International Business Machines Corporation

Huawei Technologies Co., Limited

Robin Grosman

Ian R Finlay

Xun Xue

Calisto Paul Zuzarte

Garth Allen Dickie

Naresh K Chainani

Nattavut Sutyanyong

Abhinav Goyal

Adam J Storm

Chong Chen

Mark Francis Wilding

Yuk Kuen Chan

Alexander Abrashkevich

Per-Ake Larson

Dmitri Abrashkevich

Uttam Setty

Stephen Rees

Wenbin Ma

Yuanxi Chen

Ralph James Bateman

Yinghua (Amy) Tang

Anthony Anthony

Junhan Hu

James Trevor Nisbet

Jesse Ka-Leung Lee

The disclosed systems and methods are directed to generating cache IDs for each of a plurality of AI training pipelines, accessing training data elements included in a training data set, generating IDs corresponding to the training data elements, receiving the data IDs and an associated cache ID, randomizing the data IDs, selecting a subset of the randomized data IDs, fetching the training data elements previously cached by a cache node, receiving a portion of the training data elements present in the caching server corresponding to the subset of randomized data IDs, forwarding the portion of the training data elements present in the caching server to at least one consumer node, fetching the remaining training data elements associated with the subset of randomized data IDs from the training data set, and forwarding the remaining training data elements to at least one transformation node for training the neural network.

System, method and apparatus for intelligent caching

The disclosed systems and methods for intelligent heterogeneous computation directed to receiving monitoring data and a set of training data, wherein the monitoring data includes an occupancy rate of a preprocessed data queue and a utilization factor of accelerating devices, generating a resource computation job list in accordance with the monitoring data, forwarding jobs, in the resource computation job list to be executed on a central processing unit (CPU), to a CPU worker queue, forwarding control messages to the CPU worker queue, wherein the control messages are associated with jobs in the resource computation job list to be executed on the accelerating devices, and executing, by the accelerating devices, jobs in the resource computation job list to be executed on the accelerating devices.

System, method, and apparatus for selecting a CPU or an accelerator to preprocess data based on monitored information

Methods and systems are described for managing a shared database. One or more processing nodes may access a shared database. A common log node may manage the shared database. The common log node may validate database operations requested by the one or more processing nodes. During validation, the common log node may detect conflicts that occur between database operations requested by the one or more processing nodes.

Systems and methods for managing a shared database

This application provides a data processing method and apparatus. The method includes: receiving, by a master storage node, information that is about a first transaction and that is sent by a read-write node, where the information about the first transaction is used to request to perform a write operation on first data stored on the master storage node; determining, by the master storage node, the first data based on the information about the first transaction, and executing the first transaction; generating, by the master storage node, first transaction status metadata when the first transaction ends, where the first transaction status metadata includes identification information of expired data and identification information of the first transaction; and sending, by the master storage node, the first transaction status metadata to at least one read-only node. According to the data processing method and apparatus, a read delay of a read-only node can be eliminated.

Data processing method and apparatus

A database system comprises a plurality of servers with data storage. The system holds first and second database tables. Records of the first database table are distributed among the servers according to values of a first distribution column. The servers can receive access requests and log workload data representative of the requests. Based on the workload data, the servers can determine a performance impact associated with columns of the tables and, upon determining that a performance impact associated with one of the columns is greater than a performance impact associated with the first distribution column, the servers can re-distribute the records among the plurality of servers according to values of the second column.

Distributed database systems and methods of distributing and accessing data

Although many techniques provide ways to answer queries very quickly by either avoiding joins and/or aggregation or avoiding expression computations in order to apply filtering predicates, these techniques have significant potential overhead for maintenance, and require a great deal of storage space. The present invention discloses the use of synopsis or zone tables to expressions to provide powerful filtering capabilities to base tables across joins, and aggregations on join results, thus allowing for highly effective region or stride filtering and ordering aggregate ranges to select the most likely candidate tuples for joining to produce top-N or bottom-N result sets for queries.

Expression based synopsis of database tables

According to one embodiment of the present invention, a system may execute database queries and may comprise a computer system including at least one processor. The system may select a column of a first table as a candidate column for creating a dictionary, create tokens for a foreign key in a second table that corresponds to a primary key in the first table, generate the dictionary for the selected column using the created tokens as indexes in the dictionary and modify a query accessing the column of the first table to retrieve column values based on the dictionary. Embodiments of the present invention further include a method and computer program product executing database queries in substantially the same manner described above.

Query processing using a dimension table implemented as decompression dictionaries

An indication to load data into a database table is received. A determination is made whether an existing set of frequency distribution statistics is available. In response to determining that an existing set of frequency distribution statistics is available, the data is loaded into the database table using the existing set of frequency distribution statistics.

Method of reusing existing statistics to load database tables

A method for describing activities in software products is provided. The method provides for identifying a plurality of activities in a software product that are of interest, defining an event structure to describe each identified activity as an event, associating each identified activity with one or more problem determination technologies, executing the software product, and responsive to detecting one of the plurality of identified activities during execution of the software product, creating one or more event records for the one detected activity based on the defined event structure, initiating the one or more problem determination technologies associated with the one detected activity, and passing the one or more event records to each of the one or more problem determination technologies.

Description of activities in software products

Growing community of inventors

Markham, Canada

Robin Grosman