Transaction consistency in replicated SIP call control systems

The session initiation protocol (SIP) has been selected as the main call control protocol in the 3GPP IP multimedia subsystem (IMS). SIP service users and providers require fault-tolerance with high service availability and reliability. One approach is to replicate SIP call control functionality over a number of dispersed hosts. In order to allow for midcall fail-over, call states need to be replicated, but this may cause call state inconsistency. SIP transaction inconsistency is quantified here by the misread probability. Server response time is defined by the local delay when a read access to call state is to be processed. The tradeoff relationship between misread probability and read delay is exploited to design an update commit algorithm (UCA) for adaptive consistency control. The UCA iteratively updates the read delay so that the misread rate converges towards the predefined misread probability. The algorithm has a rather low complexity and can adapt to a variety of traffic types and load, inter-server link parameters, state-sharing protocols and failure/repair random processes.