Analyzing routing protocol convergence in routed satellite networks

Routed multi-satellite networks provide several challenges that are not present in entirely ground-based wired networks. These include asymmetrical wireless links, high latencies, and higher bit error rates. For example, ground terminals generally have a unicast wireless link to connected satellites, while the satellites can generally broadcast to a larger subset of ground terminals simultaneously. These characteristics can have a significant impact on routing protocol selection and performance. This paper discusses OPNET simulation results that attempt to characterize the differences between BGP, which operates atop a reliable point to point transport protocol (TCP), with OSPFv2, operating atop an unreliable broadcast protocol (IP). We have simulated multiple router-enabled satellites at geosynchronous orbit with high bandwidth crossconnects, and a custom link model that connected multiple ground nodes to a single satellite network interface. This models the effect of a single downlink beam being composed of one uplink beam per terminal. We varied the number of ground terminals from 1 to 250 in a downlink beam. Each of these ground terminals had their own route tables representing ground-based networks behind the terminals, and redistributed these routes to all satellites and other ground terminals using each routing protocol selection. These tables were uniform in size across all ground terminals, and varied from 25 to 100 entries per ground node. We report router convergence times and overheads for cold startup and for a later ground terminal addition, and analyze and evaluate the relative performance of OSPF and BGP under these scenarios. We illustrate that routing protocol performance is affected more dramatically by individual protocol configuration than by differences between the two, and show that the scalability of each is affected primarily by the product of number of peers and routes.