Computation overhead due to networking in vitualized environments

Virtualization has become an essential component of modern applications, providing flexibility and scalability at a manageable cost. This thesis examines network performance in virtualized Linux environments, with particular focus on the impact of CPU allocation on achievable throughput and the overhead introduced by system and virtualization layers. By analyzing different communication scenarios between hosts and virtual machines, the study explores how resource constraints affect network performance. The collected data are further examined through a targeted analysis of CPU behavior and workload distribution across both host and virtualized contexts. The main objective is to quantify overhead in terms of CPU utilization and network performance, identifying conditions under which performance parameters remain within predictable ranges and revealing any deviations that might impact throughput. This analysis helps identify potential bottlenecks that arise under different configurations and serves as a foundation for developing advanced scheduling strategies in containerized and orchestrated environments. For applications that require guaranteed data transfer levels as part of specific Quality of Service (QoS) requirements, the findings provide guidance for scheduling policies that take into account CPU load, nodespecific characteristics, and workload demands. In environments such as Kubernetes, this work can support the development of intelligent scheduling policies that optimize resource placement and allocation through informed decisions on CPU affinity, node selection, and replication strategies. In doing so, this thesis offers a systematic methodology and valuable insights for improving network performance in virtualized systems, contributing to effective resource management in modern, performance-driven network infrastructures.