Performance prediction and evaluation for SOA based applications assist software consumers to estimate their applications based on service specifications created by service developers. Incorporating traditional performance models such as Stochastic Petri Nets, Queuing Networks, and Simulation present drawbacks of SOA based applications due to special characteristics of SOA such as lose coupling, self-contained and interoperability. Although, researchers have suggested many methods in this area during last decade, none of them has obtained popular industrial use. Based on this, we have conducted a comprehensive survey on these methods to estimate their applicability. This survey classified these approaches according to their performance metrics analyzed, performance models used, and applicable project stage. Our survey helps SOA architects to select the appropriate approach based on target performance metric and researchers to identify the SOA state-of-art performance prediction.