VR redefines the well known social network by introducing impressive environment sharing and revolutionary face-to-face interaction. Benefiting from better mobility and affordability, mobile social VR is gaining popularity; however, very little is known about how and how well it performs. In this article, we characterize the workflow of mobile social VR, propose a tentative architecture by drawing lessons from well known techniques, and implement this architecture as a prototype. To analyze mobile social VR's performance and identify potential bottlenecks, we conduct experiments to evaluate and decouple the prototype. We observe that mobile social VR's performance mainly depends on the scene it renders, and existing high-end mobile devices exhaust most GPU resources but still offer poor experience for complicated scenes. Finally, we highlight the major challenges and prospects for future work.