Microsoft Technology Licensing, LLC

Advanced Micro Devices Corporation

Wisconsin Alumni Research Foundation

Rathijit Sen

Carlo Aldo Curino

Alekh Jindal

Hiren Shantilal Patel

Shi Qiao

Alan D Halverson

Matteo Interlandi

Marius Dumitru

Subramaniam Venkatraman Krishnan

Konstantinos Karanasos

Malay Kumar Bag

Hans C Lehnert Merino

Wei T Huang

Indrani Paul

Clemens Alden Szyperski

David Allen Wood

Yuan Yu

Marc Todd Friedman

Gagan Gupta

Raghunath Ramakrishnan

Kwanghyun Park

Zhicheng Yin

Karthik Saligrama Ramachandra

Markus Weimer

Abhishek Roy

Fotios Psallidas

Dalitso Hansini Banda

Jesus Camacho Rodriguez

Anish Yatin Pimpley

Zhiwei Fan

Harshada Chavan

Krystian Sakowski

Zezhou Huang

Karla Jean Saur

Doris Suiyi Xin

Ivan Popivanov

Shuo Li

Yi Zhu

Vishal Lalchand Rohra

Anubha Srivastava

Supun Chathurang Nakandala

Dong He

Hossein Golestani

Wei Huang

Alan Dale Halverson

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

A method, computer program product, and computing system for optimizing query operations on run length encoding (RLE) data in a parallel processing computing system. Data is received in a plurality of columns of an input table of a parallel processing computing system for query execution; the system determines that at least a portion of the received data in a first number of columns is compressed according to run length encoding (RLE), thereby comprising RLE data columns including RLE data and that the received data in a second number of columns is not compressed according to run length encoding (RLE), thereby comprising non-RLE data columns including non-RLE data. A query operation is executed on the RLE data and the non-RLE data by prioritizing processing of the RLE data columns over processing of the non-RLE data columns.

System and method for accelerating query execution

A method, computer program product, and computing system for processing query operations on run length encoding (RLE), data in a parallel processing computing system. Data for query execution is received at a parallel processing computing system, at least a portion of the data being compressed according to RLE, thereby forming RLE data; and a query operation is executed on the RLE data without performing a decompression operation on the RLE data.

System and method for performing query operations on run length encoded data

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

Solutions for optimizing job runtimes via prediction-based token allocation includes receiving training data comprising historical run data, the historical run data comprising job characteristics, runtime results, and a token count for each of a plurality of prior jobs, and the job characteristics comprising an intermediate representation and job graph data; based at least on the training data, training a token estimator, the token estimator comprising a machine learning (ML) model; receiving job characteristics for a user-submitted job; based at least on the received job characteristics, generating, with the token estimator, token prediction data for the user-submitted job; selecting a token count for the user-submitted job, based at least on the token prediction data; identifying the selected token count to an execution environment; and executing, with the execution environment, the user-submitted job in accordance with the selected token count.

Optimizing job runtimes via prediction-based token allocation

The cloud-based query workload optimization system disclosed herein the cloud-based query workloads optimization system receives query logs from various query engines to a cloud data service, extracts various query entities from the query logs, parses query entities to generate a set of common workload features, generates intermediate representations of the query workloads, wherein the intermediate representations are agnostic to the language of the plurality of the queries, identifies a plurality of workload patterns based on the intermediate representations of the query workloads, categorizes the workloads in one or more workload type categories based on the workload patterns and the workload features, and selects an optimization scheme based on the category of workload pattern.

Cloud based query workload optimization

A serverless query processing system receives a query and determines whether the query is a recurring query or a non-recurring query. The system may predict, in response to determining that the query is the recurring query, a peak resource requirement during an execution of the query. The system may compute, in response to determining that the query is the non-recurring query, a tight resource requirement corresponding to an amount of resources that satisfy a performance requirement over the execution of the query, where the tight resource requirement is less than the peak resource requirement. The system allocates resources to the query based on an applicable one of the peak resource requirement or the tight resource requirement. The system then starts the execution of the query using the resources.

Resource optimization for serverless query processing

Embodiments described herein are directed to a microarchitecture modeling tool configured to model and analyze a microarchitecture using a dependency graph. The dependency graph may be generated based on an execution trace of a program and a microarchitecture definition that specifies various features and/or characteristics of the microarchitecture on which the execution trace is based. The dependency graph includes vertices representing different microarchitectural events. The vertices are coupled via edges representing a particular dependency therebetween. The edges are associated with a cost for performing microarchitectural event(s) corresponding to the vertices coupled thereto. The dependency graph also takes into account various policies for structural hazards of the microarchitecture. The microarchitecture modeling tool analyzes the costs associated with each of the edges to determine a design metric of the microarchitecture. A user is enabled to modify various features of the dependency graph to analyze different design choices and/or optimizations to the microarchitecture.

Performance modeling and analysis of microprocessors using dependency graphs

Methods for automatic selection of degrees of parallelism for efficient execution of queries in a database system are performed by systems and devices. An incoming query associated with a query system is received and features of the incoming query are determined. A system state of the query system and a set of executing queries are also determined, along with a query state of each executing query in the set. At runtime of the incoming query, allocation of a degree of parallelism for executing the query is determined by calculating different possible execution times for the incoming query at least partially concurrently with the set of executing queries. Execution times are calculated for different parallel thread options and based on query features, system state, or query states of executing queries. The execution of the incoming query is initialized with the parallel thread option corresponding to a specific execution completion time.

Methods for automatic selection of degrees of parallelism for efficient execution of queries in a database system

According to examples, an apparatus may include a machine readable medium on which is stored machine readable instructions that may cause a processor to, for each of a plurality of resource setting levels, determine resource usage characteristics and execution times of executed workloads, assign, based on the resource usage characteristics of the executed workloads, each of the executed workloads into one of a plurality of resource bins, determine, for each of the resource bins, an average execution time of the executed workloads in the resource bin, determine a total average execution time of the executed workloads from the determined average execution times, identify a lowest total average execution time of the determined total average execution times, determine the resource setting level corresponding to the identified lowest total average execution time, and tune a resource setting to the determined resource setting level.

Tune resource setting levels for query execution

A frequency governing method for a processor includes reading power management information from the processor. The processor operates in an active state or one of multiple power saving states. The power management information includes first information indicating occupancy of a first core of the processor in the active state and second information indicating occupancy of the first core in a first power saving state. The method includes generating an effective utilization based on the first information and the second information. The method includes setting a target frequency by performing a selected action of a first action including increasing the target frequency and a second action including decreasing the target frequency with respect to a current frequency of the processor based on the effective utilization. The selected action is selected according to the effective utilization. The method includes instructing the processor to operate the first core at the target frequency.

Core frequency management using effective utilization for power-efficient performance

Disclosed is a method of determining concurrency factors for an application running on a parallel processor. Also disclosed is a system for implementing the method. In an embodiment, the method includes running at least a portion of the kernel as sequences of mini-kernels, each mini-kernel including a number of concurrently executing workgroups. The number of concurrently executing workgroups is defined as a concurrency factor of the mini-kernel. A performance measure is determined for each sequence of mini-kernels. From the sequences, a particular sequence is chosen that achieves a desired performance of the kernel, based on the performance measures. The kernel is executed with the particular sequence.

System and method for determining concurrency factors for dispatch size of parallel processor kernels

Computational resources of an electronic computer are monitored by predictors that establish predicted trade-offs between performance and power for a particular workload while the workload is being executed. A coordinator combines prediction values to identify a limited number of combinations of operating states of the computational resources, allowing operating states of the computational resources to be readily adapted during program execution based on a particular workload. The limited number of combinations of operating states are ideally Pareto optimal combinations.

Electronic computer providing power/performance management

Growing community of inventors

Madison, WI, United States of America

Rathijit Sen