PerfExplorer

Thanks for your interest in PerfExplorer! This simple page is hopefully only temporary - more useful content will come at a later time.

For The Impatient

Click here to run PerfExplorer!

(or, right-click, save the JNLP file, and open it with Java Web Start)

Introduction

PerfExplorer is a framework for parallel performance data mining and knowledge discovery. The framework architecture enables the development and integration of data mining operations that will be applied to large-scale parallel performance profiles. PerfExplorer was developed as a research project by Kevin Huck, a Doctoral Candidate in Computer and Information Science at the University of Oregon.

The overall goal of the PerfExplorer project is to create a software framework to integrate sophisticated data mining techniques in the analysis of large-scale parallel performance data.

PerfExplorer supports clustering, summarization, association, regression, and correlation. Cluster analysis is the process of organizing data points into logically similar groupings, called clusters. Summarization is the process of describing the similarities within, and dissimilarities between, the discovered clusters. Association is the process of finding relationships in the data. One such method of association is regression analysis, the process of finding independent and dependent correlated variables in the data. In addition, comparative analysis extends these operations to compare results from different experiments, for instance, as part of a parametric study.

New Features in PerfExplorer 2.0

Analysis Process Automation: PerfExplorer 2 adds analysis process automation through the use of a scripting interface. The interface is in Python, and scripts can be used to build analysis workflows. The Python scripts control the Java classes in the application through the Jython interpreter. There are two types of components which are useful in building analysis scripts. The first type is the PerformanceResult interface, and the second is the PerformanceAnalysisComponent interface. For documentation on how to use the Java classes, see the javadoc, and the example scripts below.
Metadata Encoding and Incorporation: In order to interpret performance results within the context of the application experiment environment, PerfExplorer requires the context metadata. There are three was to get context metadata into the performance profiles - the TAU measurement runtime automatically collects some hardware information, the TAU instrumentation library provides the ability to include application context parameters into the profile data, and metadata can be added to the performcane data when the data is loaded into PerfDMF. In PerfExplorer, this data can be correlated with performance results and included in inference rules (see below) in order to explain performance behavior.
Inference Engine: In order to reason about performance causes, PerfExplorer has incorporated the JBoss Rules engine. PerfExplorer uses parallel computing, hardware and application rule sets to interpret the performance results, and provide suggestions for improving application performance.
Provenance and Persistence: Using the Hibernate Relational Persistence for Java, perfexplorer retains a provenance of analysis results, and also stores all intermediate results. Analysis scripts can use the provenance object to access the intermediate results without maintaining local references to results. Note: The provenance/persistence feature is not fully supported at this time.

Prerequisites

Java Runtime Environment (JRE) 5

PerfExplorer is a Java application, and therefore it requires that Java is installed. If you have Java, make sure the version is Java 5 or better. If you need to install Java, you can get it from Sun.

PerfDMF

PerfDMF is the Performance Data Management Framework from TAU. If you have already installed TAU on your computer, then you already have a copy of PerfDMF, and the appropriate configuration files. See the TAU website for more details. If you aren't hosting a PerfDMF data repository but rather wish to connect to a remote repository, then PerfExplorer will help you configure your connection.

Running PerfExplorer

Click here to run the PerfExplorer Java Web Start file. You can use the JNLP file to run PerfExplorer again later - your platform instructions may vary (see the Java Web Start documentation for more details). If you are running PerfExplorer on Windows, allow Java and PerfExplorer to communicate with the internet when the "Windows Security Alert" window appears.

Also, you will want the Java Console open when you run PerfExplorer. From Sun Developer Network, here are the instructions for enabling the console:

The console is one of many settings that Java Web Start sets to "off" by default. To display the console, run the Java Web Start application. Under File->Preferences, go to the Advanced Tab and select "Show Java Console".

Using PerfExplorer

The full documentation for PerfExplorer 1.0 is located at the TAU web site.

If you want to write scripts for PerfExplorer 2.0, then you need the javadoc for the glue package. The glue package is the set of analysis components and data components for scriptable analysis in PerfExplorer 2.0. Here are some examples of analysis scripts:

...coming soon...
...coming soon...
...coming soon...

If you want to write inference rules for PerfExplorer 2.0, then here are some examples:

...coming soon...
...coming soon...
...coming soon...