v7.0

The 7.0 release, including updates to the following components:

Chimbuko backend

• Added anomaly "post-pruning"; a second pass over recorded anomalies is performed at the end of the run, re-evaluating stored anomalies against the final AD model and discarding those no longer considered anomalous. This significantly reduces the number of mislabeled anomalies resulting from an unconverged AD model.
• Major refactoring of backend codebase, separating out generic functionality and that specific to analyzing performance trace data. Additional modules can now be created to analyze other streaming data types.
• Improvements to services including:
  • Finer control over network interfaces used by service components
  • Optional NUMA binding of service components
  • Additional controls over the frequency of provenance database and pserver sends from the AD modules to reduce network load
• Improvements to AD algorithm robustness for several edge cases
• Various bugfixes

Visualization

• Updated dependencies for the visualization component

Offline analysis

• Added a new tool for converting Chimbuko's provenance database to a relational database
• Added a preliminary version of a new Python library for offline analysis built around the relational database

Full Changelog: v6.5...v7.0

Chimbuko-v6.5

The 6.5 release, which includes:

• Offline analysis command-line tooling that now supports interactive parsing and summarizing of global provenance data
• Support for the Cray CXI provider used by HPE Slingshot 11 networks such as Frontier
• Improved ease-of-use for using the Infiniband verbs provider
• Significant performance optimizations and robustness/veracity improvements to the HBOS algorithm implementation
• Experimental support for launching Chimbuko's services and online anomaly detection components through a single script
• Various fixes and code improvements

Chimbuko-v6.0

The 6.0 release, which includes:

• An overhaul of the visualization frontend workflow allowing for easier isolation of problematic ranks/functions
• New offline analysis tools including summary generation for single runs and for between-run analysis
• Significant enhancements to the utility of the provenance data including incorporating system state information from TAU's monitoring plugin
• Improved workflow support
• Various fixes and optimizations

Chimbuko-v5.5

The 5.5 release, which includes:

• Significant improvements to thread scalability of the parameter server
• Major improvements to histogram creation and merging to improve the correctness of the AD model for HBOS and COPOD algorithms
• Improvements to the implementation of the HBOS and COPOD algorithms based on extensive testing
• Expanded function profile capture in the global provenance database to include runtime and anomaly statistics by function
• Expanded options for controlling Chimbuko's functionality including the ability to skip anomaly detection for certain functions and changing the frequency the analysis is performed
• Expanded functionality for interacting with the provenance database through Python scripts
• Various fixes and optimizations

Chimbuko-v5.0

The 5.0 release, which includes:
- A streaming implementation of the COPOD algorithm for improved anomaly detection,
- Changes to the provenance database to dramatically improve scalability and performance including support for running multiple independent server instances,
- New anomaly metrics for sorting and filtering the most important anomalies,
- Support for building and running Chimbuko in the absence of MPI,
- Various optimizations

Chimbuko-v4.5

The 4.5 release, which includes support for installation via Spack, a new script to simplify launching the Chimbuko services, a streaming implementation of the HBOS algorithm for improved anomaly detection, and various optimizations.

Chimbuko-v4.0

The fourth release. It contains an optimized and scalable online data analysis module complete with a new provenance database component that maintains detailed information on each detected anomaly, as well as support for traces of GPU kernel activity. Also contained is a re-implementation of the visualization module.

Chimbuko-v3.0

The third release. It contains scalable online performance data analysis and visualization.

Chimbuko-v2.0

The second release. Contains online performance data analysis and visualization.

Chimbuko-v1.0

The first release for Chimbuko.