Processing large amount of tasks with heterogeneous features by distributing them between limited resources is a difficult problem.
In my thesis I work with a system that has to process these tasks with multiple processing component instances through different processing steps. My task is to find load balancing strategies that perform well in this environment, with using limited information regarding the incoming task’s resource requirement from each processing step, where the last one is an IBM Platform Symphony server grid, containing thousands of servers.
I present different load balancing strategies I researched to help me find a solution for my problem, and I also introduce the IBM Platform Symphony server grid.
To help testing load balancing strategies, I introduce a model that I created of the system, by only keeping its main features. This model can simulate the processing of large amounts of tasks modelled from real tasks.
I also introduce the load balancing strategies I tried that could possibly improve the system, and I test them with my model, and evaluate and compare their performance.