Master 2015 2016
Stages de la spécialité SAR
Research Assistant - Database Programming


Lieu : Brown University, CS Labs
Encadrant : Ugur Cetintemel
Dates :du 01/04/2016 au 31/08/2016
Rémunération :Oui
Mots-clés : Master SAR, autre qu’ATIAM

Cliquer ici pour vous authentifier


Description

Abstract :

Help develop comprehensive performance benchmarks and optimized execution strategies for data-intensive workloads on Oracle’s new M7 Platform. Assist in the analysis of the experimental results and the support and preparation of posters, reports or other scholarly communication.

Description of responsibilities :

Oracle’s new chip set M7 (also called "software in silicon") implements several high-level data processing functionalities in hardware. These operations include filter/scan, set membership, and decryption, which are all common operations that consume significant CPU cycles. Pushing these operations to hardware, where they can be executed in full-memory speeds is a game changer : for example, this capability not only allows processing on large, compressed data sets to go very fast, it also allows the cores in the machine to use its cycles on other high-level tasks.

M7 is a breakthrough co-processor technology that can be thought of as a co-processor (called DAX, or Data Acceleration System) to which a large class of common data- and time-intensive operations can be offloaded.

This technology was released just last summer. The Database Systems group at Brown has early access to M7s through their collaboration with researchers and developers at Oracle.

The intern will be working with Brown CS database group and the Oracle group to carry out the following tasks.

1. Help develop comprehensive performance benchmarks and optimized execution strategies for data-intensive workloads on Oracle’s M7 Platform.

This task will involve taking well-established database benchmarks (e.g., TPC-H and TPC-C) and implementing them on M7s. A comprehensive performance evaluation of M7 will be done, along with comparative evaluation against alternative architectures (e.g., Intel Xeon Phi). The goal is to numerically quantify the speedups enabled by the DAX co-processor for a variety of workloads and configurations, to identify potential bottlenecks and also workloads where M7s "shine".

A second task is to consider common database operations (such as filters, joins, and aggregations) and devise optimized execution strategies for these using DAX’s low-level programming interfaces. High-level programming abstractions that implement optimized parallel execution of these operations on multiple DAXs, as well as across multiple devices will be designed and implemented. This also involves the choice of where to execute different types of operations in the presence of multiple cores and DAXs.

2. Assist in the analysis of the experimental results and the support and preparation of posters, reports or other scholarly communication.

The experimental results will need to systematically cataloged, analyzed and presented within technical reports, whitepapers, demos or conference publications. These results will also help the team identify performance bottlenecks, which will be then be addressed through new execution strategies.