As Moore’s Law comes to an end, performance advancements will increasingly rely on parallelism and special purpose processors. Software infrastructure is playing catch-up as it strives to enable development and test to effectively utilize parallelism, and unlock the performance benefits of heterogeneous processors.
We are trying to address a few pressing needs
Provide a lightweight runtime system that support computation and data movement in massive concurrency that directly leverage low level hardware resource for enterprise storage (IO intensive) and HPC workload (Compute intensive). It dynamically matches requirements from application and hardware resource, scheduling based on power, resilience, memory locality and processor capabilities.
Enable development by enhancing ARM system emulation tools, such as QEMU, by investigating the performance implications of scaling to hundreds of cores and working to upstream these developments into the QEMU open source community.
Perform research on critical aspects of heterogeneous compute, such as a heterogeneous memory management subsystem, scheduling and software packaging, distribution and deployment.