Total Link Service Rate Estimation With Packet Metadata

Guaranteeing end-to-end network latency bounds is required in many applications. A necessary condition for ensuring end-to-end latency bounds is to keep the total service rate for flows on each link lower than the capacity of the link. An admission control with setup messages before data transmission is necessary to check whether the total service rate remains within the capacity in each link through which the flow passes. Current practices of admission control require every node to maintain flow state information including the service rates of individual flows, which is not practical in large-scale networks with high link speeds and a massive number of flows. Instead, determining the total service rate of a link without maintaining any flow state information is highly desired. This study proposes a method to accurately infer the total service rate of the link using only packet metadata so that admission control can be performed without setup messages or state maintenance for each flow at the core node. By specifying the exact time of inference, the service rate can be inferred at a desired time. The effectiveness of the proposed method is proven through simulations. The simulation was conducted on a simple topology with four types of flows, each differing in service rate and service period.