Dynamic Predictive Algorithms for Seamless Scaling of Social Media Applications in Multi-Provider Clouds

Professor Wu, Chuan   (Principal Investigator (PI))

Dr Lin Xiaojun   (Co-Investigator)

Dr Li Zongpeng   (Co-Investigator)

42

2012-12-01

905425

Dynamic Predictive Algorithms for Seamless Scaling of Social Media Applications in Multi-Provider Clouds

Algorithm design, Cloud computing, Optimization, Social media application

Network

Engineering (E)

HKU 717812E

General Research Fund (GRF)

2012

Completed

1) [Joint Optimization Framework]: We formulate an optimization problem for optimal social media service deployment over the long run, which models various charging plans and inter-connection conditions of different cloud data centers; 2) [Demand Prediction]: We build an analytical model to describe information propagation following social connections and content correlations, and use the model to predict future content demand from each region; 3) [Dynamic Algorithm Design]: We design dynamic algorithms pursuing long-term optimality of the optimization problem, by exploiting forecasted demands and problem-specific structures. We will carry out extensive theoretical analysis to show their effectiveness; 4) [Implementation and Evaluation]: We realize and evaluate our algorithms using large-scale experiments on a distributed cloud testbed we are building, as well as in public clouds such as Amazon EC2.