Ideas List


1. Parallel Programming

1.1. Erlang Many-core Accelerator Support

Summary: Erlang support for many-core accelerators would be a significant boost for Erlang as a high performance computing language/framework. The project would use the Parallella hardware platform and many-core programming infrastructure as a test and development vehicle.

Expected results: support within Erlang/OTP that allows tasks to be offloaded from the host processor to many-core accelerators like the Epiphany processor on the Parallella platform.

Knowledge prerequisite: C/C++, Erlang.

Mentor: Yaniv Sapir, Adapteva.

1.2. Go Language Many-Core Accelerator Support

Summary: Go support for many-core accelerators would be a significant performance/energy-efficiency boost for applications. The project would use the Parallella hardware platform and many-core programming infrastructure as a test and development vehicle.

Expected results: support that allows Go applications running on the host to offload tsks to many-core accelerators like the Epiphany processor on the Parallella platform.

Knowledge prerequisite: C/C++, OpenCL, Go.

Mentor: Yaniv Sapir, Adapteva.

1.3. NumPy/SciPy Many-Core Accelerator Support

Summary: NumPy/SciPy support for many-core accelerators would provide a significant performance boost for Python high-performance computing applications that make use of those libraries. The project would use the Parallella hardware platform and many-core programming infrastructure as a test and development vehicle.

Expected results: support that allows Python applications to offload NumPy/SciPy functions to many-core accelerators like the Epiphany processor on the Parallella platform.

Knowledge prerequisite: C/C++, OpenCL, Python, NumPy/SciPy.

Mentor: Ed Hartley, Meta Metal.

1.4. Hadoop Many-Core Accelerator Support

Summary: many-core accelerator support in Hadoop would provide a significant performance/energy-efficiency boost for applications built using the framework. The project would use the Parallella hardware platform and many-core programming infrastructure as a test and development vehicle.

Expected results: offload of MapReduce operations from Hadoop applications running in a JVM on the host processor, to many-core accelerators like the Epiphany processor on the Parallella platform.

Knowledge prerequisite: C/C++, Java, OpenJDK development, Hadoop.

Mentor: Yaniv Sapir, Adapteva.

1.5. MPI Lite (MPI for Many-Core Accelerators)

Summary: MPI Lite will allow the MPI programming model to be used with massively parallel systems made up of processors with a small amount of local memory and where traditional MPI would prove too heavy. The project would use the Parallella hardware platform and many-core programming infrastructure as a test and development vehicle.

Expected results: a lightweight messaging passing framework that is built on top of OpenCL, and that is suitable for use with many-core accelerators like the Epiphany processor on the Parallella platform.

Knowledge prerequisite: C/C++, OpenCL and MPI or a similar message passing framework

Mentor: David Richie, Brown Deer Technology.

2. Software-defined Radio

2.1. GNU Radio Scheduler for Many-Core Accelerators

Summary: a custom scheduler for GNU Radio will allow it to efficiently schedule execution of signal processing blocks on many-core accelerators. The project would use the Parallella hardware platform and many-core programming infrastructure as a test and development vehicle.

Expected results: a GNU Radio scheduler for heterogeneous systems and that enables efficient scheduling of signal processing blocks across a large number of cores and providing features such as processor affinity.

Knowledge prerequisite: C/C++, DSP.

Mentor: Tommy Tracy II, High Performance Low Power Lab, University of Virginia.

2.2. GNU Radio DSP Block Porting

Summary: porting signal processing blocks to the Parallella platform will allow GNU Radio to take advantage of Epiphany many-core accelerator.

Expected results: signal processing blocks that can be targeted to Epiphany cores and assembled into flow graphs using Python scripts running on the host processor.

Knowledge prerequisite: C/C++, DSP.

Mentor: Tommy Tracy II, High Performance Low Power Lab, University of Virginia.

2.3. OpenBTS DSP Many-core Accelerator Support

Summary: OpenBTS support for many-core accelerators would make it possible to create multi-channel software-defined radio GSM base stations, where the host power consumption is considerably less than it would be than when using a GPP alone. The project would use the Parallella hardware platform and many-core programming infrastructure as a test and development vehicle.

Expected results: OpenBTS DSP code ported to Parallella platform and enabling support for at least four simultaneous radio channels (ARFC) when using a 16-core accelerator.

Knowledge prerequisite: C/C++, DSP.

Mentor: TBC.

3. Distributed Computing

3.1. BOINC Many-core Accelerator Support

Summary: BOINC support for many-core accelerators would provide a significant performance/energy-efficiency boost for volunteer computing applications built using the framework. The project would use the Parallella hardware platform and many-core programming infrastructure as a test and development vehicle.

Expected results: support for Parallella platform within BOINC and at least one open source volunteer computing project that uses the framework.

Knowledge prerequisite: C/C++, OpenCL, BOINC.

Mentor: Yaniv Sapir, Adapteva.