International Business Machines Corporation

Srikumar Venugopal

Christian Pinto

Stefano Braghin

Michael Johnston

Pierre Lemarinier

Ioannis Gkoufas

Vasileios Vasileiadis

Michele Gazzetti

Yiannis Gkoufas

Burkhard Ringlein

Panagiotis Koutsovasilis

Muhammad Mustafa Rafique

Khalid Hasanov

Archit Patke

Alessandro Pomponio

An embodiment includes detecting a search request of a multi-dimensional search space by a system; responsive to detecting the search request, sampling representative parameters in the multi-dimensional search space based on a sampling threshold. The embodiment includes determining a parameter range of the representative parameters in the multi-dimensional search space. The embodiment also includes transforming a part of the multi-dimensional search space based on a statistical guarantee and the parameter range, where the part of the multi-dimensional search space is a refined space for a solution of the search request.

Refinement of large multi-dimensional search spaces

In an approach for storage, search, acquisition, and composition of a digital artifact, a processor obtains the digital artifact in a digital marketplace platform. The digital artifact is a collection of digital data with automatically generated and verifiable provenance and usage data. A processor transforms the digital artifact to define an access privilege. A processor shares the digital artifact in the digital marketplace platform by providing a view of a catalogue including the digital artifact. A processor authorizes a usage request based on the access privilege. A processor rewards a source of the digital artifact based on the usage of the digital artifact.

Digital artifact marketplace

A computer-implemented method, a computer program product, and a computer system for determining optimal data access for deep learning applications on a cluster. A server determines candidate cache locations for one or more compute nodes in the cluster. The server fetches a mini-batch of a dataset located at a remote storage service into the candidate cache locations. The server collects information about time periods of completing a job on the one or more nodes, where the job is executed against fetched mini-batch at the candidate cache locations and the mini-batch at the remote storage location. The server selects, from the candidate cache locations and the remote storage location, a cache location. The server fetches the data of the dataset from the remote storage service to the cache location, and the one or more nodes execute the job against fetched data of the dataset at the cache location.

Determining optimal data access for deep learning applications on a cluster

A method, computer system, and a computer program product are provided for query management optimization in a distributed data management system. In one embodiment, at least one query is received. The query is then analyzed and related information associated with the query is obtained. When information exists in a database relating to previously executed queries similar to the received query, that information is obtained. A query execution plan is then formulated using any existing information and information relating to the similarly previously executed queries.

Database query optimization based on analytics

A computer implemented method for optimizing performance of workflow includes associating each of a plurality of workflow nodes in a workflow with a data cache and managing the data cache on a local storage device on one of one or more compute nodes. A scheduler can request execution of the tasks of a given one of the plurality of workflow nodes on one of the one of more compute nodes that hosts the data cache associated with the given one of the plurality of workflow nodes. Each of the plurality of workflow nodes is permitted to access a distributed filesystem that is visible to each of the plurality of compute nodes. The data cache stores data produced by the tasks of the given one of the plurality of workflow nodes.

Optimization of workflows with dynamic file caching

Embodiments for processing data with multiple machine learning models are provided. Input data is received. The input data is caused to be evaluated by a first machine learning model to generate a first inference result. The first inference result is compared to at least one quality of service (QoS) parameter. Based on the comparison of the first inference result to the at least one QoS parameter, the input data is caused to be evaluated by a second machine learning model to generate a second inference result.

Methods and systems for sequential model inference

Embodiments for providing intelligent data replication and distribution in a computing environment. Data access patterns of one or more queries issued to a plurality of data partitions may be forecasted. Data may be dynamically distributed and replicated to one or more existing data partitions or additional of the plurality of data partitions according to the forecasting.

Intelligent data distribution and replication using observed data access patterns

Embodiments for managing distributed computing systems are provided. Information associated with operation of a computing node within a distributed computing system is collected. A reliability score for the computing node is calculated based on the collected information. The calculating of the reliability score is performed utilizing the computing node. A remedial action associated with the operation of the computing node is caused to be performed based on the calculated reliability score.

Methods and systems for proactive management of node failure in distributed computing systems

A job execution scheduling system and associated methods are provided for accommodating a request for additional computing resources to execute a job that is currently being executed or a request for computing resources to execute a new job. The job execution scheduling system may utilize a decision function to determine one or more currently executing jobs to select for resizing. Resizing a currently executing job may include de-allocating one or more computing resources from the currently executing job and allocating the de-allocated resources to the job for which the request was received. In this manner, the request for additional computing resources is accommodated, while at the same time, the one or more jobs from which computing resources were de-allocated continue to be executed using a reduced set of computing resources.

Selecting and resizing currently executing job to accommodate execution of another job

Embodiments for providing failure tolerance to containerized applications by one or more processors. A layered filesystem is initialized to maintain checkpoint information of stateful processes in separate and exclusive layers on individual containers. A most recent checkpoint layer is transferred from a main container exclusively to an additional node to maintain an additional, shadow container.

Growing community of inventors

Dublin, Ireland

Srikumar Venugopal