Student research opportunities
Faster Dynamic Execution through Smarter Adaptive Compilation
Project Code: CECS_68
This project is available at the following levels:
Honours, Summer Scholar, Masters, PhD
Supervisor:
Professor Steve BlackburnOutline:
High performance runtime systems use adaptive compilation to perform feedback-directed optimization (FDO). One of the original, and more advanced such systems is the Adaptive Optimization System (AOS) used by Jikes RVM (formerly known as Jalapeno), and developed by Matthew Arnold et al. Such systems observe execution of a program, and based on those observations, a) select particular code for optimization, and b) target specific optimizations and levels of optimization at that code according to its dynamically observed characteristics. Jikes RVM's AOS is very aggressive, allowing speculative optimization and on-stack replacement (OSR), which means even a very long running hot loop can be optimized and replaced on the fly without breaking out of that hot loop.
Goals of this project
This project will explore and implement a number of strategies for improving the effectiveness of the AOS, with guidance from Matt Arnold and other colleagues of ours at IBM Research. These strategies include different strategies for sampling (which is used to drive the AOS's decisions), including sampling driven by hardware events rather than a timer, and sample rates changing over time to maximize the effectiveness of the tradeoff between the cost of sampling and the quality of the information gleaned by sampling. This project has the potential to improve the performance of a high performance research JVM used by researchers all over the world.
