Self-Organizing Mechanism for Cloud Services and Performance Analysis

To improve the performance and robustness in service discovery, a self-organizing mechanism for service alliances of Service Providers (SPs) is proposed in this paper. According to the similarity of service content, an SP publishes its services in a partition of SPs to construct connections between highly similar SPs. These SPs constitute a self-organized distributed environment. A self-organizing protocol is designed to ensure the correctness of the construction of the alliances. The protocol consists of four stages initiating stage, developing stage, developed stage and degradation stage. The experimental results demonstrate that this protocol ensures the self-property. The visualization of alliance developing stages illustrates that sub-alliances are split in balance and self-connected. Compared with the Random Walker algorithm, the time cost and the number of forwarded messages in alliance-based mechanism is lower in service discovery. On three typical topologies (Grid, Random-Graph, Power-Law), the success rate of service discovery is much higher, which shows that self-organized alliances are helpful to enhance the discovery performance.