- Главная
- →
- Архитектуры, масштабируемость
Optimizing Function Placement for Large-Scale Data-Center Applications Архитектуры, масштабируемость
Guilherme Ottoni is currently a Software Engineer at Facebook, where he's leading the design and implementation of the HHVM JIT compiler, as well as HHVM's overall performance effort. Before joining Facebook in 2011, Guilherme was a Staff Research Scientist at Intel Labs. Guilherme received his Ph.D. from Princeton University.
Тезисы
I'll present a practical methodology and tool that we developed to improve the performance of large-scale applications at Facebook. This methodology sorts the functions at link time using sample-based profiling data, and it doesn't require any change to the compiler or the linker. The order of the functions is obtained by a simple open-source tool that we built, called HFSort. We also show how this technique can be combined with the selective use of huge-pages support available in modern processors to further boost performance. This methodology has been successfully deployed to various large-scale server applications at Facebook, resulting in performance improvements ranging from 5% to 11%.