Visualization of Repetitive Patterns in Event Traces (original) (raw)

Abstract

Performance Tracing has always been challenged by large amounts of trace data. Software tools for trace analysis and visualization successfully cope with ever growing trace sizes. Still, human perception is unable to "scale up" with the amounts of data. With a new model of trace visualization, we try to provide less data but additional information resp. more convenient information to human users. By marking regular repetition patterns and hiding the inner details less complex visualization can offer better insight. At the same time irregular sections are revealed which are potentially interesting. The paper introduces the origin of repetition patterns and outlines the detection algorithm used. It demonstrates the new visualization scheme which has also been incorporated into Vampir NG as a prototype. Finally, it gives an outlook on further development and possible extentions.

Key takeaways

AI

1. The proposed visualization model enhances insight by reducing data while emphasizing repetitive patterns.
2. Large trace sizes often exceed tens of gigabytes due to increased parallelism and execution counts.
3. The pattern detection algorithm operates with complexity bounded by O(n•m), optimizing performance analysis.
4. Vampir NG incorporates the new visualization and pattern detection scheme as a prototype.
5. Future work includes accepting variations in patterns and improving the rendering of timeline diagrams.

Loading Preview

Sorry, preview is currently unavailable. You can download the paper by clicking the button above.

References (3)

1. Andreas Knüpfer, Wolfgang E. Nagel: Compressible Memory Data Structures for Event-Based Trace Analysis. Future Generation Computer Systems, Volume 22, Issue 3, February 2006, pages 359-368
2. Bernhard Voigt: Effiziente Erkennungs-und Visualisierungsmethoden für hierarch- ische Trace-Informationen. Diploma thesis, TU Dresden, Germany, 2006
3. Holger Brunst, Wolfgang E. Nagel, Allen D. Malony: A Distributed Performance Analysis Architecture for Clusters. Proc. of IEEE International Conference on Clus- ter Computing, Hong Kong, China, IEEE Computer Society, pages 73-81, 2003

FAQs

AI

What explains the growth of trace data in high-performance computing?add

The study finds that trace sizes primarily double with iteration counts or run-time increases and parallel process numbers, often leading to tens of gigabytes of data in modern applications.

How does the new visualization methodology differ from traditional approaches?add

The proposed methodology utilizes Process Timeline Diagrams that decompose patterns interactively, allowing users to access underlying structures without being overwhelmed by data complexity.

What is the complexity of the proposed pattern detection algorithm?add

The complexity for pattern search in the compressed complete call graphs is bounded by O(n•m), where n is the number of nodes and m is the maximum number of child nodes.

When could pattern matching allow variations in high-performance trace analysis?add

Future adaptations of the pattern matching process may allow for structural equality among loops with different iteration counts, improving the flexibility of analysis.

Why might additional information be overlooked in traditional visualization methods?add

Users often cannot perceive regular or irregular behaviors due to overwhelming details in traditional visualizations, leading to critical insights being hidden in large data sets.