Data Science experts spend lots of time solving problems using streaming data processing. There are many misconceptions about modern stream process space . Here are few of them There's no streaming without batch :  These limitations existed in earlier version of Apache Storm and are no more relevant in modern stream processing architectures such as Flink. Latency and Throughput: Choose One : A good engineer software like Flink is capable of low latency and high throughput. It has been shown to handle 10s of millions of events per second in a 10-node cluster. Micro-batching means better throughput : Though streaming framework will not rely on batch processing, but it will buffer at the physical level. Exactly once? Completely impossible: Flink is able to provide exactly one state which guarantees under failure by reading both input stream position and the corresponding state of the operator. Earlier traditional data flow had to be interrupted and stored in applications to interact, but new patterns such as CQRS can be developed on continuously flowing data. As the stream processing further evolves we will have more power computational models. You can read more at : http://dataconomy.com/2017/02/stream-processing-myths-debunked/