Watermarks, quantitative markers that indicate no future events within a stream, will have an earlier time stamp and enable order-dependent processing on out-of-order streams. They facilitate reasoning about the completeness of input data from multi-source streams (i.e. different producers) in distributed environments, where events may arrive at different consumers at different times. In this paper, Akidau et al. provide an overview of watermarks, including key use cases and related research. They analyze watermark implementations in Google Cloud Dataflow (which has higher latencies) and Apache Flink (for which grows super-linearly as pipeline depth and worker count increase) and discuss opportunities to streamline watermark generation.