Model-Driven Techniques for Virtual Network Function Rehoming for Service Chains

Background

The shift in communication networks from specialized hardware to virtualized network functions (VNFs) operating within cloud infrastructures has introduced significant complexities in resource management. Service providers frequently deploy VNF service chains on private clouds with constrained capacity, necessitating strict adherence to Service Level Objectives, particularly concerning availability for critical services. Consequently, VNFs often require relocation, or "rehoming," in response to dynamic cloud events such as resource hotspots, system failures, or infrastructure upgrades. A primary challenge with current rehoming strategies is their singular, inflexible approach, which often results in operational inefficiencies and delays. This inflexibility makes it difficult to dynamically determine the most appropriate rehoming action and timing, including whether to employ live migration, thereby hindering optimal cloud resource utilization and service continuity.

Technology

Researchers at Stony Brook University developed a system that identifies when a trigger event occurs within a cloud infrastructure, then extracts specific characteristics from one or more VNFs within a service chain. Based on these characteristics, it determines various potential rehoming actions for each VNF and predicts the associated rehoming delay or service chain downtime for each action. Subsequently, the system selects an optimal rehoming action for at least one VNF by evaluating these predicted delays or downtimes, and then executes that optimal action.