[dbis logo]

.Lehre.Abschlussarbeiten
[Institut fuer Informatik] [Leerraum] [Humboldt-Universitaet zu Berlin]

Query Execution using modern hardware [SA, DA, BA and/ or MA]

Description:

Over the last decade, several approaches for parallel query execution have emerged. Research in this area showed, that the performance of these approaches is mainly affected by the non-manageable cache hierarchy. However, the capabilities of the cache hierarchy in modern processors are exploited differently. Furthermore, comparing these different approaches is difficult due to different operator-to-resource assignments at run-time (scheduling strategies) and different numbers of tuples each operator processes in a batch (chunk size).

In previous work, we presented QTM, a query task model that generalizes the modeling of parallel query execution such that different approaches become comparable. We simulate query execution of current state-of- the-art database systems like MonetDB, DB2, PostgreSQL, or MonetDB X100 using QTM. We show, how these DBMS would execute the same QEP differently. In particular, we investigate the exploitation of the multi-level cache hierarchy by different access patterns and the opportunities for parallelism. Therefore, we achieve a major step towards a profound understanding of the reasons and circumstances that lead to different performance behavior of the same QEP in different DBMS.

The aim of this work is to investigate different aspects of this model using real world data sets and queries. According to the skills and interests of the student, different fields (e.g. GPU, scheduling, parallelization, etc.) can be investigated.

 

Requirements: 

English language for reading related literature, good C++ skills, basic knowledge of CPUs

 

Contact:

Steffen Zeuch, steffen.zeuch@informatik.hu-berlin.de

Prof. J.C. Freytag, Ph.D., freytag@informatik.hu-berlin.de



[aktiver Punkt]  Themen

[Punkt]  Masterarbeiten

[Punkt]  Bachelorarbeiten

[Punkt]  Diplomarbeiten

[Punkt]  Studienarbeiten



Ansprechpartner

+49 30 2093-3090