Intel® Trace Analyzer and Collector User and Reference Guide

Download PDF
ID 767272
Date 3/31/2023
Public
Document Table of Contents
Document Table of Contents x
Intel® Trace Analyzer and Collector User and Reference Guide
Intel® Trace Analyzer and Collector User and Reference Guide x
Introduction Install and Set Up Intel® Trace Analyzer and Collector Trace Your Applications Analyze Your Applications Intel® Trace Collector Reference Intel® Trace Analyzer Reference Notices and Disclaimers
Introduction x
Notational Conventions Get Help
Trace Your Applications x
Tracing Conventional MPI Applications Tracing Failing MPI Applications Tracing OpenSHMEM* Applications Tracing MPI File IO Handling of Communicator Names Tracing MPI Load Imbalance Tracing User Defined Events Configuring the Collector Filtering Trace Data Recording OpenMP* Regions Information Tracing System Calls (Linux* OS) Collecting Lightweight Statistics Recording Source Location Information Recording Hardware Performance Information (Linux* OS) Recording Operating System Counters Tracing Library Calls Correctness Checking Tracing Distributed Non-MPI Applications
Correctness Checking x
Correctness Checking of MPI Applications Running with Valgrind* (Linux* OS) Configuring Error Checks Analyzing the Results Debugger Integration
Debugger Integration x
TotalView* Debugger GNU* Symbolic Debugger Allinea* Distributed Debugging Tool* (DDT*)
Analyze Your Applications x
Starting Intel® Trace Analyzer Intel Trace Analyzer Graphical User Interface Navigating Timelines Concepts Viewing Correctness Checking Reports Comparing Two Trace Files Interoperability with Intel® VTune™ Profiler and Intel® Advisor OpenMP* Regions Display Support OTF2 Format Support
Navigating Timelines x
Zoom Stack
Concepts x
Level of Detail Aggregation Advanced Aggregation Tagging and Filtering
Viewing Correctness Checking Reports x
Event Timeline Correctness Checking Reports Qualitative Timeline Correctness Checking Reports Detailed Dialog
Comparing Two Trace Files x
Mappings in Comparison Views Comparison Charts
Mappings in Comparison Views x
Mapping of Processes Mapping of Functions
Comparison Charts x
Comparison Function Profile Comparison Message Profile Comparison Collective Operations Profile
Intel® Trace Collector Reference x
API Reference Configuration Reference Correctness Checking Errors Structured Tracefile Format stftool Utility Time Stamping Secure Loading of Dynamic Link Libraries* on Windows* OS
API Reference x
Initialization, Termination and Control Defining and Recording Source Locations Defining and Recording Functions or Regions Defining and Recording Scopes Defining Groups of Processes Defining and Recording Counters Recording Communication Events Additional API Calls in libVTcs C++ API
Initialization, Termination and Control x
VT_initialize VT_finalize VT_getrank VT_registerthread VT_registernamed VT_registerprefixed VT_getthrank VT_traceon VT_traceoff VT_tracestate VT_symstate VT_flush VT_timestamp VT_timestart VT_setfinalizecallback VT_getdescription VT_countsetcallback
Defining and Recording Functions or Regions x
New Interface Old Interface State Changes
C++ API x
VT_FuncDef Class Reference VT_SclDef Class Reference VT_Function Class Reference VT_Region Class Reference
Configuration Reference x
Configuration File Format Protocol File Configuration Options
Configuration Options x
ACTIVITY ALTSTACK AUTOFLUSH CHECK CHECK-LEAK-REPORT-SIZE CHECK-MAX-DATATYPES CHECK-MAX-ERRORS CHECK-MAX-PENDING CHECK-MAX-REPORTS CHECK-MAX-REQUESTS CHECK-SUPPRESSION-LIMIT CHECK-TIMEOUT CHECK-TRACING CLUSTER COMPRESS-RAW-DATA COUNTER CURRENT-DIR DEADLOCK-TIMEOUT DEADLOCK-WARNING DEMANGLE DETAILED-STATES ENTER-USERCODE ENVIRONMENT EXTENDED-VTF FLUSH-PID FLUSH-PREFIX GROUP HANDLE-SIGNALS INTERNAL-MPI KEEP-RAW-EVENTS LOGFILE-FORMAT LOGFILE-NAME LOGFILE-PREFIX LOGFILE-RANK MEM-BLOCKSIZE MEM-FLUSHBLOCKS MEM-INFO MEM-MAXBLOCKS MEM-MINBLOCKS MEM-OVERWRITE NMCMD OS-COUNTER-DELAY PCTRACE PCTRACE-CACHE PCTRACE-FAST PLUGIN PROCESS PROGNAME PROTOFILE-NAME STATISTICS STATE STF-PROCS-PER-FILE STF-USE-HW-STRUCTURE STOPFILE-NAME SYMBOL SYNC-MAX-DURATION SYNC-MAX-MESSAGES SYNC-PERIOD SYNCED-CLUSTER SYNCED-HOST TIME-WINDOWS (Experimental) TIMER TIMER-SKIP UNIFY-COUNTERS UNIFY-GROUPS UNIFY-SCLS UNIFY-SYMBOLS VERBOSE VT_START_PAUSED VT_COMPRESS_TRACE
Correctness Checking Errors x
Supported Errors How the Collection Works
How the Collection Works x
Parameter Checking Premature Exit Overlapping Memory Detecting Illegal Buffer Modifications Buffer Given to MPI Cannot Be Read or Written Distributed Memory Checking Illegal Memory Access Request Handling Datatype Handling Buffered Sends Deadlocks Checking Message Transmission Datatype Mismatches Data Modified during Transmission Checking Collective Operations Freeing Communicators
Structured Tracefile Format x
STF Components Single-File STF Configuring STF
stftool Utility x
stftool Utility Options Expanded ASCII output of STF Files
Time Stamping x
Clock Synchronization Choosing a Timer
Choosing a Timer x
gettimeofday/_ftime QueryPerformanceCounter CPU Cycle Counter Normalized CPU Cycle Counter MPI_Wtime() High Precision Event Timers POSIX* clock_gettime
Intel® Trace Analyzer Reference x
Graphical User Interface Reference Intel® Trace Analyzer Command Line Interface Reference Filter Expression Grammar otf2-to-stf Utility
Graphical User Interface Reference x
Welcome Page Summary Page Main Menu Bar View Menu Bar View Bars Charts Dialogs Settings
Main Menu Bar x
File Menu Options Menu Project Menu  Windows Menu Help Menu
View Menu Bar x
View Charts Navigate Advanced Layout Comparison Menu
Advanced x
Tagging Specific Events Filtering Events Simulating Ideal Communication Checking Application Imbalance Aggregating Results Aggregating Functions Creating Command Line for Intel® VTune™ Profiler and Intel® Advisor
View Bars x
Toolbar Trace Map Status Bar
Charts x
Event Timeline Qualitative Timeline Quantitative Timeline Counter Timeline Function Profile Message Profile Collective Operations Profile Performance Assistant Common Chart Features
Event Timeline x
Context Menu Filtering and Tagging
Qualitative Timeline x
Context Menu Filtering and Tagging
Quantitative Timeline x
Context Menu Filtering and Tagging
Counter Timeline x
Context Menu Filtering and Tagging
Function Profile x
Flat Profile Load Balance Call Tree Call Graph Context Menu Filtering and Tagging Function Profile Settings
Message Profile x
Context Menu Filtering and Tagging Aggregation Message Profile Settings
Collective Operations Profile x
Context Menu Filtering and Tagging Collective Operations Profile Settings
Dialogs x
Process Aggregation Function Aggregation Function Group Color Editor Filtering Dialog Box Tagging Dialog Box Idealization Dialog Box Imbalance Diagram Dialog Box Trace Merge Dialog Box  Details Dialog Box Source View Dialog Time Interval Selection Configuration Dialogs Find Dialog Box Command line for Intel® VTune™ Profiler and Intel® Advisor Dialog Box OTF2 to STF Conversion Dialog Box Configuration Assistant
Process Aggregation x
Comparison Mode
Function Aggregation x
Comparison Mode
Filtering Dialog Box x
Building Filter Expressions Using Graphical Interface Building Filter Expressions Manually Filter Expressions in Comparison Mode
Details Dialog Box x
Detailed Attributes of Function Events Detailed Attributes of Message Events Detailed Attributes of Collective Operation Events
Configuration Dialogs x
Load Configuration File Dialog Edit Configuration File Dialog
Settings x
Preferences Font Settings Number Formatting Settings
Preferences x
General Preferences Tracefile Preferences Event Timeline Settings Qualitative Timeline Settings Quantitative Timeline Settings Counter Timeline Settings
Notices and Disclaimers x
Appendix A Copyright and Licenses
Intel® Trace Analyzer and Collector User and Reference Guide

Deadlocks

(GLOBAL:DEADLOCK)

Deadlocks are detected through a heuristic: the background thread in each process cooperates with the MPI wrappers to detect that the process is stuck in a certain MPI call. That alone is not an error because some other processes might still make progress. Therefore the background threads communicate if at least one process appears to be stuck. If all processes are stuck, this is treated as a deadlock. The timeout after which a process and thus the application is considered as stuck is configurable with DEADLOCK-TIMEOUT.

The timeout defaults to one minute which should be long enough to ensure that even very long running MPI operations are not incorrectly detected as being stuck. In applications which are known to execute correct MPI calls much faster, it is advisable to decrease this timeout to detect a deadlock sooner.

This heuristic fails if the application is using non-blocking calls like MPI_Test() to poll for completion of an operation which can no longer complete. This case is covered by another heuristic: if the average time spent inside the last MPI call of each process exceeds the DEADLOCK-WARNING threshold, then a GLOBAL:DEADLOCK:NO_PROGRESS warning is printed, but the application is allowed to continue because the same high average blocking time also occurs in correct application with a high load imbalance. For the same reason the warning threshold is also higher than the hard deadlock timeout.

To help analyzing the deadlock, Intel® Trace Collector prints the call stack of all process. A real hard deadlock exists if there is a cycle of processes waiting for data from the previous process in the cycle. This data dependency can be an explicit MPI_Recv(), but also a collective operation like MPI_Reduce().

If message are involved in the cycle, then it might help to replace send or receive calls with their non-blocking variant. If a collective operation prevents one process from reaching a message send that another process is waiting for, then reordering the message send and the collective operation in the first process would fix the problem.

Another reason could be messages which were accidentally sent to the wrong process. This can be checked in debuggers which support that by looking at the pending message queues. In the future Intel® Trace Collector might also support visualizing the program run in Intel® Trace Analyzer in case of an error. This would help to find messages which were not only sent to the wrong process, but also received by that processes and thus do not show up in the pending message queue.

In addition to the real hard deadlock from which the application cannot recover MPI applications might also contain potential deadlocks: the MPI standard does not guarantee that a blocking send returns unless the recipient calls a matching receive. In the simplest case of a head-to-head send with two processes, both enter a send and then the receive for the message that the peer just sent. This deadlocks unless the MPI buffers the message completely and returns from the send without waiting for the corresponding receive.

Because this relies on undocumented behavior of MPI implementations this is a hard to detect portability problem. Intel® Trace Collector detects these GLOBAL:DEADLOCK:POTENTIAL errors by turning each normal send into a synchronous send. The MPI standard then guarantees that the send blocks until the corresponding receive is at least started. Send requests are also converted to their synchronous counterparts; they block in the call which waits for completion. With these changes any potential deadlock automatically leads to a real deadlock at runtime and will be handled as described above. To distinguish between the two types, check whether any process is stuck in a send function. Due to this way of detecting it, even the normally non-critical potential deadlocks do not allow the application to proceed.

Parent topic: How the Collection Works
Level Two Title