05.08.2014 Views

here - Stefan-Marr.de

here - Stefan-Marr.de

here - Stefan-Marr.de

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

8. Evaluation: Performance<br />

Conclusion To conclu<strong>de</strong>, the average measured overhead of 28% for the kernel<br />

benchmarks for AmbientTalkST and LRSTM are un<strong>de</strong>sirable. However,<br />

the 11% overhead measured for the LRSTM is a positive indication that the<br />

OMOP allows efficient implementation. Arguably, with merely 11% overhead,<br />

the OMOP-based implementation of LRSTM is on par with the ad hoc implementation.<br />

Furthermore, these results are based on a first RoarVM+OMOP<br />

implementation that does not inclu<strong>de</strong> any extensive optimizations.<br />

However, these 11% overhead are no prediction for actual application performance.<br />

Thus, only a performance evaluation with a concrete application<br />

can <strong>de</strong>termine whether this slowdown is acceptable.<br />

8.8. Conclusions<br />

This chapter compared the performance of the two proposed implementation<br />

approaches for the OMOP. Furthermore, it analyzed how the OMOP-based<br />

implementations of AmbientTalkST and LRSTM compare to the corresponding<br />

ad hoc implementations.<br />

CogVM can be 11.0x faster than RoarVM (opt). The experiments started<br />

by measuring the relative performance between CogVM and the RoarVM<br />

variants. Measurements indicate that the CogVM is about 11.0x faster (min<br />

7.1x, max 14.4x) than the RoarVM (opt) on the kernel benchmarks. T<strong>here</strong>fore,<br />

it is used for assessing the performance of the AST-OMOP on top of a VM<br />

with JIT compiler, which gives it the best possible performance.<br />

Measurements indicate that OMOP-based implementations can have on<br />

par performance with ad hoc implementations. The comparison of the<br />

ad hoc implementations with the OMOP-based implementations of AmbientTalkST<br />

and LRSTM used the CogVM with AST-OMOP as well as the<br />

RoarVM+OMOP (opt). While kernel benchmarks on top of the AST-OMOP<br />

show an average slowdown of 2.8x, the results for the RoarVM+OMOP (opt)<br />

are significantly better. VM support for the OMOP results in an average overhead<br />

of 28% (max 2.6x) over the ad hoc implementation. However, some<br />

benchmarks show overhead as low as 2%, being on par with the ad hoc<br />

implementations. Furthermore, the current implementation has an in<strong>here</strong>nt<br />

performance overhead of 17% (min 7%, max 31%), which is part of these performance<br />

numbers.<br />

232

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

Saved successfully!

Ooh no, something went wrong!