DISTRIBUTION SYSTEMS AND PARALLEL COMPUTING TECHNOLOGIES

Abstract

High performance computing (HPC) has recently become important in several sectors, including science and manufacturing.

To build powerful systems requires a combination of software models to increase system parallelism, especially dual and three-tier programming models to increase parallelism in disparate systems, including CPUs and GPUs.

However, building systems with different programming models is error-prone and complex, and difficult to verify. In addition, testing parallel programs is already a difficult task, as parallel errors are difficult to detect due to the indeterminate behavior of the parallel application. Integrating multiple programming models into a single application complicates testing, as this integration can lead to new types of errors.

Despite efforts to create tools for testing parallel systems, much remains to be done, especially when testing heterogeneous and multilevel programming models.

This paper presents the structure and functionality of zFunction, which is an adaptive distributed computing platform that supports a convenient programming model for developing applications for parallel processing. This allows developers to develop software as if they were programming for a single computer, and then it automatically takes care of data distribution and task parallelization on different nodes of the cluster or multiple CPU cores. The z function thus significantly improves the performance of a complex distributed program that processes large amounts of data in real time, these are critical systems.

This paper uses a representative case study in the field of financial services to show how zFunction can benefit from these types of programs.