StepNet: A Compositional Framework with Reduced Querying for Homing Complex Network Services

Homing or placement of network elements on cloud infrastructure is a crucial step in the orchestration of network services, involving complex interactions with several cloud and network service controllers. Network Service Providers (NSPs) currently follow a traditional approach akin to existing VM and VNF placement techniques that involves hand-crafting service specific heuristics for homing network services. However, operational experience from a Tier-1 NSP shows that existing approaches do not scale well when network services evolve and their requirements change. Further, these approaches require extensive and repetitive querying of the various controllers (e.g., to check customer eligibility or capacity), placing significant burden on the resources at the controllers. We propose StepNet, a compositional homing framework that allows service designers to easily mix and match homing requirements to create instances of the homing problem, enabling greater agility of service creation and evolution. StepNet adopts an incremental approach to querying that provides near optimal homing solutions, while reducing the cumulative time spent by all of the data sources responding to queries for each homing request (query cost). Our evaluation with production traces from a Tier-1 NSP shows a reduction in query cost of 92% for over 50% of the requests.