05.08.2014 Views

here - Stefan-Marr.de

here - Stefan-Marr.de

here - Stefan-Marr.de

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

8.3. Baseline Assessment<br />

marking framework 17 . The benchmarks are executed based on a configuration<br />

that consists of the lists of benchmarks and their parameters, as well as the<br />

VMs and the necessary command-line arguments. ReBench 18 uses this configuration<br />

to execute the benchmarks accordingly. The configuration is given in<br />

Appendix B.2.<br />

For the CogVM, an official 32-bit binary is used (cf. Sec. 8.1.3). The RoarVM<br />

is equally compiled as a 32-bit binary. The compiler used is either Clang<br />

3.0 or GNU GCC 4.2 with full compiler optimization enabled, i. e., using the<br />

compiler switch -O3.<br />

8.2.2. Presentation<br />

Beanplots Most of the following diagrams use beanplots [Kampstra, 2008].<br />

Beanplots are similar to violin and box plots. They show the distribution of<br />

measurements, and thus, enable an visual comparison of benchmark results,<br />

including an assessment of the significance of the observed differences (cf.<br />

Fig. 8.1). Note that the beanplots are directly based on the measured data<br />

points, i. e., they <strong>de</strong>pict the distribution of the actual measurements and are<br />

not synthesized from aggregated values. Asymmetric beanplots such as in<br />

Fig. 8.3 are a variant that facilitates direct comparison of two data sets by<br />

<strong>de</strong>picting both data sets for each criterion si<strong>de</strong>-by-si<strong>de</strong>. The first example in<br />

the following section discusses the semantics of these diagrams in more <strong>de</strong>tail.<br />

Geometric Mean Note that all averages reported in this dissertation are<br />

based on the geometric mean. For most experiments normalized measurements<br />

are reported, and thus, the geometric mean needs to be used because<br />

the arithmetic mean would be meaningless (cf. Fleming and Wallace [1986]).<br />

8.3. Baseline Assessment<br />

This section discusses baseline performance of the VMs used for the experiments.<br />

The goal of this discussion is to provi<strong>de</strong> an intuition of the performance<br />

characteristics of the individual VMs before discussing the impact of<br />

the modifications for the OMOP. This evaluation uses the kernel benchmarks<br />

<strong>de</strong>scribed in Appendix B.1.2. As discussed in Sec. 8.2, every benchmark is executed<br />

100 times for each of the VMs. All benchmarks have been executed<br />

with i<strong>de</strong>ntical parameters.<br />

17 http://www.squeaksource.com/SMark.html<br />

18 http://github.com/smarr/ReBench<br />

209

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!