v0.4

Major improvements:

• Data Reduction framework
  • New abstract ReductionMethod interface for extensible data reduction strategies
  • Decimation method: spatial subsampling with per-dimension stride, optional interpolation (linear, quadratic, cubic), configurable computational cost, and support for both publisher-side and subscriber-side application
  • Compression method: size reduction with preserved shape, supporting three compressor profiles (fixed ratio, SZ-inspired, and ZFP-inspired models), separate compression and decompression costs, and per-transaction ratio variability
  • Publisher-side reduction state is propagated to subscribers through inquire_variable, enabling detection and prevention of conflicting double reductions
• Reduction operations work with both File and Staging engines
• New documentation pages for the reduction feature
• Improved test coverage

API Changes:

• New Stream method:
  • Stream::define_reduction_method(name) creates a named reduction method ("decimation" or "compression") for the stream
• New Variable methods and properties:
  • Variable::set_reduction_operation(method, parameters) applies a reduction with key-value parameters to a variable
  • Variable::is_reduced(), is_reduced_by_publisher(), is_reduced_by_subscriber() query the reduction state
  • Variable::get_reduction_method() retrieves the applied ReductionMethod
• New ReductionMethod class with query methods:
  • get_reduced_variable_global_size(), get_reduced_variable_local_size()
  • get_reduced_variable_shape()
  • get_flop_amount_to_reduce_variable(), get_flop_amount_to_decompress_variable()
• Full Python bindings for all reduction operations
• Engine::put() now automatically simulates reduction cost and transports the reduced data size when the variable is reduced
• Engine::get() now automatically simulates decompression cost after receiving compressed data

v0.3

Major improvements:

• Enhanced code quality with comprehensive refactoring
  • Improved modern C++17 usage throughout the codebase
  • Reduced code complexity and technical debt
• Memory safety and robustness improvements
  • Fixed critical memory safety issue in DTL connection management
  • Fixed race condition in Engine creation
  • Plugged memory leaks
• Major refactoring with improved memory management, encapsulation, and move-only semantics
• Improved CI/CD infrastructure
  • Added Valgrind and sanitizer checks (AddressSanitizer, UndefinedBehaviorSanitizer)
  • Enhanced test coverage and validation
• Comprehensive proofread of the documentation

API Changes:

• DTL::get_stream_by_name_or_null() renamed to DTL::get_stream_by_name()
  Returns std::optional<std::shared_ptr> instead of raw pointer
  Python: dtl.get_stream_by_name() now returns None if stream not found
• Metadata management transferred from Engine to Stream
  Stream::get_metadata_file_name() replaces Engine::get_metadata_file_name()
  Python: stream.metadata_file_name replaces engine.metadata_file_name
• Stream method chaining improvements
  set_engine_type() and set_transport_method() now return Stream& instead of Stream*
  Enables cleaner fluent interface: stream.set_engine_type(...).set_transport_method(...)
• DTL::create() now accepts std::string_view and has empty default parameter
• String parameters changed to std::string_view for better performance:
  • DTL::add_stream(), DTL::get_stream_by_name()
  • Stream and Variable constructors and methods
• get_engine_type_str() and get_transport_method_str() now return std::optional<const char*>

v0.2

• New comprehensive documentation
• Full python bindings
• Improved code coverage
• Reduced technical debt
• Fixed deadlocks in File engine
• Update to SimGrid 4.1 and FSMod 0.4
• API:
  • Make metadata export configurable at stream level.
  • Have a simpler version of Engine::put() that automatically uses the local size of the Variable.

v0.1

This is the first release of the SimGrid Data Transport Layer Module (DTLMod), which provides a versatile simulated data transport layer for use in any SimGrid-based simulator.