiments employ Locality Sensitive Hashing (LSH) functions as part of the clustering and remove redundant data vectors from the data set to mitigate unbalanced computation. The disclosed embodiments may facilitate the analysis of very large and/or very high dimensional datasets with reasonable runtimes.

Nearest neighbor clustering determination and estimation algorithm that hashes centroids into buckets and redistributes vectors between clusters

A value is distributed in a distributed computing system having a master system in communication with a plurality of worker systems. Partitions of a graph are assigned to the worker systems. The graph represents relationships among a set of tangible items that model a real-world condition having an associated problem. Configuration information is determined that describes a configuration of the distributed computing system. A distribution scheme is selected for distributing a value from the master system to the plurality of worker systems based on the configuration information. The value is distributed from the master system to the worker systems according to the selected distribution scheme. The worker systems are configured to use the value to produce an output representing a solution to the real-world problem.

Growing community of inventors

Tokyo, Japan

Grzegorz Malewicz