Table of Contents
This document describes darshan-util, a collection of tools for parsing and summarizing log files produced by Darshan instrumentation. The darshan-util package can be installed and used on any system regardless of where the logs were originally generated. Darshan log files are platform-independent.
More information about Darshan can be found at the Darshan web site.
Darshan-util has only been tested in Linux environments, but will likely work in other Unix-like environments as well.
Hard requirements
Optional requirements
Configure and build example.
tar -xvzf darshan-<version-number>.tar.gz cd darshan-<version-number>/ ./prepare cd darshan-util/ ./configure make make install
The darshan-util package is intended to be used on a login node or
workstation. For most use cases this means that you should
either leave CC to its default setting or specify a local compiler. This is
in contrast to the darshan-runtime documentation, which suggests setting CC
to mpicc because the runtime library will be used in the compute node
environment.
You can specify --prefix to install darshan-util in a specific location
(such as in your home directory for non-root installations). See
./configure --help for additional optional arguments, including how to
specify alternative paths for zlib and libbz2 development libraries.
darshan-util also supports VPATH or "out-of-tree" builds if you prefer that
method of compilation.
The --enable-shared argument to configure can be used to enable
compilation of a shared version of the darshan-util library.
The --enable-apmpi-mod and --enable-apxc-mod configure
arguments must be specified to build darshan-util with support for AutoPerf
APMPI and APXC modules, respectively.
AutoPerf log analysis code is provided as Git submodules to Darshan’s main repository, so if building Darshan source that has been cloned from Git, it is neccessary to first retrieve the AutoPerf submodules by running the following command:
git submodule update --init
Each time a darshan-instrumented application is executed, it will generate a single log file summarizing the I/O activity from that application. See the darshan-runtime documentation for more details, but the log file for a given application will likely be found in a centralized directory, with the path and log file name in the following format:
<YEAR>/<MONTH>/<DAY>/<USERNAME>_<BINARY_NAME>_<JOB_ID>_<DATE>_<UNIQUE_ID>_<TIMING>.darshan
This is a binary format file that summarizes I/O activity. As of version
2.0.0 of Darshan, this file is portable and does not have to be analyzed on
the same system that executed the job. Also, note that Darshan logs generated
with Darshan versions preceding version 3.0 will have the extension darshan.gz
(or darshan.bz2 if compressed using bzip2 format). These logs are not compatible
with Darshan 3.0 utilities, and thus must be analyzed using an appropriate version
(2.x) of the darshan-util package.
You can generate a graphical summary
of the I/O activity for a job by using the darshan-job-summary.pl graphical summary
tool as in the following example:
darshan-job-summary.pl carns_my-app_id114525_7-27-58921_19.darshan.gz
This utility requires Perl, pdflatex, epstopdf, and gnuplot in order to
generate its summary. By default, the output is written to a multi-page
pdf file based on the name of the input file (in this case it would
produce a carns_my-app_id114525_7-27-58921_19.pdf output file).
You can also manually specify the name of the output file using the
--output argument.
An example of the output produced by darshan-job-summary.pl can be found HERE.
NOTE: The darshan-job-summary tool depends on a few LaTeX packages that may not
be availalble by default on all systems, including: lastpage, subfigure, and
threeparttable. These packages can be found and installed using your system’s package
manager. For instance, the packages can be installed on Debian or Ubuntu systems as
follows: apt-get install texlive-latex-extra
This utility is similar to darshan-job-summary.pl, except that it produces a separate pdf summary for every file accessed by an application. It can be executed as follows:
darshan-summary-per-file.sh carns_my-app_id114525_7-27-58921_19.darshan.gz output-dir
The second argument is the name of a directory (to be created) that will contain the collection of pdf files. Note that this utility probably is not appropriate if your application opens a large number of files.
You can produce a summary for a specific file of interest with the following commands:
darshan-convert --file HASH carns_my-app_id114525_7-27-58921_19.darshan.gz interesting_file.darshan.gz darshan-job-summary.pl interesting_file.darshan.gz
The "HASH" argument is the hash of a file name as reported by
darshan-parser. The interesting_file.darshan.gz file produced by
darshan-convert is like a normal Darshan log file, but it will only contain
instrumentation for the specified file.
You can use the darshan-parser command line utility to obtain a
complete, human-readable, text-format dump of all information contained
in a log file. The following example converts the contents of the
log file into a fully expanded text file:
darshan-parser carns_my-app_id114525_7-27-58921_19.darshan.gz > ~/job-characterization.txt
The format of this output is described in the following section.
The beginning of the output from darshan-parser displays a summary of
overall information about the job. Additional job-level summary information
can also be produced using the --perf, --file, or --total command line
options. See the Additional summary output section for more
information about those options.
The following table defines the meaning of each line in the default header section of the output:
| output line | description |
|---|---|
"# darshan log version" | internal version number of the Darshan log file |
"# exe" | name of the executable that generated the log file |
"# uid" | user id that the job ran as |
"# jobid" | job id from the scheduler |
"# start_time" | start time of the job, in seconds since the epoch |
"# start_time_asci" | start time of the job, in human readable format |
"# end_time" | end time of the job, in seconds since the epoch |
"# end_time_asci" | end time of the job, in human readable format |
"# nprocs" | number of MPI processes |
"# run time" | run time of the job in seconds |
The next portion of the parser output displays the size of each region contained within the given log file. Each log file will contain the following regions:
All regions of the log file are compressed (in libz or bzip2 format), except the header.
The next portion of the output shows a table of all general purpose file systems that were mounted while the job was running. Each line uses the following format:
<mount point> <fs type>
The remainder of the output will show characteristics for each file that was opened by the application. Each line uses the following format:
<module> <rank> <record id> <counter name> <counter value> <file name> <mount point> <fs type>
The <module> column specifies the module responsible for recording this piece
of I/O characterization data. The <rank> column indicates the rank of the process
that opened the file. A rank value of -1 indicates that all processes opened the
same file. In that case, the value of the counter represents an aggregate across all
processes. The <record id> is a 64 bit hash of the file path/name that was opened.
It is used as a way to uniquely differentiate each file. The <counter name> is
the name of the statistic that the line is reporting, while the <counter value> is
the value of that statistic. A value of -1 indicates that Darshan was unable to
collect statistics for that particular counter, and the value should be ignored.
The <file name> field shows the complete file name the record corresponds to. The
<mount point> is the mount point of the file system that this file belongs to and
<fs type> is the type of that file system.
The following tables show a list of integer statistics that are available for each of
Darshan’s current instrumentation modules, along with a description of each. Unless
otherwise noted, counters include all variants of the call in question, such as
read(), pread(), and readv() for POSIX_READS.
Table 1. POSIX module
| counter name | description |
|---|---|
POSIX_OPENS | Count of how many times the file was opened (INCLUDING |
POSIX_FILENOS | Count of POSIX fileno operations |
POSIX_DUPS | Count of POSIX dup operations |
POSIX_READS | Count of POSIX read operations |
POSIX_WRITES | Count of POSIX write operations |
POSIX_SEEKS | Count of POSIX seek operations |
POSIX_STATS | Count of POSIX stat operations |
POSIX_MMAPS | Count of POSIX mmap operations |
POSIX_FSYNCS | Count of POSIX fsync operations |
POSIX_FDSYNCS | Count of POSIX fdatasync operations |
POSIX_RENAME_SOURCES | Number of times this file was the source of a rename operation |
POSIX_RENAME_TARGETS | Number of times this file was the target of a rename operation |
POSIX_RENAMED_FROM | If this file was a rename target, the Darshan record ID of the first rename source |
POSIX_MODE | Mode that the file was last opened in |
POSIX_BYTES_READ | Total number of bytes that were read from the file |
POSIX_BYTES_WRITTEN | Total number of bytes written to the file |
POSIX_MAX_BYTE_READ | Highest offset in the file that was read |
POSIX_MAX_BYTE_WRITTEN | Highest offset in the file that was written |
POSIX_CONSEC_READS | Number of consecutive reads (that were immediately adjacent to the previous access) |
POSIX_CONSEC_WRITES | Number of consecutive writes (that were immediately adjacent to the previous access) |
POSIX_SEQ_READS | Number of sequential reads (at a higher offset than where the previous access left off) |
POSIX_SEQ_WRITES | Number of sequential writes (at a higher offset than where the previous access left off) |
POSIX_RW_SWITCHES | Number of times that access toggled between read and write in consecutive operations |
POSIX_MEM_NOT_ALIGNED | Number of times that a read or write was not aligned in memory |
POSIX_MEM_ALIGNMENT | Memory alignment value (chosen at compile time) |
POSIX_FILE_NOT_ALIGNED | Number of times that a read or write was not aligned in file |
POSIX_FILE_ALIGNMENT | File alignment value. This value is detected at runtime on most file systems. On Lustre, however, Darshan assumes a default value of 1 MiB for optimal file alignment. |
POSIX_MAX_READ_TIME_SIZE | Size of the slowest POSIX read operation |
POSIX_MAX_WRITE_TIME_SIZE | Size of the slowest POSIX write operation |
POSIX_SIZE_READ_* | Histogram of read access sizes at POSIX level |
POSIX_SIZE_WRITE_* | Histogram of write access sizes at POSIX level |
POSIX_STRIDE[1-4]_STRIDE | Size of 4 most common stride patterns |
POSIX_STRIDE[1-4]_COUNT | Count of 4 most common stride patterns |
POSIX_ACCESS[1-4]_ACCESS | 4 most common POSIX access sizes |
POSIX_ACCESS[1-4]_COUNT | Count of 4 most common POSIX access sizes |
POSIX_FASTEST_RANK | The MPI rank with smallest time spent in POSIX I/O (cumulative read, write, and meta times) |
POSIX_FASTEST_RANK_BYTES | The number of bytes transferred by the rank with smallest time spent in POSIX I/O (cumulative read, write, and meta times) |
POSIX_SLOWEST_RANK | The MPI rank with largest time spent in POSIX I/O (cumulative read, write, and meta times) |
POSIX_SLOWEST_RANK_BYTES | The number of bytes transferred by the rank with the largest time spent in POSIX I/O (cumulative read, write, and meta times) |
POSIX_F_*_START_TIMESTAMP | Timestamp that the first POSIX file open/read/write/close operation began |
POSIX_F_*_END_TIMESTAMP | Timestamp that the last POSIX file open/read/write/close operation ended |
POSIX_F_READ_TIME | Cumulative time spent reading at the POSIX level |
POSIX_F_WRITE_TIME | Cumulative time spent in write, fsync, and fdatasync at the POSIX level |
POSIX_F_META_TIME | Cumulative time spent in open, close, stat, and seek at the POSIX level |
POSIX_F_MAX_READ_TIME | Duration of the slowest individual POSIX read operation |
POSIX_F_MAX_WRITE_TIME | Duration of the slowest individual POSIX write operation |
POSIX_F_FASTEST_RANK_TIME | The time of the rank which had the smallest amount of time spent in POSIX I/O (cumulative read, write, and meta times) |
POSIX_F_SLOWEST_RANK_TIME | The time of the rank which had the largest amount of time spent in POSIX I/O (cumulative read, write, and meta times) |
POSIX_F_VARIANCE_RANK_TIME | The population variance for POSIX I/O time of all the ranks |
POSIX_F_VARIANCE_RANK_BYTES | The population variance for bytes transferred of all the ranks |
Table 2. MPI-IO module
| counter name | description |
|---|---|
MPIIO_INDEP_OPENS | Count of non-collective MPI opens |
MPIIO_COLL_OPENS | Count of collective MPI opens |
MPIIO_INDEP_READS | Count of non-collective MPI reads |
MPIIO_INDEP_WRITES | Count of non-collective MPI writes |
MPIIO_COLL_READS | Count of collective MPI reads |
MPIIO_COLL_WRITES | Count of collective MPI writes |
MPIIO_SPLIT_READS | Count of MPI split collective reads |
MPIIO_SPLIT_WRITES | Count of MPI split collective writes |
MPIIO_NB_READS | Count of MPI non-blocking reads |
MPIIO_NB_WRITES | Count of MPI non-blocking writes |
MPIIO_SYNCS | Count of MPI file syncs |
MPIIO_HINTS | Count of MPI file hints used |
MPIIO_VIEWS | Count of MPI file views used |
MPIIO_MODE | MPI mode that the file was last opened in |
MPIIO_BYTES_READ | Total number of bytes that were read from the file at MPI level |
MPIIO_BYTES_WRITTEN | Total number of bytes written to the file at MPI level |
MPIIO_RW_SWITCHES | Number of times that access toggled between read and write in consecutive MPI operations |
MPIIO_MAX_READ_TIME_SIZE | Size of the slowest MPI read operation |
MPIIO_MAX_WRITE_TIME_SIZE | Size of the slowest MPI write operation |
MPIIO_SIZE_READ_AGG_* | Histogram of total size of read accesses at MPI level, even if access is noncontiguous |
MPIIO_SIZE_WRITE_AGG_* | Histogram of total size of write accesses at MPI level, even if access is noncontiguous |
MPIIO_ACCESS[1-4]_ACCESS | 4 most common MPI aggregate access sizes |
MPIIO_ACCESS[1-4]_COUNT | Count of 4 most common MPI aggregate access sizes |
MPIIO_FASTEST_RANK | The MPI rank with smallest time spent in MPI I/O (cumulative read, write, and meta times) |
MPIIO_FASTEST_RANK_BYTES | The number of bytes transferred by the rank with smallest time spent in MPI I/O (cumulative read, write, and meta times) |
MPIIO_SLOWEST_RANK | The MPI rank with largest time spent in MPI I/O (cumulative read, write, and meta times) |
MPIIO_SLOWEST_RANK_BYTES | The number of bytes transferred by the rank with the largest time spent in MPI I/O (cumulative read, write, and meta times) |
MPIIO_F_*_START_TIMESTAMP | Timestamp that the first MPIIO file open/read/write/close operation began |
MPIIO_F_*_END_TIMESTAMP | Timestamp that the last MPIIO file open/read/write/close operation ended |
MPIIO_F_READ_TIME | Cumulative time spent reading at MPI level |
MPIIO_F_WRITE_TIME | Cumulative time spent write and sync at MPI level |
MPIIO_F_META_TIME | Cumulative time spent in open and close at MPI level |
MPIIO_F_MAX_READ_TIME | Duration of the slowest individual MPI read operation |
MPIIO_F_MAX_WRITE_TIME | Duration of the slowest individual MPI write operation |
MPIIO_F_FASTEST_RANK_TIME | The time of the rank which had the smallest amount of time spent in MPI I/O (cumulative read, write, and meta times) |
MPIIO_F_SLOWEST_RANK_TIME | The time of the rank which had the largest amount of time spent in MPI I/O (cumulative read, write, and meta times) |
MPIIO_F_VARIANCE_RANK_TIME | The population variance for MPI I/O time of all the ranks |
MPIIO_F_VARIANCE_RANK_BYTES | The population variance for bytes transferred of all the ranks at MPI level |
Table 3. STDIO module
| counter name | description |
|---|---|
STDIO_OPENS | Count of stdio file open operations (INCLUDING |
STDIO_FDOPENS | Count of stdio fdopen operations |
STDIO_READS | Count of stdio read operations |
STDIO_WRITES | Count of stdio write operations |
STDIO_SEEKS | Count of stdio seek operations |
STDIO_FLUSHES | Count of stdio flush operations |
STDIO_BYTES_WRITTEN | Total number of bytes written to the file using stdio operations |
STDIO_BYTES_READ | Total number of bytes read from the file using stdio operations |
STDIO_MAX_BYTE_READ | Highest offset in the file that was read |
STDIO_MAX_BYTE_WRITTEN | Highest offset in the file that was written |
STDIO_FASTEST_RANK | The MPI rank with the smallest time spent in stdio operations (cumulative read, write, and meta times) |
STDIO_FASTEST_RANK_BYTES | The number of bytes transferred by the rank with the smallest time spent in stdio operations (cumulative read, write, and meta times) |
STDIO_SLOWEST_RANK | The MPI rank with the largest time spent in stdio operations (cumulative read, write, and meta times) |
STDIO_SLOWEST_RANK_BYTES | The number of bytes transferred by the rank with the largest time spent in stdio operations (cumulative read, write, and meta times) |
STDIO_META_TIME | Cumulative time spent in stdio open/close/seek operations |
STDIO_WRITE_TIME | Cumulative time spent in stdio write operations |
STDIO_READ_TIME | Cumulative time spent in stdio read operations |
STDIO_*_START_TIMESTAMP | Timestamp that the first stdio file open/read/write/close operation began |
STDIO_*_END_TIMESTAMP | Timestamp that the last stdio file open/read/write/close operation ended |
STDIO_F_FASTEST_RANK_TIME | The time of the rank which had the smallest time spent in stdio I/O (cumulative read, write, and meta times) |
STDIO_F_SLOWEST_RANK_TIME | The time of the rank which had the largest time spent in stdio I/O (cumulative read, write, and meta times) |
STDIO_F_VARIANCE_RANK_TIME | The population variance for stdio I/O time of all the ranks |
STDIO_F_VARIANCE_RANK_BYTES | The population variance for bytes transferred of all the ranks |
Table 4. H5F module
| counter name | description |
|---|---|
H5F_OPENS | Count of H5F opens |
H5F_FLUSHES | Count of H5F flushes |
H5F_USE_MPIIO | Flag indicating whether MPI-IO is used for accessing the file |
H5F_F_*_START_TIMESTAMP | Timestamp that the first H5F open/close operation began |
H5F_F_*_END_TIMESTAMP | Timestamp that the last H5F open/close operation ended |
H5F_F_META_TIME | Cumulative time spent in H5F open/close/flush operations |
Table 5. H5D module
| counter name | description |
|---|---|
H5D_OPENS | Count of H5D opens |
H5D_READS | Count of H5D reads |
H5D_WRITES | Count of H5D writes |
H5D_FLUSHES | Count of H5D flushes |
H5D_BYTES_READ | Total number of bytes read from the dataset using H5D |
H5D_BYTES_WRITTEN | Total number of bytes written to the dataset using H5D |
H5D_RW_SWITCHES | Number of times that access toggled between read and write in consecutive H5D operations |
H5D_REGULAR_HYPERSLAB_SELECTS | Number of H5D read/write ops with regular hyperslab selections |
H5D_IRREGULAR_HYPERSLAB_SELECTS | Number of H5D read/write ops with irregular hyperslab selections |
H5D_POINT_SELECTS | Number of read/write ops with point selections |
H5D_MAX_READ_TIME_SIZE | Size of the slowest H5D read operation |
H5D_MAX_WRITE_TIME_SIZE | Size of the slowest H5D write operation |
H5D_SIZE_READ_AGG_* | Histogram of total size of read accesses at H5D level |
H5D_SIZE_WRITE_AGG_* | Histogram of total size of write accesses at H5D level |
H5D_ACCESS[1-4]_ACCESS | Sizes of 4 most common H5D accesses |
H5D_ACCESS[1-4]_LENGTH_D[1-5] | Access lengths along last 5 dimensions (D5 is fastest changing) of 4 most common H5D accesses |
H5D_ACCESS[1-4]_STRIDE_D[1-5] | Access strides along last 5 dimensions (D5 is fastest changing) of 4 most common H5D accesses |
H5D_ACCESS[1-4]_COUNT | Count of 4 most common H5D aggregate access sizes |
H5D_DATASPACE_NDIMS | Number of dimensions in dataset’s dataspace |
H5D_DATASPACE_NPOINTS | Number of points in dataset’s dataspace |
H5D_DATATYPE_SIZE | Total size of dataset elements in bytes |
H5D_CHUNK_SIZE_D[1-5] | Chunk sizes in the last 5 dimensions of the dataset (D5 is the fastest changing dimension) |
H5D_USE_MPIIO_COLLECTIVE | Flag indicating use of MPI-IO collectives |
H5D_USE_DEPRECATED | Flag indicating whether deprecated create/open calls were used |
H5D_FASTEST_RANK | The MPI rank with smallest time spent in H5D I/O (cumulative read, write, and meta times) |
H5D_FASTEST_RANK_BYTES | The number of bytes transferred by the rank with smallest time spent in H5D I/O (cumulative read, write, and meta times) |
H5D_SLOWEST_RANK | The MPI rank with largest time spent in H5D I/O (cumulative read, write, and meta times) |
H5D_SLOWEST_RANK_BYTES | The number of bytes transferred by the rank with the largest time spent in H5D I/O (cumulative read, write, and meta times) |
H5D_F_*_START_TIMESTAMP | Timestamp that the first H5D open/read/write/close operation began |
H5D_F_*_END_TIMESTAMP | Timestamp that the last H5D open/read/write/close operation ended |
H5D_F_READ_TIME | Cumulative time spent reading at H5D level |
H5D_F_WRITE_TIME | Cumulative time spent writing at H5D level |
H5D_F_META_TIME | Cumulative time spent in open/close/flush at H5D level |
H5D_F_MAX_READ_TIME | Duration of the slowest individual H5D read operation |
H5D_F_MAX_WRITE_TIME | Duration of the slowest individual H5D write operation |
H5D_F_FASTEST_RANK_TIME | The time of the rank which had the smallest amount of time spent in H5D I/O (cumulative read, write, and meta times) |
H5D_F_SLOWEST_RANK_TIME | The time of the rank which had the largest amount of time spent in H5D I/O (cumulative read, write, and meta times) |
H5D_F_VARIANCE_RANK_TIME | The population variance for H5D I/O time of all the ranks |
H5D_F_VARIANCE_RANK_BYTES | The population variance for bytes transferred of all the ranks at H5D level |
H5D_FILE_REC_ID | Darshan file record ID of the file the dataset belongs to |
Table 6. PNETCDF_FILE module
| counter name | description |
|---|---|
PNETCDF_FILE_CREATES | PnetCDF file create operation counts |
PNETCDF_FILE_OPENS | PnetCDF file open operation counts |
PNETCDF_FILE_REDEFS | PnetCDF file re-define operation counts |
PNETCDF_FILE_INDEP_WAITS | PnetCDF independent file wait operation counts (for flushing non-blocking I/O) |
PNETCDF_FILE_COLL_WAITS | PnetCDF collective file wait operation counts (for flushing non-blocking I/O) |
PNETCDF_FILE_SYNCS | PnetCDF file sync operation counts |
PNETCDF_FILE_BYTES_READ | PnetCDF total bytes read for all file variables (includes internal library metadata I/O) |
PNETCDF_FILE_BYTES_WRITTEN | PnetCDF total bytes written for all file variables (includes internal library metadata I/O) |
PNETCDF_FILE_WAIT_FAILURES | PnetCDF file wait operation failure counts (failures indicate that variable-level counters are unreliable) |
PNETCDF_FILE_F_*_START_TIMESTAMP | Timestamp that the first PNETCDF file open/close/wait operation began |
PNETCDF_FILE_F_*_END_TIMESTAMP | Timestamp that the last PNETCDF file open/close/wait operation ended |
PNETCDF_FILE_F_META_TIME | Cumulative time spent in file open/close/sync/redef/enddef metadata operations |
PNETCDF_FILE_F_WAIT_TIME | Cumulative time spent in file wait operations (for flushing non-blocking I/O) |
Table 7. PNETCDF_VAR module
| counter name | description |
|---|---|
PNETCDF_VAR_OPENS | PnetCDF variable define/inquire operation counts |
PNETCDF_VAR_INDEP_READS | PnetCDF variable independent read operation counts |
PNETCDF_VAR_INDEP_WRITES | PnetCDF variable independent write operation counts |
PNETCDF_VAR_COLL_READS | PnetCDF variable collective read operation counts |
PNETCDF_VAR_COLL_WRITES | PnetCDF variable collective write operation counts |
PNETCDF_VAR_NB_READS | PnetCDF variable nonblocking read operation counts |
PNETCDF_VAR_NB_WRITES | PnetCDF variable nonblocking write operation counts |
PNETCDF_VAR_BYTES_* | total bytes read and written at PnetCDF variable layer (not including internal library metadata I/O) |
PNETCDF_VAR_RW_SWITCHES | number of times access alternated between read and write |
PNETCDF_VAR_PUT_VAR* | number of calls to different ncmpi_put_var* APIs (var, var1, vara, vars, varm, varn, vard) |
PNETCDF_VAR_GET_VAR* | number of calls to different ncmpi_get_var* APIs (var, var1, vara, vars, varm, varn, vard) |
PNETCDF_VAR_IPUT_VAR* | number of calls to different ncmpi_iput_var* APIs (var, var1, vara, vars, varm, varn) |
PNETCDF_VAR_IGET_VAR* | number of calls to different ncmpi_iget_var* APIs (var, var1, vara, vars, varm, varn) |
PNETCDF_VAR_BPUT_VAR* | number of calls to different ncmpi_bput_var* APIs (var, var1, vara, vars, varm, varn) |
PNETCDF_VAR_MAX_*_TIME_SIZE | size of the slowest read and write operations |
PNETCDF_VAR_SIZE_*AGG* | histogram of PnetCDf total access sizes for read and write operations |
PNETCDF_VAR_ACCESS*_* | the four most common total accesses, in terms of size and length/stride (in last 5 dimensions) |
PNETCDF_VAR_ACCESS*_COUNT | count of the four most common total access sizes |
PNETCDF_VAR_NDIMS | number of dimensions in the variable |
PNETCDF_VAR_NPOINTS | number of points in the variable |
PNETCDF_VAR_DATATYPE_SIZE | size of each variable element |
PNETCDF_VAR_*_RANK | rank of the processes that were the fastest and slowest at I/O (for shared datasets) |
PNETCDF_VAR_*_RANK_BYTES | total bytes transferred at PnetCDF layer by the fastest and slowest ranks (for shared datasets) |
PNETCDF_VAR_F_*_START_TIMESTAMP | timestamp of first PnetCDF variable open/read/write/close |
PNETCDF_VAR_F_*_END_TIMESTAMP | timestamp of last PnetCDF variable open/read/write/close |
PNETCDF_VAR_F_READ/WRITE/META_TIME | cumulative time spent in PnetCDF read, write, or metadata operations |
PNETCDF_VAR_F_MAX_*_TIME | duration of the slowest PnetCDF read and write operations |
PNETCDF_VAR_F_*_RANK_TIME | fastest and slowest I/O time for a single rank (for shared datasets) |
PNETCDF_VAR_F_VARIANCE_RANK_* | variance of total I/O time and bytes moved for all ranks (for shared datasets) |
PNETCDF_VAR_FILE_REC_ID | Darshan file record ID of the file the variable belongs to |
Each heatmap module record reports a histogram of the number of bytes read or written, per process, over time, for a given I/O API. It provides a synopsis of I/O intensity regardless of how many files are accessed. Heatmap records are never aggregated across ranks.
The file name field is used to indicate the API that produced the histogram record. For exmaple, "heatmap:POSIX" indicates that the record is reporting I/O traffic that passed through the POSIX module.
The number of BIN fields present in each record may vary depending on the job’s execution time and the configurable maximum number of bins chosen at execution time.
Table 8. HEATMAP module
| counter name | description |
|---|---|
HEATMAP_F_BIN_WIDTH_SECONDS | time duration of each heatmap bin |
HEATMAP_READ|WRITE_BIN_* | number of bytes read or written within specified heatmap bin |
Table 9. Lustre module (if enabled, for Lustre file systems)
| counter name | description |
|---|---|
LUSTRE_OSTS | number of OSTs (object storage targets) for the file system |
LUSTRE_MDTS | number of MDTs (metadata targets) for the file system |
LUSTRE_STRIPE_OFFSET | OST id offset specified at file creation time |
LUSTRE_STRIPE_SIZE | stripe size for the file in bytes |
LUSTRE_STRIPE_WIDTH | number of OSTs over which the file is striped |
LUSTRE_OST_ID_* | indices of OSTs over which the file is striped |
Table 10. APXC module header record (if enabled, for Cray XC systems)
| counter name | description |
|---|---|
APXC_GROUPS | total number of groups for the job |
APXC_CHASSIS | total number of chassis for the job |
APXC_BLADES | total number of blades for the job |
APXC_MEMORY_MODE | Intel Xeon memory mode |
APXC_CLUSTER_MODE | Intel Xeon NUMA configuration |
APXC_MEMORY_MODE_CONSISTENT | Intel Xeon memory mode consistent across all nodes |
APXC_CLUSTER_MODE_CONSISTENT | Intel Xeon cluster mode consistent across all nodes |
Table 11. APXC module per-router record (if enabled, for Cray XC systems)
| counter name | description |
|---|---|
APXC_GROUP | group this router is on |
APXC_CHASSIS | chassis this router is on |
APXC_BLADE | blade this router is on |
APXC_NODE | node connected to this router |
APXC_AR_RTR_x_y_INQ_PRF_INCOMING_FLIT_VC[0-7] | flits on VCs of x y tile for router-router ports |
APXC_AR_RTR_x_y_INQ_PRF_ROWBUS_STALL_CNT | stalls on x y tile for router-router ports |
APXC_AR_RTR_PT_x_y_INQ_PRF_INCOMING_FLIT_VC[0,4] | flits on VCs of x y tile for router-nic ports |
APXC_AR_RTR_PT_x_y_INQ_PRF_REQ_ROWBUS_STALL_CNT | stalls on x y tile for router-nic ports |
Table 12. APMPI module header record (if enabled, for MPI applications)
| counter name | description |
|---|---|
MPI_TOTAL_COMM_TIME_VARIANCE | variance in total communication time across all the processes |
MPI_TOTAL_COMM_SYNC_TIME_VARIANCE | variance in total sync time across all the processes, if enabled |
Table 13. APMPI module per-process record (if enabled, for MPI applications)
| counter name | description |
|---|---|
MPI_PROCESSOR_NAME | name of the processor used by the MPI process |
MPI_*_CALL_COUNT | total call count for an MPI op |
MPI_*_TOTAL_BYTES | total bytes (i.e., cumulative across all calls) moved with an MPI op |
MPI_*_MSG_SIZE_AGG_* | histogram of total bytes moved for all the calls of an MPI op |
MPI_*_TOTAL_TIME | total time (i.e, cumulative across all calls) of an MPI op |
MPI_*_MIN_TIME | minimum time across all calls of an MPI op |
MPI_*_MAX_TIME | maximum time across all calls of an MPI op |
MPI_*_TOTAL_SYNC_TIME | total sync time (cumulative across all calls of an op) of an MPI op, if enabled |
MPI_TOTAL_COMM_TIME | total communication (MPI) time of a process across all the MPI ops |
MPI_TOTAL_COMM_SYNC_TIME | total sync time of a process across all the MPI ops, if enabled |
Table 14. BG/Q module (if enabled on BG/Q systems)
| counter name | description |
|---|---|
BGQ_CSJOBID | Control system job ID |
BGQ_NNODES | Total number of BG/Q compute nodes |
BGQ_RANKSPERNODE | Number of MPI ranks per compute node |
BGQ_DDRPERNODE | Size of compute node DDR in MiB |
BGQ_INODES | Total number of BG/Q I/O nodes |
BGQ_ANODES | Dimension of A torus |
BGQ_BNODES | Dimension of B torus |
BGQ_CNODES | Dimension of C torus |
BGQ_DNODES | Dimension of D torus |
BGQ_ENODES | Dimension of E torus |
BGQ_TORUSENABLED | Bitfield indicating enabled torus dimensions |
BGQ_F_TIMESTAMP | Timestamp of when BG/Q data was collected |
The following sections describe addtitional parser options that provide summary I/O characterization data for the given log.
NOTE: These options are currently only supported by the POSIX, MPI-IO, and stdio modules.
Job performance information can be generated using the --perf command-line option.
Example output.
# performance # ----------- # total_bytes: 134217728 # # I/O timing for unique files (seconds): # ........................... # unique files: slowest_rank_io_time: 0.000000 # unique files: slowest_rank_meta_only_time: 0.000000 # unique files: slowest_rank: 0 # # I/O timing for shared files (seconds): # (multiple estimates shown; time_by_slowest is generally the most accurate) # ........................... # shared files: time_by_cumul_io_only: 0.042264 # shared files: time_by_cumul_meta_only: 0.000325 # shared files: time_by_open: 0.064986 # shared files: time_by_open_lastio: 0.064966 # shared files: time_by_slowest: 0.057998 # # Aggregate performance, including both shared and unique files (MiB/s): # (multiple estimates shown; agg_perf_by_slowest is generally the most # accurate) # ........................... # agg_perf_by_cumul: 3028.570529 # agg_perf_by_open: 1969.648064 # agg_perf_by_open_lastio: 1970.255248 # agg_perf_by_slowest: 2206.983935
The total_bytes line shows the total number of bytes transferred
(read/written) by the job. That is followed by three sections:
I/O timing for unique files. This section reports information about any files that were not opened by every rank in the job. This includes independent files (opened by 1 process) and partially shared files (opened by a proper subset of the job’s processes). The I/O time for this category of file access is reported based on the slowest rank of all processes that performed this type of file access.
I/O timing for shared files. This section reports information about files that were globally shared (i.e.
opened by every rank in the job). This section estimates performance for
globally shared files using four different methods. The time_by_slowest
is generally the most accurate, but it may not available in some older Darshan
log files.
shared files: time_by_cumul_*: adds the cumulative time across all processes and divides by the number of processes (inaccurate when there is high variance among processes).
Aggregate performance. Performance is calculated by dividing the total bytes by the I/O time (shared files and unique files combined) computed using each of the four methods described in the previous output section. Note the unit for total bytes is Byte and for the aggregate performance is MiB/s (1024*1024 Bytes/s).
Use the --file option to get totals based on file usage.
Each line has 3 columns. The first column is the count of files for that
type of file, the second column is number of bytes for that type, and the third
column is the maximum offset accessed.
Example output.
# <file_type> <file_count> <total_bytes> <max_byte_offset> # total: 5 4371499438884 4364699616485 # read_only: 2 4370100334589 4364699616485 # write_only: 1 1399104295 1399104295 # read_write: 0 0 0 # unique: 0 0 0 # shared: 5 4371499438884 4364699616485
Use the --total option to get all statistics as an aggregate total rather
than broken down per file. Each field is either summed across files and
process (for values such as number of opens), set to global minimums and
maximums (for values such as open time and close time), or zeroed out (for
statistics that are nonsensical in aggregate).
Example output.
total_POSIX_OPENS: 1024 total_POSIX_READS: 0 total_POSIX_WRITES: 16384 total_POSIX_SEEKS: 16384 total_POSIX_STATS: 1024 total_POSIX_MMAPS: 0 total_POSIX_FOPENS: 0 total_POSIX_FREADS: 0 total_POSIX_FWRITES: 0 total_POSIX_BYTES_READ: 0 total_POSIX_BYTES_WRITTEN: 68719476736 total_POSIX_MAX_BYTE_READ: 0 total_POSIX_MAX_BYTE_WRITTEN: 67108863 ...
The darshan-dxt-parser utility can be used to parse DXT traces out of Darshan
log files, assuming the corresponding application was executed with the DXT
modules enabled. The following example parses all DXT trace information out
of a Darshan log file and stores it in a text file:
darshan-dxt-parser shane_ior_id25016_1-31-38066-13864742673678115131_1.darshan > ~/ior-trace.txt
The preamble to darshan-dxt-parser output is identical to that of the traditional
darshan-parser utility, which is described above.
darshan-dxt-parser displays detailed trace information contained within a Darshan log
that was generated with DXT instrumentation enabled. Trace data is captured from both
POSIX and MPI-IO interfaces. Example output is given below:
Example output.
# *************************************************** # DXT_POSIX module data # *************************************************** # DXT, file_id: 16457598720760448348, file_name: /tmp/test/testFile # DXT, rank: 0, hostname: shane-thinkpad # DXT, write_count: 4, read_count: 4 # DXT, mnt_pt: /, fs_type: ext4 # Module Rank Wt/Rd Segment Offset Length Start(s) End(s) X_POSIX 0 write 0 0 262144 0.0029 0.0032 X_POSIX 0 write 1 262144 262144 0.0032 0.0035 X_POSIX 0 write 2 524288 262144 0.0035 0.0038 X_POSIX 0 write 3 786432 262144 0.0038 0.0040 X_POSIX 0 read 0 0 262144 0.0048 0.0048 X_POSIX 0 read 1 262144 262144 0.0049 0.0049 X_POSIX 0 read 2 524288 262144 0.0049 0.0050 X_POSIX 0 read 3 786432 262144 0.0050 0.0051 # *************************************************** # DXT_MPIIO module data # *************************************************** # DXT, file_id: 16457598720760448348, file_name: /tmp/test/testFile # DXT, rank: 0, hostname: shane-thinkpad # DXT, write_count: 4, read_count: 4 # DXT, mnt_pt: /, fs_type: ext4 # Module Rank Wt/Rd Segment Length Start(s) End(s) X_MPIIO 0 write 0 262144 0.0029 0.0032 X_MPIIO 0 write 1 262144 0.0032 0.0035 X_MPIIO 0 write 2 262144 0.0035 0.0038 X_MPIIO 0 write 3 262144 0.0038 0.0040 X_MPIIO 0 read 0 262144 0.0048 0.0049 X_MPIIO 0 read 1 262144 0.0049 0.0049 X_MPIIO 0 read 2 262144 0.0049 0.0050 X_MPIIO 0 read 3 262144 0.0050 0.0051
This module provides details on each read or write access at the POSIX layer. The trace output is organized first by file then by process rank. So, for each file accessed by the application, DXT will provide each process’s I/O trace segments in separate blocks, ordered by increasing process rank. Within each file/rank block, I/O trace segments are ordered chronologically.
Before providing details on each I/O operation, DXT provides a short preamble for each file/rank trace block with the following bits of information: the Darshan identifier for the file (which is equivalent to the identifers used by Darshan in its traditional modules), the full file path, the corresponding MPI rank the current block of trace data belongs to, the hostname associated with this process rank, the number of individual POSIX read and write operations by this process, and the mount point and file system type corresponding to the traced file.
The output format for each indvidual I/O operation segment is:
# Module Rank Wt/Rd Segment Offset Length Start(s) End(s)
If the MPI-IO interface is used by an application, this module provides details on each read or write access at the MPI-IO layer. This data is often useful in understanding how MPI-IO read or write operations map to underlying POSIX read or write operations issued to the traced file.
The output format for the DXT MPI-IO module is essentially identical to the DXT POSIX module, except that the offset of file operations is not tracked.
The darshan-util package includes a number of other utilies that can be summarized briefly as follows:
--bzip2 flag is given, then the output file will be re-compressed in
bzip2 format rather than libz format. It also has command line options for
anonymizing personal data, adding metadata annotation to the log header, and
restricting the output to a specific instrumented file.
PyDarshan is a Python package that provides functionality for analyzing Darshan log files, first introduced as part of Darshan 3.3.0. This package provides easier to use Python interfaces to Darshan log file data (compared to the C-based darshan-util library), enabling Darshan users to develop their own custom log file analysis utilities.
PyDarshan has independent documentation outlining how to install and use this package which can be found HERE.