Project: NUMA4HANA

Team: Prof. Dr. Andreas Polze (HPI), Felix Eberhardt (HPI), Kai Schliewenz (HPI), Bernhard Höppner (SAP ICP)

Research institution: HPI Potsdam & SAP ICP

Abstract: On NUMA systems, good thread and memory placement is the key factor for optimal performance. Using SAP HANA as system under test, we want to use state-of-the-art approaches to analyze and optimize the behavior of software on NUMA systems. To gather needed system metrics, we will exploit certain hardware events, performance counters and use instruction sampling. Additionally we try to investigate how race conditions can be detected on INTEL x86 CPU’s with respect to parts of the SAP HANA database. There are two ways to detect data races in software. On the one hand static code analysis tools such as RacerX or LockSmith and on the other hand dynamical code analysis tools like Helgrind, Jumble or FeatherLite.

Last modified 8 years ago Last modified on Sep 27, 2013 4:32:44 PM