Rapid increases in computer performance, according to Moores law, have enabled software architects to design software of ever increasing functionality and complexity. Consequently, consumers expect software that grows in functionality and can process larger data sets with every new release (e.g., huge spreadsheets or streaming high-quality video). Unfortunately, since Moores law will likely not hold in the near future, owing to physical limitations such as wire delays and heat dissipation, hardware speedups will not keep pace with the demands of software. Thus, to meet the performance requirements of new software, one must look to software techniques.
The proposed work will explore a new approach, algorithmic optimizations in virtual machines, to provide the performance for next generations of software.
Intellectual Merit
-A system for marking implementations of algorithms, data structures, and their uses. A VM that knows about this system can exploit the information for connecting uses with suitable implementations. A VM that does not know about this system still executes the code correctly but without its performance benefits. -An implementation within a Java virtual machine that exploits the information for improving performance. This implementation support includes modifications to the garbage collector (for migrating data from one representation to another), compiler (for inlining away the calls), and scheduler (for scheduling profile collection threads with lower priority). -A collection of libraries and applications marked to exploit algorithmic optimizations. This will form part of the evaluation for the proposed work.
Broader Impact
The infrastructure developed as part of this research will be used in undergraduate algorithms classes to give students deep knowledge about the strengths and limitations of commonly used algorithms and data structures. This research will provides the performance necessary for meeting the needs of future software systems.
