• parallel computing
• data stream processing
• benchmarking
• Apache Storm
• Apache Flink
• performance
• WindFlow

In recent years, the need for continuous processing and real-time analysis of data streams has increased rapidly. In order to achieve high-throughput and low-latency requirements, a stream application can be implemented choosing one of the various Data Stream Processing frameworks that offer suitable abstractions for operator parallelization and distribution.
This work shows a comparison in terms of performance (bandwidth and latency) between traditional Data Stream Processing systems (Apache Storm and Flink) and the WindFlow C++17 library, which is an efficient streaming library developed by the Parallel Programming Models group at the Department of Computer Science of the University of Pisa.
Four real-world Data Stream Processing applications have been implemented using Storm, Flink and WindFlow. Experiments are conducted on a single multi-core machine showing a significant throughput improvement and latency reduction by using the C++ solution with respect to the state-of-the-art frameworks.