Microsoft Technology Licensing, LLC

Konstantinos Karanasos

Carlo Aldo Curino

Sriram S Rao

Subramaniam Venkatraman Krishnan

Rathijit Sen

Matteo Interlandi

Srikanth Kandula

Yuan Yu

Alekh Jindal

Hiren Shantilal Patel

Milan Vojnovic

Raghunath Ramakrishnan

Kwanghyun Park

Christopher William Douglas

Jeffrey Thomas Rasley

Yiwen Zhu

Markus Weimer

Fotios Psallidas

Dalitso Hansini Banda

Jesus Camacho Rodriguez

Sudhir Darbha

Isha Tarte

Dong He

Rodrigo Lopes Cancado Fonseca

Supun Chathurang Nakandala

Joana Matos Fonseca Da Trindade

Ivan Popivanov

Doris Suiyi Xin

Andrew F Chung

Karla Jean Saur

Example aspects include techniques for query processing over deep neural network runtimes. These techniques may include receiving a query including one or more query operators and determining a query representation based on the one or more query operators. In addition, the techniques may include determining a neural network program based on the query representation, the neural network program including one or more neural network operators for performing the query in a neural network runtime, generating a neural network data structure based on a dataset associated with the query, and executing the neural network program in the neural network runtime over the neural network data structure to generate a query result.

Method and system for query processing over tensor runtimes

The description relates to executing an inference query relative to a database management system, such as a relational database management system. In one example a trained machine learning model can be stored within the database management system. An inference query can be received that applies the trained machine learning model on data local to the database management system. Analysis can be performed on the inference query and the trained machine learning model to generate a unified intermediate representation of the inference query and the trained model. Cross optimization can be performed on the unified intermediate representation. Based upon the cross-optimization, a first portion of the unified intermediate representation to be executed by a database engine of the database management system can be determined, and, a second portion of the unified intermediate representation to be executed by a machine learning runtime can be determined.

Query processing with machine learning

An automated tuning service is used to automatically tune, or modify, the operational parameters of a large-scale cloud infrastructure. The tuning service performs automated and fully data/model-driven configuration based from learning various real-time performance of the cloud infrastructure. Such performance is identified through monitoring various telemetric data of the cloud infrastructure. The tuning service leverages a mix of domain knowledge and principled data-science to capture the essence of our cluster dynamic behavior in a collection of descriptive machine learning (ML) models. The ML models power automated optimization procedures for parameter tuning, and inform administrators in most tactical and strategical engineering/capacity decisions (such as hardware and datacenter design, software investments, etc.). Rich 'observational' models (models collected without modifying the system) are combined with judicious use of “fighting” (testing in production), allowing the tuning service to automatically configure operational parameters of a large cloud infrastructure for a broad range of applications.

Tuning large data infrastructures

Methods, systems, and computer program products are provided for generating and utilizing materialized graph views. A system according to one embodiment includes a graph database including a graph and schema, a workload analyzer, a view enumerator, a query rewriter and an execution engine. The workload analyzer is configured to receive and analyze queries in a query workload. The view enumerator is configured to use an inference engine to operate on facts derived from the graph and a query, and view templates comprising inference rules to enumerate candidate views. The workload analyzer is further configured to selects a candidate view to materialize, provide the selected view to the execution engine that is configured to generate the materialized view. The workload analyzer may select the at least one candidate view based on factors such as query evaluation cost estimates, candidate view performance improvement estimates, view size estimates and view creation cost estimates.

Materialized graph views for efficient graph analysis

Embodiments for efficient queue management for cluster scheduling and managing task queues for tasks which are to be executed in a distributed computing environment. Both centralized and distributed scheduling is provided. Task queues may be bound by length-based bounding or delay-based bounding. Tasks may be prioritized and task queues may be dynamically reordered based on task priorities. Job completion times and cluster resource utilization may both be improved.

Efficient queue management for cluster scheduling

Described herein is a system and method for ranking and/or taking an action regarding execution of jobs of a shared computing cluster based upon predicted user impact. Information regarding previous executions of a plurality of jobs is obtained, for example, from job execution log(s). Data dependencies of the plurality of jobs are determined. Job impact of each of the plurality of jobs as a function of the determined data dependencies is calculated. User impact of each of the plurality of jobs as a function of the determined data dependencies, the calculated job impact, and time is calculated. The plurality of jobs are ranked in accordance with the calculated user impact. An action is taken in accordance with the ranking of the plurality of jobs. The action can include automatic scheduling of the jobs and/or providing information regarding the rankings to a user.

Provenance driven job relevance assessment

Described herein is a system and method for selecting subexpressions to be materialized. For a predefined storage budget, subexpressions of a set of candidate subexpressions to be materialized to minimize query evaluation cost are selected based upon a calculated utility of the set of candidate subexpressions, interactions of the candidate subexpressions, and, a cost of evaluating the candidate subexpressions. Based upon the subexpressions selected to be materialized, subexpression(s) of the set of candidate subexpressions to use when evaluating particular queries of the set of queries to minimize query evaluation cost are determined.

Selection of subexpressions to materialize for datacenter scale

Growing community of inventors

San Francisco, CA, United States of America

Konstantinos Karanasos