Multi-Core Simulator using GPGPU Platforms | Consortium for Embedded Systems | SIU

Abstract: The proposed project is a continuation of previous work for a multicore simulator implemented using the OpenMPI interface. Previous development in the project was focused on implementation of a simulator which can process data, and estimate timing information for a set multi-core architectures. This allows for an exploration of design choices of the underlying hardware to optimize the performance of the algorithms running on the platform.

It has been found that on typical general computing platforms such as the x86, performance bottlenecks exist. This is primarily caused by the limited number of cores available on the host’s platform compared to the many simulated cores. The proposed project seeks to migrate the simulator to a General-Propose computing on Graphics Processing Units (GPGPU) platform, and provide metrics for the order of speed up that is achieved. These platforms have significantly more cores available and may allow for near real-time simulation of multi-core architectures.

Problem:

• Current simulator experiences bottlenecks on high resolution images due to the lack of cores available on traditional desktop platforms
• Bottlenecks erode the ability to process simulation data in real-time, which is a key feature of the simulator

Rationale /Approach:

• The simulator will be converted from a CPU platform to a GPGPU platform
• GPGPUs offer many more processing cores than traditional CPUs allowing more scalable implementations

Novelty:

• Through estimations the current simulator platform is able to rapidly simulate timing of a multi-core system
• Allows developers to explore the design space prior to committing to a hardware platform

Potential Member Company Benefits:

• Allows members to choose a platform which maximizes performance and minimizes cost
• Increase in simulator performance allows for real-time data simulation giving better indication on the simulated platforms performance

Deliverables for the proposed year:

• Survey of GPGPU Platforms and Trade-Offs
• Multi-Core Simulator Running on a GPGPU Platform
• Benchmarks on the a Multi-Core Simulator using Common Image Processing Algorithms
• Documentation of software tool use

Milestones for the proposed year:

• Select a GPGPU Platform – 9/30/2014
• Migrate Multi-Core Tool To GPGPU – 02/28/2015
• Performance Analysis Using Image Processing Algorithms – 04/30/2015
• Transfer Multi-Core Tool and Documentation – 07/31/2015

Progress to Date:

• Implemented high level multicore simulator
• Added capability to stream real-time data to the simulator