Радіоелектронні і комп'ютерні системи (Dec 2019)

ANALYSIS OF TOOLS AND TECHNOLOGIES OF FaaS DEVELOPMENT

  • Инна Николаевна Заризенко,
  • Артём Евгеньевич Перепелицын

DOI
https://doi.org/10.32620/reks.2019.4.10
Journal volume & issue
Vol. 0, no. 4
pp. 88 – 93

Abstract

Read online

This article has analyzed the most effective integrated development environments from leading programmable logical device (PLD) manufacturers. Heterogeneous calculations and the applicability of a general approach to the description of hardware accelerator designs are considered. An analytical review of the use of the OpenCL language in the construction of high-performance FPGA-based solutions is performed. The features of OpenCL language usage for heterogeneous computing for FPGA-based accelerators are discussed. The experience of a unified description of projects for solutions based on CPU, GPU, signal processors and FPGA is analyzed. The advantages of using such a description for tasks that perform parallel processing are shown. Differences in productivity and labor costs for developing FPHA systems with parallel data processing for hardware description languages and OpenCL language are shown. The results of comparing commercially available solutions for building services with FPGA accelerators are presented. The advantages of the Xilinx platform and tools for building an FPGA service are discussed. The stages of creating solutions based on FaaS are proposed. Some FaaS related tasks are listed and development trends are discussed. The SDAccel platform of the Xilinx SDx family is considered, as well as the possible role of these tools in creating the FPGA computing platform as a service. An example of using SDAccel to develop parallel processing based on FPGA is given. The advantages and disadvantages of the use of hardware description languages with such design automation tools are discussed. The results of comparing the performance of the simulation speed of the system described with the use of programming languages and hardware description languages are presented. The advantages of modeling complex systems are discussed, especially for testing solutions involving the processing of tens of gigabytes of data and the impossibility of creating truncated test sets. Based on practical experience, the characteristics of development environments, including undocumented ones, are formulated.

Keywords