WindFlow 4.2.0

New release with some modifications:
-various bug fixing in the implementation of persistent operators (db_handle);
-new operator included: IntervalJoin with two parallel execution modes (key parallelism, data parallelism).

WindFlow 4.0.0

This new release of WindFlow comes with several optimizations and code refactoring activities. Furthermore, it includes a new set of persistent operators where the internal state of operators is kept on RocksDB (a popular Key-Value Store). They can be useful in scenarios with "larger-than-memory states".

WindFlow release 3.6.0

This new version of WindFlow provides the following main changes:

1. the source code of the library has been intensively improved and re-factorized in order to avoid code duplication as much as possible and to have a better class hierarchy in several parts of the project;
2. a new operator has been added with the name Ffat_Windows_GPU. It replaces the old Ffat_Aggregator_GPU operator by improving its performance a lot. Now, window-based operators running associative aggregates can be efficiently executed on GPU with out-of-order input streams.

WindFlow release 3.4.0

WindFlow 3.4.0 released on 3rd November 2022.

Major changes:

• added the withBroadcast() distribution policy for Map, Filter, FlatMap and Sink basic operators.
  Minor changes:
• bug fixing
• code improvements and cleaning.

WindFlow release 3.2.0

Main new feature is the integration with Apache Kafka. This is done by two special operators (Kafka_Source and Kafka_Sink), which allow WindFlow applications to subscribe to Kafka topics, and to publish results to Kafka brokers.

WindFlow release 3.1.0

New release of WindFlow with an optimized implementation of the runtime system for GPU operators with full support to CUDA unified memory. Minor cleaning of the code.

WindFlow version 3.0.0

First version of the new branch 3.x. This is a completely redesigned version of the WindFlow library, with code fixing, support for GPU operators, batching and other extensions. Further information can be found in the readme and in the web page of the project.

WindFlow version 2.8.8

New version of WindFlow. Major changes:
1-Web dashboard to monitor the execution of WindFlow applications through a Web REST server;
2-Removed dummy_emitter nodes in the run-time system thanks to a recent modification of FastFlow;
3-Cleaned the code of GPU operators to work on embedded GPUs with ARM processors.

WindFlow version 2.8.6

New release of the library with following important extensions:
1-Added a new processing mode of the PipeGraph named PROBABILISTIC. It allows streams to be processed in order in a probabilistic manner (tupes might be dropped);
2-Cleaned and improved the code of GPU-based windowed operators;
3-Removed the GRAPHVIZ_WINDFLOW macro. Its behavior is now subsumed by the TRACE_WINDFLOW macro, which enables the logging at the WindFlow level (now in JSON format) and the generation of a graphical representation of the PipeGraph.

WindFlow version 2.8.4

New release of WindFlow with the following main features added to the source code:
1-Key_Farm, Key_Farm_GPU, Key_FFAT and Key_FFAT_GPU now work with DEFAULT execution mode (out-of-order processing) with count-based window semantics;
2-Minor modifications to the CUDA kernels used by the Key_FFAT_GPU operator.