Model for Automated Parallelization in Microservices Using Blackboard Systems and Linda Tuple Spaces

Abstract

This thesis proposes the development of a parallelization framework that optimizes task execution in distributed systems based on microservices. The prevailing methods of parallel computation characterized by static scheduling, central coordination, and human intervention are not addressing the dynamically changing needs of modern applications as cloud and microservices-based structures spread among industries. For facilitating decentralized task management, dynamic workload management, and self-synchronization purposes, this work proposes a completely novel framework integrating Linda tuplespaces and blackboard systems augmented by intelligent tuples. The basic concept of the proposed framework is the utilisation of a common knowledge area — delivered by blackboard-based systems — as a means of allowing multiple processes to collaborate towards the solution of complex problems. The model of decoupled communication of Linda tuplespaces complements this by allowing distributed services to interact reliably and asynchronously. The two paradigms put together cause the performance of concurrent systems typically to suffer as a result of delays due to synchronisation as well as communication overhead. The framework minimizes manual setup and facilitates on-time adaptation of the changing workloads through the automation of task delegation as well as execution. The thesis also analyzes significant theoretical foundations of parallel processing like the implications of Amdahl’s Law and the perspectives on scalability provided by Gustafson’s Law. These theories bring into relief the shortcomings of traditional sequential bottlenecks as well as the likely advantages of effective parallelization. Several application domains like IoT-based emergency networks and high-frequency trading networks are applied especially for the evaluation of the proposed model. Although efficient distribution of loads plays a crucial role in timely computation of the data and dependability of the system within the context of emergency networks, even minimal delays due to task synchronization cause lost opportunities and considerable economic losses in the context of high-frequency trading. The combination of smart tuples, blackboard systems, and enhanced Linda tuplespaces greatly increases system throughput, fault tolerance, and resource utilization, as shown by experimental evaluations and performance analyses. According to the results, many of the difficulties present in distributed computing environments can be resolved with a decentralized and flexible task management approach, offering a scalable, reliable, and effective solution.