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<br />

E VA L U AT I O N : P E R F O R M A N C E<br />

The goal of this chapter is to evaluate the performance of the AST-OMOP<br />

and the RoarVM+OMOP, by comparing the OMOP-based and ad hoc implementations<br />

of LRSTM and AmbientTalkST. This assessment is based on the<br />

assumption that application <strong>de</strong>velopers already use concurrent programming<br />

based on systems similar to LRSTM or AmbientTalkST. T<strong>here</strong>fore, this evaluation<br />

needs to show that applications can reach on par performance when<br />

LRSTM and AmbientTalkST are implemented on top of the OMOP compared<br />

to being implemented with ad hoc strategies.<br />

This chapter first <strong>de</strong>tails the evaluation methodology and the generalizability<br />

of the results. These results allow conclusions for other byteco<strong>de</strong> interpreters.<br />

However, they preclu<strong>de</strong> conclusions for high-performance VMs.<br />

Subsequently, the performance comparison of the ad hoc and OMOP-based<br />

implementations of AmbientTalkST and LRSTM shows that the OMOP can be<br />

used to implement concurrent programming concepts efficiently. Currently,<br />

the OMOP provi<strong>de</strong>s a sweet spot for implementations that rely on the customization<br />

of state access policies as used in LRSTM. The evaluation further<br />

shows that the overhead of the AST-OMOP is high, while the overhead of the<br />

RoarVM+OMOP is significantly lower. The customization constant optimization<br />

introduces in<strong>here</strong>nt overhead, but offers performance benefits when only<br />

a small number of intercession handlers are customized. Overall, the comparison<br />

of AST-OMOP and RoarVM+OMOP indicates the need of VM support<br />

for the OMOP to obtain performance on par with ad hoc implementations.<br />

201

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

Saved successfully!

Ooh no, something went wrong!