Social networking sites such as Orkut, MySpace, Hi5, and Facebook attract billions of visits a day, surpassing the page views of Web Search. These social networking sites provide applications for individuals to establish communities, to upload and share documents/photos/videos, and to interact with other users. Take Orkut as an example. Orkut hosts millions of communities, with hundreds of communities created and tens of thousands of blogs/photos uploaded each hour. To assist users to find relevant information, it is essential to provide effective collaborative filtering tools to perform recommendations such as friend, community, and ads matching. In this talk, I will first describe both computational and storage challenges to traditional collaborative filtering algorithms brought by aforementioned information explosion. To deal with huge social graphs that expand continuously, an effective algorithm should be designed to 1) run on thousands of parallel machines for sharing storage and speeding up computation, 2) perform incremental retraining and updates for attaining online performance, and 3) fuse information of multiple sources for alleviating information sparseness. In the second part of the talk, I will present algorithms we recently developed including parallel Spectral Clustering [1], parallel PF-Growth [2], parallel combinational collaborative filtering [3], parallel LDA, parallel spectral clustering, and parallel Support Vector Machines [4].