Mathematical Model for Approximation the Efficiency of Parallel Computing on Single Board Cluster with Least-squares Approximation

Abstract

This research aims to study the relationship between parallel processing efficiency and several nodes on a single board cluster using a mathematical model, approximating least squares. This research tested on the Raspberry Pi single-board in the form of a high-performance computing system. It divided the tasks that need to be processed in each particular part and sent it to each unit to process simultaneously via the MPI (Messaging Passing Interface). This process is the standard division of work with communication between processors in the form of messages on the cluster system. It consists of eight nodes of Raspberry Pi. It measures the instruction set's ability to perform decimal operations per second or Floating-point Operation Per Second (FLOPS) with High-Performance Linpack Benchmarks (HPL). As a result, the efficiency of the ability to process instruction set in decimal per second increases the performance continuously when increasing the number of the node on the cluster. Which corresponds to the mathematical model obtained f(x) = 1.0684x^(0.8256).It shows a relationship between parallel processing performance values and the number of nodes on the cluster and can be estimated with the mathematical model above.