In this paper we propose cloud-based demand response (CDR), a novel demand response architecture for fast response times in large scale deployments. The proposed architecture is in contrast to master/slave based demand response where the participants directly interact with the utility using host address-centric communication. CDR leverages data-centric communication, publisher/subscriber and topic-based group communication to make demand response secure, scalable and reliable. To the utility, CDR appears to be a black box function call that takes an input from the utility, e.g., power deficit and gives an output to the utility, e.g., power reduction per customer and the corresponding price incentive. Using this implementation framework, we propose two market-based distributed algorithms (bisection and Illinois methods). The proposed algorithms exhibit at least exponentially fast convergence with O(1) iteration as the number of customers grows and outperform prior work of the dual gradient method in terms of convergence speed while keeping the same messaging overhead.