darshan-util.txt 25.2 KB
Newer Older
1 2 3 4 5
Darshan-util installation and usage
===================================

== Introduction

6 7 8 9 10
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.
11

12 13 14
More information about Darshan can be found at the 
http://www.mcs.anl.gov/darshan[Darshan web site].

15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
== Requirements

Darshan-util has only been tested in Linux environments, but will likely
work in other Unix-like environments as well.  

.Hard requirements
* C compiler
* zlib development headers and library (zlib-dev or similar)

.Optional requirements
* libbz2 development headers and library (libbz2-dev or similar)
* Perl
* pdflatex
* gnuplot 4.2 or later
* epstopdf

31
== Compilation and installation
32 33 34 35 36 37 38 39 40 41

.Configure and build example
----
tar -xvzf darshan-<version-number>.tar.gz
cd darshan-<version-number>/darshan-util
./configure
make
make install
----

42 43 44 45 46 47 48 49
[NOTE]
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.

50 51 52 53 54 55 56
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.

57 58 59
The `--enable-shared` argument to configure can be used to enable
compilation of a shared version of the darshan-util library.

60 61 62 63 64 65 66 67 68
== Analyzing log files

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:

----
69
<YEAR>/<MONTH>/<DAY>/<USERNAME>_<BINARY_NAME>_<JOB_ID>_<DATE>_<UNIQUE_ID>_<TIMING>.darshan
70 71 72 73
----

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
74 75 76 77 78
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.
79 80 81 82

=== darshan-job-summary.pl

You can generate a graphical summary
Philip Carns's avatar
Philip Carns committed
83
of the I/O activity for a job by using the `darshan-job-summary.pl` graphical summary
84 85 86 87 88 89 90 91 92 93 94 95 96
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.

97 98
An example of the output produced by darshan-job-summary.pl can be found
link:http://www.mcs.anl.gov/research/projects/darshan/docs/ssnyder_ior-hdf5_id3655016_9-23-29011-12333993518351519212_1.darshan.pdf[HERE].
99

100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
=== darshan-summary-per-file.sh

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.

If you would like to produce a summary for a single
specific file, then you can run the following command to produce a quick
list of the files opened by an application and the amount of time spent
performing I/O to each of them:

----
darshan-parser --file-list carns_my-app_id114525_7-27-58921_19.darshan.gz
----

Once you have identified a specific file of interest, then you can produce a
summary for that specific file 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 listed in the
+darshan-parser --file-list+ output.  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.

136 137
=== darshan-parser

138 139 140 141
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:
142 143 144 145 146

----
darshan-parser carns_my-app_id114525_7-27-58921_19.darshan.gz > ~/job-characterization.txt
----

Philip Carns's avatar
Philip Carns committed
147
The format of this output is described in the following section.
148 149 150

=== Guide to darshan-parser output

151
The beginning of the output from darshan-parser displays a summary of
152 153 154 155 156 157 158 159
overall information about the job. Additional job-level summary information
can also be produced using the `--perf`, `--file`, `--file-list`, or
`--file-list-detailed` command line options.  See the
<<addsummary,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:
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175

[cols="25%,75%",options="header"]
|====
|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
|====

176 177 178 179 180 181 182 183 184 185 186 187 188
==== Log file region sizes

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:

* header - constant-sized uncompressed header providing data on how to properly access the log
* job data - job-level metadata (e.g., start/end time and exe name) for the log
* record table - a table mapping Darshan record identifiers to full file name paths
* module data - each module (e.g., POSIX, MPI-IO, etc.) stores their I/O characterization data in distinct regions of the log

All regions of the log file are compressed (in libz or bzip2 format), except the header.

189 190 191 192 193 194 195
==== Table of mounted file systems

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:

----
196
<mount point> <fs type>
197 198 199 200 201 202 203 204
----

==== Format of I/O characterization fields

The remainder of the output will show characteristics for each file that was
opened by the application. Each line uses the following format:

----
205
<module> <rank> <record id> <counter name> <counter value> <file name> <mount point> <fs type>
206 207
----

208 209 210 211 212 213 214 215 216 217 218 219
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.
220 221 222

==== I/O characterization fields

223 224 225 226
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.
227

228
.POSIX module
229 230
[cols="40%,60%",options="header"]
|====
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
| counter name | description
| POSIX_OPENS | Count of how many times the file was opened
| 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_FOPENS | Count of POSIX stream open operations
| POSIX_FREADS | Count of POSIX stream read operations
| POSIX_FWRITES | Count of POSIX stream write operations
| POSIX_FSEEKS | Count of POSIX stream seek operations
| POSIX_FSYNCS | Count of POSIX fsync operations
| POSIX_FDSYNCS | Count of POSIX fdatasync operations
| 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
258 259
runtime on most file systems.  On Lustre, however, Darshan assumes a default
value of 1 MiB for optimal file alignment.
260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
| 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 of the rank with smallest time spent in POSIX I/O
| POSIX_FASTEST_RANK_BYTES | The number of bytes transferred by the rank with smallest time spent in POSIX I/O
| POSIX_SLOWEST_RANK | The MPI rank of the rank with largest time spent in POSIX I/O
| POSIX_SLOWEST_RANK_BYTES | The number of bytes transferred by the rank with the largest time spent in POSIX I/O
| POSIX_F_OPEN_TIMESTAMP | Timestamp of first time that the file was opened
| POSIX_F_READ_START_TIMESTAMP | Timestamp that the first POSIX read operation began
| POSIX_F_WRITE_START_TIMESTAMP | Timestamp that the first POSIX write operation began
| POSIX_F_READ_END_TIMESTAMP | Timestamp that the last POSIX read operation ended
| POSIX_F_WRITE_END_TIMESTAMP | Timestamp that the last POSIX write operation ended
| POSIX_F_CLOSE_TIMESTAMP | Timestamp of the last time that the file was closed
| 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
| 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
|====

.MPI-IO module
[cols="40%,60%",options="header"]
|====
| 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 of the rank with smallest time spent in MPI I/O
| MPIIO_FASTEST_RANK_BYTES | The number of bytes transferred by the rank with smallest time spent in MPI I/O
| MPIIO_SLOWEST_RANK | The MPI rank of the rank with largest time spent in MPI I/O
| MPIIO_SLOWEST_RANK_BYTES | The number of bytes transferred by the rank with the largest time spent in MPI I/O
| MPIIO_F_OPEN_TIMESTAMP | Timestamp of first time that the file was opened at MPI level
| MPIIO_F_READ_START_TIMESTAMP | Timestamp that the first MPI read operation began
| MPIIO_F_WRITE_START_TIMESTAMP | Timestamp that the first MPI write operation begin
| MPIIO_F_READ_END_TIMESTAMP | Timestamp that the last MPI read operation ended
| MPIIO_F_WRITE_END_TIMESTAMP | Timestamp that the last MPI write operation ended
| MPIIO_F_CLOSE_TIMESTAMP | Timestamp of the last time that the file was closed at MPI level
| MPIIO_READ_TIME | Cumulative time spent reading at MPI level
| MPIIO_WRITE_TIME | Cumulative time spent write and sync at MPI level
| MPIIO_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
| CP_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)
| CP_F_SLOWEST_RANK_TIME | The time of the rank which had the largest amount of time spent in MPI I/O
| CP_F_VARIANCE_RANK_TIME | The population variance for MPI I/O time of all the ranks
| CP_F_VARIANCE_RANK_BYTES | The population variance for bytes transferred of all the ranks at MPI level
335 336
|====

337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356
.HDF5 module
[cols="40%,60%",options="header"]
|====
| counter name | description
| HDF5_OPENS | Count of HDF5 opens
| HDF5_F_OPEN_TIMESTAMP | Timestamp of first time that the file was opened at HDF5 level
| HDF5_F_CLOSE_TIMESTAMP | Timestamp of the last time that the file was closed at HDF5 level
|====

.PnetCDF module
[cols="40%,60%",options="header"]
|====
| counter name | description
| PNETCDF_INDEP_OPENS | Count of PnetCDF independent opens
| PNETCDF_COLL_OPENS | Count of PnetCDF collective opens
| PNETCDF_F_OPEN_TIMESTAMP | Timestamp of first time that the file was opened at PnetCDF level
| PNETCDF_F_CLOSE_TIMESTAMP | Timestamp of the last time that the file was closed at PnetCDF level
|====

===== Additional modules 
357

358
.BG/Q module (if enabled on BG/Q systems)
359 360
[cols="40%,60%",options="header"]
|====
361 362 363 364 365 366 367 368 369 370 371 372 373
| 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
374 375 376
|====

==== Additional summary output
377
[[addsummary]]
378

379 380 381 382 383
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 and MPI-IO modules.

384 385
===== Performance

386
Job performance information can be generated using the `--perf` command-line option.
387 388 389 390 391

.Example output
----
# performance
# -----------
392 393 394 395
# total_bytes: 134217728
#
# I/O timing for unique files (seconds):
# ...........................
396
# unique files: slowest_rank_io_time: 0.000000
397
# unique files: slowest_rank_meta_only_time: 0.000000
398
# unique files: slowest_rank: 0
399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416
#
# 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
417 418
----

419 420 421 422 423 424 425 426 427 428 429 430 431 432
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.

* unique files: slowest_rank_io_time: total I/O time for unique files
  (including both metadata + data transfer time)
433
* unique files: slowest_rank_meta_only_time: metadata time for unique files
434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466
* unique files: slowest_rank: the rank of the slowest process

.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).
** shared files: time_by_cumul_io_only: include metadata AND data transfer
   time for global shared files
** shared files: time_by_cumul_meta_only: metadata time for global shared
   files
* shared files: time_by_open: difference between timestamp of open and
  close (inaccurate if file is left open without I/O activity)
* shared files: time_by_open_lastio: difference between timestamp of open
  and the timestamp of last I/O (similar to above but fixes case where file is
  left open after I/O is complete)
* shared files: time_by_slowest : measures time according to which rank was
  the slowest to perform both metadata operations and data transfer for each
  shared file. (most accurate but requires newer log version)

.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).

467 468
===== Files
Use the `--file` option to get totals based on file usage.
469 470 471
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.
472 473 474 475

* total: All files
* read_only: Files that were only read from
* write_only: Files that were only written to
476
* read_write: Files that were both read and written
477 478 479
* unique: Files that were opened on only one rank
* shared: File that were opened by more than one rank

480

481 482 483 484 485 486 487 488 489 490 491 492 493 494
.Example output
----
# files
# -----
# total: 1542 236572244952 154157611
# read_only: 3 133998651 122805519
# write_only: 1539 236438246301 154157611
# read_write: 0 0 0
# unique: 2 11193132 11193063
# shared: 1540 236561051820 154157611
----

===== Totals

495 496 497 498 499
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).
500 501 502

.Example output
----
503 504 505 506 507 508 509 510 511 512 513 514 515
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
516 517
...
----
518

519 520 521 522 523 524 525 526 527
===== File list

Use the `--file-list` option to produce a list of files opened by the
application along with estimates of the amount of time spent accessing each
file.

.Example output
----
# Per-file summary of I/O activity.
528 529
# <record_id>: darshan record id for this file
# <file_name>: full file name
530 531 532 533
# <nprocs>: number of processes that opened the file
# <slowest>: (estimated) time in seconds consumed in IO by slowest process
# <avg>: average time in seconds consumed in IO per process

534 535
# <record_id>   <file_name> <nprocs>    <slowest>   <avg>
5041708885572677970 /projects/SSSPPg/snyder/ior/ior.dat 1024    16.342061   1.705930
536 537
----

538 539 540 541 542 543 544
This data could be post-processed to compute more in-depth statistics, such as
the total number of MPI files and total number of POSIX files used in a
job, categorizing files into independent/unique/local files (opened by
1 process), subset/partially shared files (opened by a proper subset of
processes) or globally shared files (opened by all processes), and ranking
files according to how much time was spent performing I/O in each file.

545 546 547 548
===== Detailed file list

The `--file-list-detailed` is the same as --file-list except that it
produces many columns of output containing statistics broken down by file.
549
This option is mainly useful for more detailed automated analysis.
550

551
=== Other darshan-util utilities
552 553 554 555 556

The darshan-util package includes a number of other utilies that can be
summarized briefly as follows:

* darshan-convert: converts an existing log file to the newest log format.
557 558
If the `--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
559 560
anonymizing personal data, adding metadata annotation to the log header, and
restricting the output to a specific instrumented file.
561
* darshan-analyzer: walks an entire directory tree of Darshan log files and
Philip Carns's avatar
Philip Carns committed
562
produces a summary of the types of access methods used in those log files.
563
* darshan-logutils*: this is a library rather than an executable, but it
Kevin Harms's avatar
Kevin Harms committed
564
provides a C interface for opening and parsing Darshan log files.  This is
565 566 567 568
the recommended method for writing custom utilities, as darshan-logutils
provides a relatively stable interface across different versions of Darshan
and different log formats.