Communication Affinity Aware Process Placement: Incorporating Physical Network Considerations

Abstract

Abstract Microservice architecture, widely adopted in the modern software industry, offers numerous benefits through the decentralized distribution of components across a network. However, the inter-service communication costs in this architecture impose significant overhead on overall application performance. Drawing inspiration from microservices, this challenge extends beyond specific architectures, impacting any distributed processes communicating over a network. While complete elimination of communication costs remains impractical, efforts have been taken to mitigate their impact. Beyond the inter-process affinities, effective solutions must also consider resource utilization and physical network properties, adding complexity to the problem. This dissertation presents a novel approach to address the process placement problem in networked environments by incorporating communication affinity among the processes and physical network properties. The objective is to minimize communication costs by strategically placing processes within the network. The study explores the use of subgraph isomorphism-based virtual network embedding, leveraging the VF2 algorithm to map an application-specific communication affinity graph onto the underlying physical network topology. The proposed method considers both application-level constraints, such as communication affinities and CPU requirements, and physical network constraints, including node CPU capacities and link bandwidths. Through experimentation and analysis, the study evaluates the effectiveness and limitations of the proposed approach, highlighting its potential for improving application performance in a distributed environment. This study shows how the communication cost between a set of distributed process can be minimized by considering the underlying physical network properties and lays a solid foundation for tackling the process placement problem, offering valuable insights and methodologies for optimizing communication affinity-based placements in network environments. Additionally, the thesis discusses avenues for future research to enhance process placement strategies in real-world network environments. Overall, this research contributes to the advancement of process placement techniques and lays the groundwork for further exploration in this domain.