Grape is a distributed system designed specifically for ``real-time'' continuous processing of ``big'' streaming data on a large cluster of commodity machines. The unique characteristics of this emerging application domain have led to a significantly different design from now ``traditional'' MapReduce style batch data processing. In particular, we advocate a powerful new abstraction called resilient substitution that caters to the specific needs in this new computation model.