Hybrid distributed computing sharing platform

This dissertation presents a solution to the problem of straggling tasks in heterogeneous hybrid distributed computing networks when the task sizes and the number of tasks are unknown. The aim of this work is to create a hybrid distributed computing platform for heterogeneous computing networks. The hybrid distributed computing platform architecture proposed in this dissertation uses the created task distribution method for tasks distribution between clusters. Although this method is not new, it has not yet been applied to hybrid distributed computing platforms. The proposed method uses a task stalling buffer, thus avoiding straggling tasks in slow-performing computers. Although this method does not require preliminary data on the task queue or its tasks, it can only work on systems with two clusters. Therefore, an improved task distribution algorithm has been developed. The new algorithm works in systems with more than two clusters and adapts to distributed computing environment performance changes. A comparative analysis of the proposed algorithms was performed with the classical algorithm capable of operating under heterogeneous system conditions when there is no preliminary data on the task queue or its tasks. This dissertation provides an overview of the distributed computing model, existing private and volunteer computing solutions, and cloud computing alternatives. It provides an overview of the existing distributed computing platforms and their applications. In addition, hierarchical and non-hierarchical task distribution algorithms are reviewed, which are suitable to use in a hybrid distributed computing platform and solve the straggling task problem