Load Distribution Performance of the Reliable Server Pooling Framework

The Reliable Server Pooling (RSerPool) protocol suite currently under standardization by the IETF is designed to build systems providing highly available services by providing mechanisms and protocols for establishing, configuring, accessing and monitoring pools of server resources. While availability is one main aspect of RSerPool, load distribution is another. Since most of the time a server pool system runs without component failures, optimal performance is an extremely important issue for the productivity and costefficiency of the system. In this paper, we therefore focus especially on the load distribution performance of RSerPool in scenarios without failures, presenting a quantitative performance comparison of the different load distribution strategies (called pool policies) defined in the RSerPool specifications. Based on the results, we propose some new pool policies providing significant performance enhancements compared to those currently defined in the standards documents. 1 The Reliable Server Pooling Architecture The convergence of classical circuit-switched networks (i.e. PSTN/ISDN) and data networks (i.e. IP-based) is rapidly progressing. This implies that SS7 PSTN signalling [1] has to also be transported over IP networks. Since SS7 signalling networks offer a very high degree of availability (e.g. at most 10 minutes downtime per year for any signalling relation between two signalling endpoints; for more information see [2]), all links and components of the network devices must be fault-tolerant, and this is achieved through having multiple links, and using the link redundancy concept of SCTP [3]. When transporting signalling over IP networks, such concepts also have to be applied to achieve the required availability. Link redundancy in IP networks is supported using the Stream Control Transmission Protocol (SCTP) providing multiple network paths and fast failover [4,5]; redundancy of network device components is supported by the SGP/ASP (signalling gateway process/application server process) concept. However, this concept has some limitations: there is no support of dynamic addition and removal of components; it has only limited ways of server selection and no specific failover procedures and inconsistent application to different SS7 adaptation layers. To cope with the challenge of creating a unified, lightweight, realtime, scalable and extendable redundancy solution (see [6] for details), the IETF Reliable Server Pooling Working Group was founded to specify and define the Reliable Server Pooling Concept. An overview of the architecture currently under standardization and described by several Internet Drafts is shown in figure 1. ... Application Protocol ASAP Protocol ASAP Protocol