SERvartuka : Enhancing SIP Server Scalability with Dynamic State Management

A growing class of applications, including VoIP, IMS and Presence, are enabled by the Session Initiation Protocol (SIP). Requests in SIP typically traverse through multiple proxies. The availability of multiple proxies offers the flexibility to distribute proxy functionality across several nodes. In particular, after experimentally demonstrating that the resource consumption of maintaining state is significant, we define the problem of state distribution across multiple nodes when the goal is to increase overall call throughput. Our approach for solving this problem leads to the design and evaluation of SERvartuka, a more scalable SIP server that dynamically determines the optimal fraction of SIP requests to be handled statefully per server in order to maximize call throughput. This design is in contrast to existing SIP servers that are statically configured to be either stateful or stateless and therefore result in sub-optimal call throughput. We implemented SERvartuka on top of OpenSER and measured performance benefits of different server configurations on a BladeCenter. An example of our results is a 15% percent increase in call throughput when using our algorithm for a configuration of two servers in sequence.