Keywords: medical application, hardware-based solution, high performance computing, HPC, hybrid parallelization, hybrid small size HPC resource, HuSSaR, Xeon PHI, GPU, hybrid hardware scheduling, hybrid optimization.

The increasing ubiquity and essentiality of information technology has led to the development of never before seen computing capabilities across most architectures. Examples of innovative new hardware include General-purpose computing on graphics processing units (GPGPU), the Intel Many Integrated Core Architecture (MIC), the field-programmable gate array (FPGA), and more recently the Neural Processing Unit (NPU). These processing units (PU) can take advantage of using fusion-based approaches with higher computing efficiency instead of considering competitive individual algorithms. Despite the need for high-performance computing (HPC) in cutting edge medical technology that produces large volumes of data, there is still limited access to vital resources. There is also a lack of dedicated HPC solutions for specific medical tasks. To widen this bottleneck, we present a novel hardware accelerated solution for the efficient adoption of the complex and fusion-based approaches for real-time medical applications. Our primary aim was to build a small mobile hybrid high-performance computer which could be placed in an operation room or placed on a truck for mobile screening to support different detection or treatment-related computational issues. While the number of PUs is limited on mainboards, co-processor cards are the only option for building a high PU density machine in a regular size case to overcome computational barriers of the regular personal computers. When there is no particular algorithm predefined, it is hard to tell in advance which hardware technology of PU should be used. It has been proven that the aggregation of different solutions can solve a complex problem more efficiently compared with any individual approach. To achieve our objective, we have developed a Hybrid Small Size high-performance computing Resource (HuSSaR) which efficiently allows usage of the processing unit with various architectures. It is also mobile and has its own cooling system to support easy mobility and wide applicability. We include practical examples from the clinical data processing domain in this work. We explain how we used our HuSSaR HPC and the optimization of our computations to fit the hybrid HPC environment including the field of the distributed computation. The optimal exploitation of the system is demonstrated on a skin lesion classification, CT segmentation and 3D visualization including virtual reality tasks, and the third-party application so-called SWhybrid. For all the applications, a task scheduler (grid engine/slurm) manages task assignment to the hardware. The main goal is to maximize the utilization of the hardware and reduce the computation time. This way the hardware acceleration is driven by optimal task scheduling based on the hardware requirements for the specific task.

purchase the full-text of this paper (price £22)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £25 +P&P)