Speed up parallel application runs, locate hotspots and bottlenecks, and increase productivity with the MPI performance analysis features of Intel Trace Analyzer and Collector 7.0, with this rich set of features:
- Interface and Displays
- Metrics Tracking
- Scalability
- Instrumentation and Tracing
- Compatibility
Trace File Comparison - Simplify the development cycle by comparing two trace files and/or two regions (in one or two trace files) with statistical analysis and aligned timeline displays.
Counter Timeline – Gain insight of parallel system behaviour through recording performance counter data. Analyze counter data collected through provided PAPI and OS modules or through manual use of Intel Trace Collector API.
MPI Checking, with an optional correctness checking library - A novel MPI correctness technology to detect errors with data types, buffers, communicators, point-to-point messages and collective operations, deadlocks, or data corruption.
- Intercepts and analyzes MPI calls as the application runs
- Reports errors to console as they are detected
- Can trigger debugger breakpoints for in-place analysis
- Works with Intel MPI 2.x and 3.x
- Over 20 distinct MPI programming and run-time errors are detected
Improved usability – New ‘undo’ function, and customization options for fonts, number format, colors, and other features.
Integrated online help for Intel Trace Analyzer – Just press F1 to view the user manual or context sensitive help.
Easy run-time loading of Intel Trace Collector –Instrument existing MPI binary applications, either by dynamically loading the Trace Collector Library at run-time, or binary instrumentation of an MPI application executable.
Advanced GUI: User-friendly interface, high-level scalability, support for structured trace file (SFT) trace data, runs on Linux and Windows.
Aggregating and Filtering: Detailed views of runtime behavior grouped by functions or processes.
MPI Communicator: Display of communication metrics for an arbitrary time interval for MPI.
Fail-Safe Tracing: Improved functionality on prematurely terminating applications with deadlock detection.
Intel MPI Library Interface: Support tracing on internal MPI states.
Support for MPI-IO, ROMIO: Extended support for MPI-2 standard parallel file I/O.
Interface and Displays
Timeline Views and Parallelism Display
- Displays concurrent behavior of parallel applications
- Calculates statistics for specific time intervals, processes, or functions
- Displays application activities, event source code locations, and message-passing along time axis
- Provides timeline displays, call-graph, performance profile for function groups and communication in a specific phase of parallel execution.
Advanced GUI
- Manages displays and statistics with fast interface
- Uses object point-and-zoom for enhanced detail, context-sensitive sub-menus, and coupled displays
- Supplies automatic updates of recomputed statistics
Display Scalability
- Navigates through trace data levels of abstraction: cluster, node, process, thread, and function
Detailed and Aggregate Views
- Examines aspects of application runtime behavior, grouped by functions or processes
- Easily identifies the amount of time spent in MPI communication
- Easily see the performance differences between two program runs
Ease of Use
- Offers user-friendly application programming interface (API) to control or record user events
- Adds versatile recording and analysis of counter data (see Figure 3)
Metrics Tracking
Communication Statistics
- Displays concurrent behavior of parallel applications
- Displays metrics for an arbitrary time interval
- Keeps track of the performance increase of an algorithm change
Execution Statistics
- Provides subroutine execution metrics or call-tree characteristics
Profiling Library
- Records distributed, event-based trace data
Statistics Readability
- Logs information for function calls, sent messages, and collective operations
Scalability
Low Overhead
- Provides structured trace file (STF) format for scalability
- Generates trace files faster
- Allows random access to portions of a trace, making it suitable for analysis of large amounts of trace data
Thread Safety
- Traces multi-threaded MPI applications for event-based tracing to non-MPI applications
Fail-Safe Mode
- Writes trace files, even for premature exit of program
Filtering and Memory Handling
- Caches trace data in memory to reduce runtime overhead and memory consumption
Instrumentation and Tracing
Low Intrusion Instrumentation
- Supports MPI applications with C, C++, or Fortran
- Library supports distributed non-MPI applications, with C, C++, Fortran, or Java
- Automatically records performance data from parallel threads in C, C++, Fortran, or Java multithreaded processes
Binary Instrumentation (for IA-32 and Intel® 64 architecture)
- Creates and adds instrumentation to existing binary executables which allows automatic monitoring of function entry/exit points for detailed analysis of user code and application runtime
- Traces C++ and Java class hierarchies with instrumentation capabilities
Intel MPI Library
- Supports tracing of internal MPI states
Hardware
Minimum Requirements:
- Intel Pentium III processor 450-MHz or later
- 512 MB of RAM
- 100 MB of free hard disk space
Recommended:
IA-32-based systems or Intel 64 architecture-based systems:
- Intel Xeon processor
- 1 GB of RAM
- 100 MB of free hard disk space
Itanium 2-based systems:
- Itanium 2 processor
- 1 GB of RAM
- 1 GB of free hard disk space
Software
Minimum Requirements:
IA-32-based systems or Intel 64 architecture-based systems:
- Red Hat Enterprise Linux* 3.0
Itanium 2-based systems:
- Red Hat Enterprise Linux 3.0
Recommended
IA-32-based systems or Intel 64 architecture-based systems:
- Red Hat Enterprise Linux 4.0
- SUSE LINUX Enterprise Server* 9 or 10
- Each with Linux MPI implementations such as:
Intel® MPI Library
MPICH (or compatible)
Itanium 2-based systems:
- Red Hat Enterprise Linux 4.0
- SUSE LINUX Enterprise Server 9 or 10
- SGI Altix
- Each with Linux MPI implementations such as:
Intel MPI Library
MPICH (or compatible)
SGI Message Passing Toolkit
Supported
- Intel MPI Library
- Intel® MPI Benchmarks
- Intel® Math Kernel Library Cluster Edition
- Intel® Cluster Toolkit
- Intel® C++ Compiler for Linux
- Intel® Fortran Compiler for Linux
- GNU C, C++, and FORTRAN77 Compilers