Operation and Use, Volume 2

What is the PE Benchmarker?

The PE Benchmarker is a suite of applications and utilities that you can use to analyze the performance of programs run within the IBM Parallel Environment for AIX. The PE Benchmarker suite consists of:

• the Performance Collection Tool (PCT). This tool enables you to collect either MPI and user event data or hardware and operating system profiles for one or more application processes (or "tasks"). This tool is built on dynamic instrumentation technology, the Dynamic Probe Class Library (DPCL). Unlike more traditional tools for collecting message-passing and other performance information, the PCT, because it is built on DPCL, enables you to insert and remove instrumentation probes into the target application while the target application is running. More traditional tools require the application to be instrumented through compilation or linking. This often results in more instrumentation being inserted into the application than is actually needed, and so such tools are more likely to create situations in which the instrumented version of the application is no longer representative of the actual, uninstrumented, version of the application. Since the PCT enables you to make the decision of what data is collected at run time, this typically results in a more acceptable intrusion cost of the instrumentation. What's more, the files output by the PCT are output on each machine running instrumented processes rather than on a single, centralized, machine. This means that your analysis can be efficiently scaled to collect information on a large number of processes running on a large number of nodes.
• a set of Unified Trace Environment (UTE) utilities. When you collect MPI and user event traces using the PCT, the collected information is saved, on each machine running instrumented processes, as a standard AIX event trace file. The UTE utilities enable you to convert one or more of these AIX trace files into UTE interval files. While an AIX event trace file has a time stamp indicating the point in time when an event occurred, UTE interval files take this information to also determine how long an event lasts before encountering the next event. Because they include this duration information, UTE interval files are easier to visualize than traditional AIX event trace files. The UTE utilities are:
  • the uteconvert utility which converts AIX event trace records into UTE interval trace files.
  • the utemerge utility which merges multiple UTE interval files into a single UTE interval file.
  • the utestats utility which generates statistics tables from UTE interval files.
  • the slogmerge utility which converts and merges UTE interval files into a single SLOG file for analysis within Argonne National Laboratory's Jumpshot tool.
• the Profile Visualization Tool (PVT). When you collect hardware and operating system profiles using the PCT, the collected profile information is saved, on each machine running instrumented processes, as netCDF (network Common Data Form) files. The PVT can read netCDF files and summarize the profile information in reports.

The following figure illustrates how the various tools in the PE Benchmarker toolset work together to enable you to analyze the performance of programs run within the IBM AIX Parallel Environment. Please note that Jumpshot is not part of the PE Benchmarker toolset, but is instead a public domain tool developed at Argonne National Laboratory. It is shown in the figure below, because PE Benchmarker provides the slogmerge utility for converting UTE files into the SLOG format required by Jumpshot.

Figure 36. Overview of the PE Benchmarker Toolset

View figure.

The preceding figure illustrates the procedure for collecting and analyzing data using the PE Benchmarker toolset. This procedure starts with the PCT. When using the PCT, you must select the type of data you are collecting -- either MPI and user event trace data or hardware and operating system performance data. You use the PCT to connect to existing processes, or start processes running (which also connects to the processes). By "connect to processes" we mean the PCT establishes a communication connection that enables it to control the process' execution (suspend, resume, and terminate the process), and also instrument the process with data collection probes. Data files containing the collected information will be generated on each machine running at least one instrumented process. The format of the files generated depends on the type of data you are collecting.

• If you are collecting MPI and user event trace data, standard AIX trace files will be generated. You will first need to take the AIX trace files output by the PCT and convert them, using the uteconvert utility, into UTE interval files. If you want to view statistical tables of the information contained in the UTE interval files, you can use the utestats utility. You can optionally merge multiple UTE files into a single UTE file using the utemerge utility before using the utestats utility to generate the statistical tables. If you instead want to view the information contained in the UTE interval files graphically, you can convert them into SLOG files which are readable by Argonne National Laboratory's Jumpshot tool. To convert UTE interval files into SLOG files, you use the slogmerge utility. The slogmerge utility can convert a single UTE interval file into a single SLOG file, or it can convert multiple UTE interval files into a single, merged, SLOG file.
• If you are collecting hardware performance data, netCDF files will be generated. You can use the PVT to generate graphs and reports of the information contained in the netCDF files.