A Pregel program is a sequence of iterations called supersteps, in each of which a vertex can receive inbound messages that are sent by its neighbors in the previous iteration, and modify its attribute and its edges. In addition, each vertex also sends messages to its neighbors by the end of each superstep. By thinking as a vertex, this abstraction makes it simple to reason about parallel graph processing. All we need to think about is the type of message that each vertex should be receiving, the processing that it should do on its inbound messages, and the message that its neighbors need for the next superstep. Luckily, this message-passing approach is flexible enough to express a large class of graph algorithms. More importantly, a graph algorithm can make use of Spark's scalable architecture to process the messages in bulk and in a synchronous manner. This synchronous model of computation makes it easy to express most graph-parallel algorithms.