About

Power consumption plays an increasingly important role in multicore and manycore processors. Unfortunately, the power cost of cache-coherent shared memory increases disproportionately with the number of cores, leading the field towards processors without cache-coherent memories. Such architectures require additional effort to manually program and control the memory hierarchy, but at the same time demonstrate better scalability and power-efficiency.

We aim to develop the techniques and tools required towards programming languages, runtime systems and language virtual machines that is specifically tailored for low power consumption in such architectures, without sacrificing the ease of programming and fast development times of such a high-level language. We will implement our results into a functional Java Runtime Environment and power-efficient data-structure libraries tailored for a partially coherent manycore architecture. Specifically, we will provide a collection of highly-efficient synchronization primitives and concurrent data structures taking into consideration locality issues, as a power efficient version of the Java concurrency utility package for such architectures. We will furthermore extend the architecture with power-aware metrics and counters, and use these extensions to evaluate the energy efficiency of our system.

GreenVM develops technology that:

• Lowers power consumption on scalable manycore architectures.
• Maintains ease of programming by realizing a functional Java Runtime Environment for scalable manycores.
• Optimizes data-structure libraries for power efficiency on a partially coherent manycore architecture.