A case study of the trade-offs in a parallel protocol processing architecture under transient link failures

Signaling Transfer Points (STPs) are large packet switches in the Signaling System Number 7 (SS7) network. They route signaling messages to set-up and tear-down calls and perform network management functions in the event of link and node failures and overload conditions. These functions are part of the Message Transfer Part (MTP) Layer 3 protocol functions in the SS7 protocol stack and require the support of routing tables that maintain the status of the routes in the network. One key factor in the design of the STP architecture is the parallelism used in performing MTP Layer 3 protocol functions. This paper attempts to quantify the effect of parallelism on call throughput when the network experiences multiple simultaneous link failures. Results show that centralized architectures tend to over-control processor congestion subsequent to link failures, resulting in a higher number of calls being throttled at the source. Architectures with multiple processors can perform better since the network management task is shared by multiple processors and any subsequent processor congestion impacts a fewer number of sources. However, when the parallelism is very high, the routing table synchronization cost becomes high resulting in delays in synchronizing the routing tables which causes large number of messages to be dropped in the STP. These and other architectural trade-offs are addressed in this paper based on a detailed simulation analysis.