DTHMM ExaLB: discrete-time hidden Markov model for load balancing in distributed exascale computing environment

Defining Parameters for the Oscillation Model of Load Flow of Global Activities in a Fully Distributed Exascale System

“…The HPC process is the smallest parallelizable part of whole application running on HPC system. The purpose of the load balancer is to create an optimal match between process requirements and resource attributes so that no process should be left in the queue and all resources should be loaded to the maximum (Khaneghah, E. M., Mollasalehi, F., Aliev, A. R., Ismayilova, N., & Bakhishoff, U., 2018; Bakhishoff, U., Khaneghah, E. M., Aliev, A. R., & Showkatabadi, A. R., 2020). This rule defined with the following formula:…”

Challenges of Using Different Mathematical Models for Load Balancing Optimization in Multi-Core Computing Systems

“…Multi-core computing systems or Exascale distributed systems are characterizing by the occurring of dynamic and interactive events that changes process requirements and resource attributes in the system and requires reloading the task assignment process. The mentioned problem can be solved using two approaches: to allow the occurrence of the dynamic and interactive event and then reload the balancing process or to predict the occurrence of the dynamic and interactive event and to reload the system in a given period (Bakhishoff, U., et al, 2020). This paper was investigated different approaches based on graph theory and techniques of artificial intelligence for reloading the task matching process periodically during the execution process in multi-core computing systems.…”