For the Storm is:
1, it is recommended that require pure real-time, can not stand at more than 1 second delay using the scene, such as real-time financial system that requires pure real-time analysis of financial transactions and
2, in addition, if the function for real-time computing and reliable transaction mechanism and the required reliability mechanisms that process the data is completely accurate, nor a more, a no less, may also consider using Storm
3, if needed for peak-peak periods, dynamically adjust in real-time computing program parallelism to maximize the use of cluster resources (usually small companies in a tight cluster resources), you can also consider using Storm
4, if a large data applications, it is pure real-time calculation, need not be performed in the middle interactive SQL queries, complex transformation operator, etc., then with the Storm is a better choice
for Spark Streaming is:
1, if the above applies to the three-point Storm, a real-time scenes are not met, that is not required pure real-time, does not require a powerful and reliable transaction mechanism is not required to dynamically adjust the degree of parallelism, it can be To consider the use of Spark Streaming
2, consider using a Spark Streaming most important factor should be considered for the entire macro-project, that is, if a project is in addition to the real-time calculation, also includes off-line batch processing, interactive query and other services features, and real-time calculation, may also involve high latency batch, interactive query and other functions, then it should be the first choice ecological Spark, Spark Core development with off-line batch processing, interactive development with Spark SQL query development by Spark Streaming real-time calculation, the three can be seamlessly integrated system to provide very high scalability
advantages and disadvantages analysis Spark Streaming and Storm in
fact, Spark Streaming absolutely out of the question better than Storm. Both frameworks in the field of real-time computing, are excellent, but good at the breakdown scene is not the same.
Spark Streaming only on the throughput to be better than Storm, and the throughput of this, is always very Spark Streaming, who banished Storm emphasized. But the problem is, is not at all real-time computing scene, are so focused on throughput? Not really. Thus, the throughput of said Spark Streaming stronger than Storm, do not fly.
In fact, Storm on the degree of real-time latency, was much better than the Spark Streaming former is pure real-time, which is a quasi-real time. Moreover, Storm transaction mechanisms, robustness / fault tolerance, dynamic adjustment of the degree of parallelism and other characteristics, to be more outstanding than the Spark Streaming.
Spark Streaming, one thing is absolutely not as Storm, that is: it is located in the Spark ecological technology stack, and can therefore Spark Streaming Spark Core, Spark SQL seamless integration, which means that we can come out of the middle of real-time processing data, seamless immediately delayed batch, interactive query operations in the program. This feature greatly enhances the benefits and features of Spark Streaming.
Personal humble opinion wrong place, please correct me Gangster