TU Berlin

Fachgebiet Datenbanksysteme und InformationsmanagementInferring Program Runtime Estimates from Database Statistics

Logo FG DIMA-new  65px


zur Navigation


Bearbeiter: Andreas Wolf

Betreuer: Max Heimel

Angestrebter Abschluss: Master


Applications usually process and optimize data flow and control flow separately. However, the workload for databases has changed over the last few decades. Data-driven analytics require to run complex analytic algorithms on a tremendous amounts of input data. To efficiently execute these algorithms, hybrid query languages have emerged that allow a joint optimization of the control and data flow. While also being more complex, these optimizations can utilize the strengths of both paradigms.

This talk is about a profile generator that infers program runtime estimates from database statistics to assist classical code optimizations. The profile generator was implemented in the course of the Data Programmability project with the SAP HANA Platform group. To generate profile data, this profiler identifies relational control structures, augments them with database statistics, calculates and propagates selectivity estimates and accumulates the resulting execution counts to path profiles for each individual path in the control flow graph. By combining data and control flow information, the profile generator is able to accurately identify most frequently executed paths in a program. Consequently, this information can be used to efficiently guide compile-time optimization.



Schnellnavigation zur Seite über Nummerneingabe